Tasks
Filter with ?source_site= or ?session_id=. See the grounding for what this schema is.
In progress (1)
Build the cron Worker that wires subagentcache.com to live Cloudflare GraphQL Analytics
workers/subagentcache/README.md documents this as the next step: a scheduled Worker (same [triggers] cron pattern as workers/cron) holding a CF_API_TOKEN secret scoped to Zone Analytics Read only, running a fixed httpRequests1dGroups query, and POSTing the result to subagentcache.com/api/cache-snapshots with source=graphql_api. Not yet built — every existing row is manual_dashboard_paste.
Building the live cache-analytics cron Worker…
closes contract: ctr_cache_analytics_freshness
Pending (16)
Expand subagentidentities.com's registry as new real identity boundaries are documented
Currently seeded with 5 real rows (macos_desktop_cowork, cloud_docker_mcp, engineering-coworker binary mediation, the shared Cloudflare account, vault/Keychain credential pattern). Add more only when a new REAL boundary is documented somewhere in this repo or the mirrored claude-tag docs — never pad with invented entries. See ctr_identity_registry's done_detail.
closes contract: ctr_identity_registry
Expand subagentbrands.com guidelines as more real voice patterns are documented
Currently seeded with 6 real rows observed from this repository's own already-shipped pages. Add more only when a new REAL pattern is observed in an actual shipped file, mirroring tsk_identity_registry_more_rows' discipline of never padding with invented rows.
closes contract: ctr_brand_voice_grounding
Document POST and PATCH usage on subagentcli.com's commands page
The script itself already supports --method and --data flags, but the /commands docs page currently only shows GET examples. Add a worked write-path example once each site's WRITE_SECRET provisioning process is documented for end users.
Resync subagentdatamodels.com sibling-site references with full 37-site family
AUDIT 2026-07-02 (scripts/resync-audit.py): cannot complete — https://subagentdatamodels.com is unreachable because the domain is UNREGISTERED (blocker tsk_b228566b-1c3, operator purchase decision). Local roster/hub checks pass; the llms.txt check needs a live domain. Re-run the audit after registration; do not hand-audit before then.
Register subagentdatamodels.com (purchase decision — operator only)
ROOT CAUSE CORRECTED 2026-07-02: the domain is UNREGISTERED — whois against whois.verisign-grs.com (the .com registry itself) returns No match. Every credential approach (wrangler OAuth scopes, docker mcp secrets, dashboard Add Site) was moot: there is no domain to add a zone for. Fix: operator registers subagentdatamodels.com (~$10/yr, Cloudflare Registrar recommended so NS wire automatically) — a financial transaction Claude will not execute, and gated by cwc-finance spend_commit anyway. Once registered: zone create (dashboard or token), [[routes]] per workers/subagentdatamodels/README.md, deploy, add to subagentapi.com FAMILY_LINKSET_EXTRAS + MCP set — all automated. Interim mitigation SHIPPED: subagentdomains.com directory now links the live workers.dev URL instead of the dead apex (version 7ba1d000), so the fleet no longer teaches agents a dead link.
SDK backlog: dedupe esc() (14 workers) and htmlHeaders (53 workers)
Duplication scan 2026-07-02: function esc( defined in 14 workers, htmlHeaders const in 53, WebMCP inline script in 5 (semi-bespoke per-site tools). Low individual value, high churn — fold opportunistically into sites already being touched for other reasons (a new kit text/html module), never as a standalone fleet sweep. Editing existing kit files makes all their dependents stale per worker-build.sh — batch with a planned fleet deploy.
BUDGET-5K #1: approve runway lock — 12mo claude.ai pre-fund ($1,200)
spend_commit gate, operator decision. Existing spend class (the $100/mo AI-workforce line), lowest-risk approval in the plan. Plan: plugins/cwc-finance/skills/startup-finance/BUDGET-5K.md
BUDGET-5K #2: approve entity formation via Stripe Atlas ($800) — on formation merits; credits at realizable value
CORRECTED 2026-07-02 after operator challenged the $100k face value — challenge verified correct against cloudflare.com/forstartups: the $100k is Tier 2 of the standard startup program (1yr window, Workers AI capped $25k at list pricing, R2 $10k, AI Gateway excluded, Registrar renewals NOT eligible). Realizable value at current fleet scale ≈ $60-600, not $100k. Atlas route stands on formation merits alone (DE+EIN+equity+83(b)+agent, fee historically $500). New standing rule in BUDGET-5K.md: vendor credits booked at realizable value, never face value; no credit number without its terms page fetched and dated. spend_commit + attorney-review gate unchanged.
BUDGET-5K #3: trademark clearance then filing ($900) — after entity
Clearance search BEFORE filing money (legal rule); entity owns the mark; USPTO fees verified live at filing time. spend_commit-gated.
BUDGET-5K #4: portfolio protection ($650) — 1yr renewals + register datamodels
Pre-fund ~50 renewals (~$550) + register subagentdatamodels.com (~$11, closes tsk_b228566b-1c3 and the 1/50 gap on four flywheel metrics). Existing spend class. spend_commit-gated.
BUDGET-5K #5+6: conditional lines — CF paid capability ($600, cancels if Launchpad credits land) + monetization pilot reserve ($350, gateway-access trigger)
Both trigger-gated per BUDGET-5K.md sequencing; #5 reallocates to trademark buffer/contingency if the $250k Startup Enterprise credits arrive. Any real x402 transaction is operator-executed.
WBR-W27 rec #1: instrument the north star (agent-UA / Crawl Control baseline)
From reports/wbr/2026-W27.md — the business's output metrics are all class-2 (measurable, not instrumented). This week: manual AI Crawl Control dashboard paste to subagentcache.com (source=manual_dashboard_paste, needs operator dashboard access); automation path is existing tsk_wire_live_cache_graphql (CF_API_TOKEN GraphQL). Predicted movement: output metrics class 2 → class 1; next WBR reads agent-UA share for the first time. Gate: engineering build gate + operator access.
Review zone bot protection vs agent-readiness: Python-urllib UAs get 403
Found 2026-07-03 by scripts/territory-score.py (differential-verified: Python-urllib/3.12 → 403, identified UAs → 200, on subagentapi.com; likely zone-wide Bot Fight Mode/managed rule). Contradiction: llms.txt + MCP invite agents while the zone layer blocks the most default agent client. Operator judgment needed (security posture): allow known-agent UAs via WAF skip, rely on AI Crawl Control instead, or accept the block and document it in llms.txt. Ties to WBR rec #1 instrumentation.
Follow-up: fleet-wide 14→15 roster sweep beyond deck/plugins-page
PARTIAL 2026-07-03: subagentcoworkers.com homepage overview (3 spots) + plugins page + deck + CLAUDE.md now say 15 and science. STILL TODO on OTHER sites: subagentroles.com seed data (bio-research role row → science; +agent-resources row when ROLES_WRITE_SECRET available), subagentdomains.com any coworker counts, subagentdevices.com architecture plugin-layer count. Grep 14/fourteen/bio-research per site.
Refactor T3 (deferred): distiller from regex to stdlib html.parser
crawlers/claude_docs/md.py distills HTML via regex; stdlib html.parser.HTMLParser would be structurally robust (nesting, attributes, malformed tags). Deferred — working code behind the CommonMark gate; a rewrite needs golden-file comparison across ~10 mirrored pages plus the cm gate. Do as its own pass.
deploy-verify-task
deploy verify
Completed (197)
Verify remaining candidate board counts
Confirm final counts of Greenhouse/Lever/Ashby candidate boards before seeding.
Seed dim_board for reconciled + new companies
Insert reconciled and newly discovered company boards into dim_board.
Trigger live crawls for all new/reconciled GH+Lever boards
Kick off live crawl runs against the expanded Greenhouse + Lever board list.
Manually ingest OpenAI (Ashby) jobs into D1
One-off manual ingestion of OpenAI's Ashby-hosted postings into fact_job_posting.
Update cron worker source with expanded board lists + Ashby
workers/cron index.ts: add Ashby to the 3-way platform dispatch, expand board lists.
Create ETL worker source with Ashby support + dim_board upsert fix
workers/etl: Ashby crawl endpoint plus an ON CONFLICT upsert fix for dim_board.
Update board YAML docs + new migration file
Docs and a new Postgres migration reflecting the expanded board catalog.
Research Netflix analytics/data engineer openings
One-off research task on Netflix's open analytics/data-engineering roles.
Final verification + summary for user
Wrap-up verification pass and summary for the board-expansion work.
Research existing cwc-design plugin + agent-gen schema
Survey the existing cwc-design plugin and the agent-gen schema before building cwc-data.
Find Cloudflare zone for subagentetl.com
Confirm the subagentetl.com zone exists on the account before scaffolding.
Build workers/subagentetl Cloudflare Worker
Scaffold the Kimball-style ETL pipeline dashboard Worker.
Update cloudflare.toml with subagentetl entry
Register the new worker + zone in the repo's cloudflare.toml.
Build plugins/cwc-data plugin scaffold
Data-coworker plugin: warehouse analyst over subagentjobs-dwh, read-only by design.
Verify + final summary with deploy commands
Pre-deploy verification pass and a summary of the exact deploy commands.
Deploy subagentetl/etl/cron and live-verify via Desktop Commander
Real deploy of all three touched workers, verified live.
Research Managed Agents/Claude Tag/Office Agents docs
Initial research pass across Anthropic's Managed Agents, Claude Tag, and Office Agents documentation.
Write sessions/handoff-2026-07-01.json
Structured, enum-constrained session handoff document.
Append session entry to session-index.json
Add this session's entry to the running session index.
Update CLAUDE.md for handoff discoverability
Point CLAUDE.md at the new handoff file.
Verify handoff artifacts
Confirm the handoff JSON and index entries are internally consistent.
Fetch and save Claude Tag docs into docs/ mirror
Mirror Anthropic's Claude Tag documentation verbatim into docs/docs/claude.com/docs/claude-tag.
Investigate knowledge-engineering sister repo
Resolve the sister-repo identity and its relationship to subagentknowledge.com.
Confirm Cloudflare zones for both new domains
Verify subagentcowork.com and subagentcoworkers.com zones are active on the account.
Finalize content/architecture design for both sites
Scope the docs-mirror vs. original-writing split before scaffolding.
Scaffold workers/subagentcowork
Static docs Worker mirroring Cowork + Claude Tag documentation.
Scaffold workers/subagentcoworkers
Original docs Worker for the operator's own multi-agent coworkers platform.
Deploy both Workers via Desktop Commander
First live deploy of subagentcowork-web and subagentcoworkers-web.
Live-verify both domains + update cloudflare.toml/CLAUDE.md
Confirm both sites live, document them in the repo.
Research CMA + Claude Tag primitive mapping
Compare Cloudflare's Claude Managed Agents architecture against Claude Tag's own primitives.
Author cloudflare/ section for subagentcowork.com
Five original pages mapping CMA primitives against Claude Tag's, including the open-models adaptation.
Author cloudflare-primitives page for subagentcoworkers.com
Map the coworkers platform's own primitives against CMA's.
Read wc2026-bracket + web design system source
Locate the canonical shared CSS design system already in production.
Check for shared design-system doc or constants
Confirm there was no separate design-system doc to reference instead of the source.
Rewrite subagentcowork.com shell() to match design system
Port the full shared token set + progressive-enhancement techniques.
Rewrite subagentcoworkers.com shell() to match design system
Same port, green accent instead of cyan.
Deploy + verify redesigned sites on real device viewport
Live redeploy and visual verification of both redesigned sites.
Ground subagentcontracts.com in claude-tag's durable/contract vocabulary
Read claude-tag's durability + definition-of-done docs to define the Contract primitive precisely.
closes contract: ctr_ship_subagentcontracts
Fix: add SHA256 drift-detection to mirrored docs
Content-hash every mirrored page so upstream Anthropic doc changes surface as drift.
closes contract: ctr_ship_subagentcontracts
Scaffold subagentcontracts.com worker
D1-backed Worker implementing the durable Contract + contract_ledger primitive.
closes contract: ctr_ship_subagentcontracts
Wire dogfooding: cowork + coworkers sites consume Contracts API
Server-side fetch from both docs sites into the live Contracts API.
closes contract: ctr_ship_subagentcontracts
Build Figma/Canva-style design-system artifact page
Live, interactive style-guide page with a real accent picker.
closes contract: ctr_ship_subagentcontracts
Update CLAUDE.md + cloudflare.toml for subagentcontracts.com
Document the new worker, zone, and D1 database.
closes contract: ctr_ship_subagentcontracts
Deploy and live-verify all changes
Real deploy of all three touched workers, cross-site data flow verified live.
closes contract: ctr_ship_subagentcontracts
Write brief PRD for subagenttasks.com
Self-steered product scoping grounded in the literal TaskCreate/TaskUpdate/TaskList schema, without asking the user to pick.
closes contract: ctr_ship_subagenttasks
Scaffold subagenttasks.com D1 schema + worker
New Cloudflare Worker: Task + task_dependencies primitive, D1-backed, same design-system family, red accent.
closes contract: ctr_ship_subagenttasks
Seed real session data: this session's 44 tasks
Insert the actual 44 completed tasks from this session as genuine seed data, with real dependency edges and contract links.
closes contract: ctr_ship_subagenttasks
Cross-wire: contracts shows backing tasks; cowork/coworkers link to tasks
subagentcontracts.com pulls backing tasks from subagenttasks.com; cowork/coworkers get nav links + blurbs.
closes contract: ctr_ship_subagenttasks
Update CLAUDE.md + cloudflare.toml for subagenttasks.com
Document the new worker, zone, D1 database, and the four-way relationship.
closes contract: ctr_ship_subagenttasks
Deploy all touched workers + live-verify
Deploy subagenttasks (new), subagentcontracts, subagentcowork, subagentcoworkers; verify cross-site data flows.
closes contract: ctr_ship_subagenttasks
Retrofit subagentjobs.com, subagentdata.com, subagentworkers.com, subagentdb.com, subagentetl.com onto the full shared design system
CLOSED AS ALREADY-DONE 2026-07-02 (stale ledger, not missing work): the retrofit this task describes was completed 2026-07-01 per CLAUDE.md Design-system remediation + the retrofit strikethrough item. Grep-verified today: web/db/workers/etl all have @view-transition=1, rgb(from var()) relative-color (e.g. workers/web/src/index.ts:300,301,465), text-wrap>=2, display-p3>=1; wc2026-bracket has view-transition x3 + p3 x6 (it originated the techniques). Earlier audits missed relative-color because they grepped color(from — the shipped form is rgb(from.
Wire an automated evaluation pipeline instead of a one-time manual grading pass
DONE 2026-07-02: scripts/evaluate-workers.py (built by an opus subagent under a fixed JSON I/O contract, verified by lead). Deterministic: rubric fetched from subagentrubrics.com/rbc_worker_quality, site list from build-graph.json (routes+D1 metadata), 7 mechanical checks (d1_binding, grid_overflow_fix, p3_block, build_lock, mcp_live, rel_alternate, changes_or_citeas), fixed result mapping, every explanation cites check evidence + rubric fragment. 62 eval_auto_* rows POSTed live (e.g. eval_auto_wc2026-bracket); secret read at runtime from the local store, never in repo (py_compile clean, no 32+ hex literals). Rerun on any deploy: python3 scripts/evaluate-workers.py --post.
closes contract: ctr_evaluation_evidence
Fix subagentdomains.com CSS overflow on iPhone 16 Pro/Chrome
Real user-reported bug: .site-card and .card grid items had no min-width:0, and their text children (domain names, descriptions) had no overflow-wrap/word-break, so a long unbroken token forced the grid track wider than the viewport. Fixed by adding min-width:0 to the card/grid-item rules and overflow-wrap:anywhere;word-break:break-word to text-bearing children, plus flex-wrap on badge rows. Typechecked, deployed (Version ID 1966d9c5-438f-4947-a178-cda220859e52), live-verified via curl.
Fix CSS grid-overflow gap on remaining 18 terminal-aesthetic sites
After fixing subagentdomains.com, a repo-wide grep found 18 more sites sharing the exact same vulnerable "sharedCss()"/".site-card" CSS lineage with zero overflow-wrap protection: subagentapi, subagentbrands, subagentcache, subagentcitations, subagentcli, subagentcode, subagentcontracts, subagentcowork, subagentcoworkers, subagentdevices, subagentevaluations, subagentidentities, subagentmemories, subagentoutcomes, subagentprompts, subagentroles, subagentrubrics, subagentskills, subagenttasks. Fixed via 3 parallel dispatched agents (6-7 sites each), each locating its own real card/grid class names and applying the same min-width:0 + overflow-wrap pattern. All 19 sites typechecked, deployed in 3 batches, and live-verified via curl (min-width:0 count 3-5 per site).
Extend design-system-audit.sh to check for overflow-wrap/min-width grid gap
The audit script only ever checked display-p3/reduced-motion/safe-area/view-transition/text-wrap/backdrop-filter/scrollbar-width, so the grid-overflow bug class was structurally invisible to it across multiple prior "full family audit" passes. Added a third hard-gate criterion: any file whose CSS defines a repeating-card grid (via grid-template-columns) must also contain both overflow-wrap and min-width:0, or the audit fails it. Re-running the extended script immediately caught two more, previously-unknown real gaps predating this whole incident: subagentsubdomains.com's .card rule (overflow-wrap present, min-width:0 missing) and web/subagentjobs.com's .bcard/.bnm board-grid (neither present). Both fixed, typechecked, deployed, and the audit now passes clean (exit 0) across all 36 sites.
Resync subagentapi.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: CATALOG(37 domains, handoffs=yes); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentbrands.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcache.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcitations.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcli.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: RESOURCES(37 domains, handoffs=yes); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcode.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcontracts.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcowork.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentcoworkers.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentdb.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentdevices.com sibling-site references with full 37-site family
DONE 2026-07-02: subagenthandoffs.com live-verified shipping all 8 family-wide techniques (curl marker check: display-p3 x5, safe-area x3, view-transition x3, rgb(from x1, text-wrap x2, backdrop-filter x2, scrollbar-width x1, reduced-motion x2; the two wc2026-only techniques correctly excluded) and added to the 8 TECHNIQUES sites lists. Shipped via worker-build gate (the gate caught a missing-comma bug on first attempt — exit 1 as designed) → deploy 6d35610b → --record. Live: /adoption column <th>handoffs</th> present, /techniques/display-p3 lists the domain, resync-audit verdict now pass.
Resync subagentdomains.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: SITES(49 domains, handoffs=yes); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentevaluations.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagenthandoffs.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagenthooks.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentidentities.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentmemories.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentmodels.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentoutcomes.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentpermissions.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentprompts.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentpython.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentroles.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentrubrics.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentrust.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentsessions.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentskills.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentsubdomains.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagenttasks.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagenttools.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagenttypescript.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentvaults.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentworkers.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentjobs.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Resync subagentdata.com sibling-site references with full 37-site family
CLOSED by deterministic batch audit (scripts/resync-audit.py, 2026-07-02): llms.txt family links live-verified=True; roster arrays >=15 domains: none (hub sites are the canonical roster surface); hub link present=True (src reference or shared-kit llmsTxt, which emits the family section). Full verdicts JSON archived with the audit script; rerun anytime: python3 scripts/resync-audit.py.
Mirror Figma wireframe board in durable ledger
Architecture board https://www.figma.com/board/f3POlLrQzTL7JmDIbi7BX3 (full-system: 14 cwc-* plugins, identity directives, primitives, 50-site fleet) is recorded in plugins/cwc-design/wireframes.json; in-fleet rendering is subagentdevices.com/architecture (operates edges live 2026-07-02). This row is the durable mirror per the architecture-wireframes skill ledger rule.
Flywheel: add 13 missing sites to the RFC 9727 api-catalog
DONE 2026-07-02: linkset now 49 entries (was 37) via FAMILY_LINKSET_EXTRAS in workers/subagentapi/src/index.ts (kept separate from CATALOG so /health routes[0] probing and the primitives-count copy stay correct). subagentdatamodels.com deliberately excluded — no zone yet; a dead anchor is anti-flywheel. Deployed 74fb554c, curl-verified: .well-known/api-catalog returns 49.
Flywheel: subagentevaluations.com llms.txt missing family catalog link
DONE 2026-07-02: fixed by redeploy — the live worker predated the shared kit's llmsTxt family section; no code change needed. curl-verified: llms.txt now links subagentapi.com/api/catalog. Steady-state crosslink metric 48/50 = subagentsubdomains.com (by design, it IS the directory) + subagentdatamodels.com (no zone).
Flywheel: replicate the /mcp endpoint + server card fleet-wide
DONE 2026-07-02: workers/_shared/kit/mcp.ts factors subagentapi.com's Fable-5-reviewed handler (notifications→202, in-protocol errors over HTTP 200, parse-error id:null, GET→405, CORS, isError only for recognized-tool runtime failures) into a reusable handleMcp factory. Wired into all 24 D1 primitives (3 parallel lanes, sequential deploy+curl-verify per site — every site handshake-verified: initialize, notification 202, tools/call returning real rows). subagentapi.com catalog now advertises POST /mcp on all 24 entries and server-card describedby on 25 linkset entries (version 5d0fb34d). scripts/flywheel-metrics.sh measures 25/50 live MCP endpoints, up from 1/50 baseline.
Flywheel: link rel=alternate llms.txt in every shell() head
DONE 2026-07-02: link rel=alternate type=text/plain href=/llms.txt added to workers/_shared/kit/shell.ts head; all 63 kit workers redeployed (3 batches); 6 bespoke-shell sites (web/jobs, db, workers, etl, wc2026/data, models — models turned out bespoke, not kit) patched individually and deployed. curl-verified 49/50 homepages advertise it; only zoneless subagentdatamodels.com missing.
Flywheel: cite_as canonical URL on API rows (stage 6)
DONE 2026-07-02: citeBase added to workers/_shared/kit/d1crud.ts (list/get append cite_as: <citeBase>/<id>); 58 CRUD specs wired across 48 workers (auto-derived table→route plus an 11-site explicit map for non-derivable routes like legal_coworker_facts→/api/legal-facts); placement + table↔route pairing audited fleet-wide (0 misplaced, 0 mispaired); 51 workers deployed. Live-verified samples: roles/citations/identities/brands/news rows all carry resolvable cite_as URLs. Note: my first pass-2 insert script corrupted 3 files via stale offsets while mutating — caught by typecheck, repaired with reverse-order insertion, documented here so the pattern isn't repeated.
Flywheel: /api/changes polling feeds on contracts/tasks/memories (stage 5)
DONE 2026-07-02: GET /api/changes live on subagentcontracts.com (contract_ledger), subagenttasks.com (tasks.updated_at), subagentmemories.com (memory_versions) — thin newest-first {type, ref, at, cite_as} events, cursor+limit verified (page-2 strictly older than cursor; cursor must be URL-encoded, it contains a space). llms.txt lines added; subagentapi.com catalog advertises all three (version 838c624e); scripts/flywheel-metrics.sh now reports per-feed freshness (contracts 2026-07-01 20:55, tasks 2026-07-02 19:09, memories 2026-07-02 01:54).
Worker build system: dependency graph + kit-hash locks
DONE 2026-07-02, adversarially verified by execution: scripts/build-graph.py generates workers/build-graph.json (71 workers, 7 kit files; 259 kit_import + 55 runtime_fetch + 647 reference edges, all with file:line evidence; deterministic output, generated_at from git HEAD; scanner reads wrangler [vars] URLs — caught subagentdatamodels' env.CATALOG_URL). scripts/worker-build.sh: tsc + kit-hash drift gate vs per-worker .build-lock.json (66 locks bootstrapped = exactly the 66 kit-importing workers); exit 2 names the changed kit file; --stale fleet mode. Gate PROVEN: kit edit → --stale exit 1 naming http.ts (66 workers) / css.ts flip → worker exit 2 + 63 stale — then byte-exact revert, sha256-verified clean. --check-refs live: all 49 runtime-fetch targets HTTP 200. Real fan-out: wellknown 66, shell 63, http 57, d1crud 49, mcp 24.
Build-system skills across 4 plugins (design/pm/engineering/operations)
DONE 2026-07-02, all four registered in their plugin.json manifests and fabrication-checked against the real tools: cwc-engineering:worker-build (build sequence, lock semantics, exit codes, new-dependency protocol), cwc-operations:dependency-drift-audit (fleet staleness as blast-radius-ordered redeploy queue; weekly review updated to include it), cwc-design:dependency-graph-sync (+ scripts/arch-map-drift.py — real run found 26 drift findings between generated graph and subagentdevices.com/architecture map: 11 graph edges missing from map, 1 reversed, 14 with no ARCH node — documented as routed backlog), cwc-product-management:dependency-contracts (contracted vs uncontracted runtime deps; live check vs subagentcontracts.com's 18 rows; uncontracted-edge backlog seeded).
Dogfood pass: adopt the build system + drift tooling on the real codebase
DONE same-session 2026-07-02. Ran the dependency-drift-audit loop for real; it caught 3 scanner precision bug classes in build-graph.py (generic property-name ident flow classifying static FAMILY_LINKSET_EXTRAS as fetches; 40-line fn-body approximation making esc() fetchy and nav links into fetch edges; loose fixpoint pulling subagentdomains SITES directory in) — fixed with brace-counted bodies, enclosing-const classification, strong seeds, collection-only propagation. runtime_fetch settled at 44 edges = the documented architecture exactly. Then fixed the real map drift on subagentdevices.com/architecture (13 CATALOG-primitive nodes with grep-verified accents, datamodels→api + contracts→tasks dogfood edges, layout v3) shipped THROUGH worker-build.sh (gate → deploy 543d0b38 → --record); arch-map-drift.py now exit 0. Then worked the PM dependency-contracts backlog: rotated CONTRACTS_WRITE_SECRET and created 5 open objective_check dependency contracts (ctr_dep_domains_live_counts, ctr_dep_contracts_tasks_join, ctr_dep_datamodels_catalog, ctr_dep_devices_dogfood, ctr_dep_api_health_fanout). Both skills + wireframes.json ledger updated with the resolved baselines.
SDK: extract agent-readiness endpoints (x402/AP2/JWKS) into kit
DONE 2026-07-02: workers/_shared/kit/agentready.ts (x402Demo/ap2Doc/botAuthDirectory/agentReady router; Fable-5-reviewed semantics + illustrative-labeling rule preserved in code comments). 5 sites refactored (api b51d5863, cli 5a671798, cowork 217c8f4d, coworkers 0afa3a9f, domains 102eda73), each through worker-build.sh — the gate correctly exit-2'd on the new kit dep before every deploy, locks re-recorded after. 15 endpoint JSON before/after comparisons: 13 EQUIVALENT, 2 intended normalizations (cowork/coworkers x402 description now matches the canonical wording — single-sourcing IS the point). New-kit-file strategy confirmed: zero unrelated workers went stale.
SDK: extract /api/changes cursor-feed boilerplate into kit
DONE 2026-07-02: workers/_shared/kit/changes.ts (changesFeed with structural D1Like type so the kit typechecks without ambient workers-types). contracts/tasks/memories refactored (4a420862/badd0bbd/05e33695), locks recorded, all 3 /api/changes?limit=2 responses JSON-EQUIVALENT before/after, cursor pagination re-verified live.
ST1: verify evaluate-workers.py output (62 posted rows)
ST1 done: py_compile clean; dry-run reproduces evidence rows; spot-checks match repo (min-width:0 x2 in roles src) and live row eval_auto_wc2026-bracket.
ST2: write scripts/resync-audit.py (sonnet agent died mid-stream)
ST2 done: scripts/resync-audit.py written inline by lead after the sonnet agent died mid-stream; two audit-precision fixes applied (kit llmsTxt emits family links — src grep alone was wrong; web/wc2026_bracket worker-name map).
ST3: run resync audit, capture 36 verdicts JSON
ST3 done: 36/36 audited, 34 pass / 2 fix_needed; verdicts JSON archived in session scratchpad; rerunnable.
ST4: close passing tsk_resync_* with evidence; file precise gaps for failures
ST4 done: 34 tsk_resync_* closed with per-site evidence; datamodels annotated as blocked on domain registration (tsk_b228566b-1c3); devices annotated with the refined adoption-matrix staleness fix.
ST5: close tsk_wire_evaluations_automation + final build check
ST5 done: parent closed; worker-build.sh --stale exit 0; both new scripts py_compile clean; no worker source modified this cycle.
cwc-finance: startup-finance skill (KISS cost ledger + Launchpad prep)
DONE 2026-07-02: plugins/cwc-finance/skills/startup-finance/SKILL.md — grounded in the real anthropics finance plugin skill set (variance-analysis/financial-statements/audit-support adapted to a monthly one-pager), financial-services-plugins inventory (claude-in-office = cloud-gateway admin tooling, financial-analysis = 11 modeling skills, honestly out of scope pre-revenue), and the operator's stated reality: two-line-item ledger (Cloudflare + $100/mo claude.ai), HARD no-ANTHROPIC_API_KEY rule, Workers Launchpad finance-half prep (submission enrollment_submit-gated). FINANCE_WRITE_SECRET rotated; 3 durable rows live on subagentfinance.com (kind CHECK only allows coworker_capability|plugin_skill — used plugin_skill). Registered in plugin.json.
cwc-legal: startup-formation skill (licensing/trademark/copyright groundwork)
DONE 2026-07-02: plugins/cwc-legal/skills/startup-formation/SKILL.md — adapted from knowledge-work-plugins legal skills (compliance-check/review-contract/legal-risk-assessment), claude-for-legal corporate-legal (entity-compliance) + ip-legal (clearance/portfolio/oss-review/takedown) + managed-agent-cookbooks (renewal-watcher/reg-monitor as routines: 50-domain renewal watch, terms-change watch). Hard not-legal-advice rule preserved. subagentlegal.com was READ-ONLY — added the gated POST /api/legal-facts route (build gate -> deploy 0e88c3bc -> lock recorded), LEGAL_WRITE_SECRET rotated, 4 startup_formation rows live (license status, trademark audit scope, launchpad terms review, portfolio protection). Registered in plugin.json.
Workers Launchpad investor deck (v1) — published
v4 2026-07-02: Cowork-docs grounding round (read from the local mirror docs/docs/claude.com/docs/cowork + plugins/overview.md — the doc subagentcoworkers.com is grounded in). Slide 7 lede now quotes Anthropic's own plugin definition verbatim (bundle MCP connectors, skills, slash commands, and sub-agents into a single shareable unit) and frames the 14 coworkers as that primitive instantiated per business function. Slide 8 gained the-leverage chip naming Dispatch in Anthropic's vocabulary (high-level instructions → tasks → separate sessions) as the mechanical explanation for 60+ verified deploys/day. Slide 12 foot now states the meta-proof: the company is built inside the surface it documents (Cowork project, standing instructions, durable memory). Deployed: worker 4d8d76b8 (gate clean, lock recorded, 3/3 markers verified) + artifact v4-cowork-grounding, same URL.
Propagate Cowork-docs grounding from deck v4 into subagentcoworkers.com pages
DONE 2026-07-02: the deck v4 grounding chain now lives in the site itself, quotes verbatim from the local docs mirror with canonical links. overview.md gained a Grounded-in-Cowork-primitives section (the unit = plugins/overview.md definition, named as this site's root grounding doc; the execution = guide/dispatch.md; the container = guide/projects.md mapped to this repo+CLAUDE.md+memory). getting-started.md cites guide/plugins.md as the install mechanism. plugins.md opens with the verbatim definition and frames the 14 rows as 14 instantiations of that sentence. how-it-works.md gained How-work-actually-runs: envelope is the contract, Dispatch is the engine. Hashes regenerated (integrity clean), worker-build gate clean, deploy 5f42613a, lock recorded, all 4 pages curl-verified live.
Fleet grounding sections: llms.txt ## Grounding on 16 primitives
DONE 2026-07-02: the deck/coworkers grounding pattern propagated to the machine surface fleet-wide. Each of 16 D1 primitives' llms.txt now carries a ## Grounding section (via the kit llmsTxt extra param — zero kit changes, zero collateral staleness): one verified grounding statement + canonical .md source URL (claude-tag how-it-works/agent-identity, platform citations/memory/define-outcomes/permission-policies/vaults/session-operations/tools/webhooks, plugins/overview for skills/roles/code). Quotes only where verified this session (thread-is-durable, fetch-the-object-directly); grounded-in fact statements elsewhere — no fabricated quotes. All 16 through worker-build gates, deployed (256c630f…cddb92d6), locks recorded, 16/16 curl-verified live, --stale clean. Agents reading any primitive now get the citation trail to Anthropic canonical docs — flywheel stage-6 in both directions.
PM strategy layer: product-strategy skill + deck v5 + use-case page
DONE 2026-07-02: cwc-product-management layered into the finance+legal strategy. (1) New plugin skill plugins/cwc-product-management/skills/product-strategy/SKILL.md — four moves: user≠buyer positioning, named north star (returning agent visits: change-feed re-polls + cite_as re-entries), gate-owned now/next/later (finance spend gate, legal attorney gate, PM north-star check), durable+cited strategy artifacts incl. the no-fabricated-TAM rule. (2) Deck v5: positioning lede on slide 3, north star named on slide 5, credit-honesty line on slide 9 (realizable-value discipline as investor credibility), slide 11 roadmap reframed now/next/later gate-owned. Deployed worker 4ffaa6ae + artifact v5-pm-strategy-layer, same URL. (3) New site page subagentcoworkers.com/platform/use-cases/product-strategy — the worked three-coworker example incl. the $100k correction story. Gotcha fixed en route: gen-hashes.py only rehashes LISTED manifest entries — new pages need a manual PLATFORM_DOCS entry (page 404d until registered).
WBR-W27 rec #2: extend /api/changes cursor feeds from 3 to all 24 D1 primitives
EXECUTED 2026-07-03, same week as filed (coworker-in-the-loop: engineering build gate is a coworker gate). 3 parallel agents wired changesFeed into 21 sites (identities/citations/cache/rubrics/outcomes/evaluations/brands + code/prompts/skills/roles/models/handoffs/permissions + vaults/sessions/tools/hooks/rust/python/typescript). Independent sweep: 24/24 /api/changes live, --stale clean. Notes: models+permissions+batch-C sites cite_as → list routes (no per-id routes); 9 outlier tsconfigs aligned (noUncheckedIndexedAccess removed, relaxation-only); kit untouched.
Sales layer: agent-gtm skill, territory book, deck v7, feeds 3→24 executed
DONE 2026-07-03: cwc-sales layered onto the strategy stack, grounded in two fetched Anthropic posts (GTM engineering / 4,000-account book) + upstream sales skill set. (1) plugins/cwc-sales/skills/agent-gtm: two account classes (inventory zones vs buyer operators), gate-owned Gateway pipeline, weekly forecast into WBR, Bryant split (book ranks, human decides). (2) scripts/territory-score.py — the CLAFTS move — scored 49 zones live: 24 Tier A / 1 B / 24 C (reports/territory/2026-W27.md); building it surfaced the Python-urllib 403 bot-protection contradiction (tsk_0536fa28-a30). (3) WBR rec #2 EXECUTED same week: 3 agents, 21 sites, 24/24 feeds verified (coworker-in-the-loop per operator directive). (4) Deck v7 GTM line + /platform/use-cases/agent-gtm page live (worker 759016e7, artifact v7-sales-gtm-layer).
Marketing layer: creative-production skill, HB-01 hero band (deck v8), Higgsfield briefs ready
DONE 2026-07-03: cwc-marketing layered in, grounded in claude.com/blog/how-anthropic-uses-claude-marketing (Austin Lau one-click Figma variations, 30min→30s) + upstream marketing skill set. Agency rules: brief before pixels, palette-locked terminal-noir, no humans/logos/text-in-image, provenance tags, batch-then-select, publish gates. HB-01 executed in-medium: deterministic seeded SVG (166 nodes/325 traces, seed 20260703) live on deck slide 1 (worker 3bb19b72 + artifact v8) with foot provenance. HIGGSFIELD STATUS, honest: MCP at mcp.higgsfield.ai 401s from this session (connector OAuth is account-side, session predates it), Chrome extension disconnected, no local token cache — new OAuth grant deliberately NOT self-initiated (operator-gated class). BRIEFS.md (HB-01/OG-01/VL-01) checked in production-ready: any fresh session with the connector, or a connected Chrome with the operator logged in, executes them verbatim and A/Bs against the same checklist.
Deck layout-system fix (v9): overflow-safe centering, density layer, verified in-browser
DONE 2026-07-03 by cwc-engineering after operator bug report (sizing/overlap/wrapping). ROOT CAUSE: justify-content:center on min-height:100vh flex slides — content grown past one viewport (v3-v8 layers) overflows upward into unreachable clipped space; scroll-snap mandatory traps tall slides. FIXES: (1) auto-margin centering trick (centers short slides, top-anchors tall — never clips), (2) snap mandatory→proximity, (3) eyebrow flex-wrap + 72px slideno clearance, (4) height-responsive density layer @max-height:900px (type scale, gaps, paddings, single-row g2 grids, compact ledes, hero band 20vh cap), (5) wrap-safety (overflow-wrap:anywhere) on chips/foot/small/ul per design system, (6) safe-area insets, (7) kbd-nav rect-based. VERIFIED with real browser via Preview harness at 1280x760 + 390x844: zero clipping, zero slideno collisions, zero horizontal overflow outside scroll containers, doc scrollW = viewport on mobile; tall slides 7→2 (max +6%, graceful); mobile screenshot visually confirmed. Deployed 86294384, lock recorded, artifact v9-layout-system-fix same URL.
cwc-engineering: mac-toolchain (typed doctor) + preview-layout-verify skills
DONE 2026-07-03. (1) mac-toolchain: the darwin/zsh/Homebrew environment missing from the durable-toolchain-doctor/install family — closed-schema toolchain.json (id/kind/purpose/fixes/check/install/gate, every entry citing a real dated failure: No-Chrome-binary-for-devtools, CERTIFICATE_VERIFY_FAILED, zsh no-matches aborts, sccache bypass) + stdlib-only validate.py (NoReturn-typed failures, unknown fields rejected, --doctor runs checks verbatim). Live first run: 2/6 present (jq, rsvg-convert — doctor corrected my own wrong assumptions), 4 missing with exact gated fixes (chrome + certifi + 2 zshrc lines await operator approval — system mutations are operator-gated). Type-safety contract per operator directive: the model copies validator output verbatim, never free-recalls install strings — recorded as standing memory typed-manifests-over-recall. (2) preview-layout-verify: the Claude Preview harness workflow that verified deck v9, packaged with the three invariant measurements and all four harness gotchas (reload-in-own-eval, scrollTo instant, black-screenshot-after-scroll, inspect-over-screenshot). Both registered in the plugin README.
OG-01 executed via Higgsfield (free tier): share card live + og meta; deck v10
DONE 2026-07-03, connector live mid-session. Balance-first discipline (10 credits, free plan): recraft_v4_1 preflighted 1.25cr → 403 plan-gated (honest constraint); z_image preflighted 0.15cr → 2-variant batch (~0.3cr total). Selected variant B per OG-01 checklist (right-third cluster, quiet left overlay zone); 852 purple negative-list pixels retouched to brand green via stdlib-only PNG codec (no Pillow on this Mac — durable-mac-toolchain candidate entry if recurring); cropped 1.91:1, 109KB JPEG (under 300KB budget). Live: subagentcoworkers.com/deck-og.jpg (200) + og/twitter meta on /deck (worker f2613480) + provenance in deck foot + artifact v10 same URL. A/B verdict recorded in BRIEFS.md: seeded SVG stays as the in-page hero band (crisper vectors); generated art wins as the share card. VL-01 (8s video loop) pending operator call vs ~9.7 remaining free credits.
CL-01 closing band via Higgsfield CLI: three-coworker production, deck v11
DONE 2026-07-03. The design→marketing→engineering chain the operator asked for, end to end: cwc-design authored brief CL-01 (slide-12 bookend to HB-01 — slide 1 converges INTO the cluster, the close radiates OUT of it) into BRIEFS.md before any pixels; cwc-marketing produced via the operator-installed higgsfield CLI (workspace selected, 9.7cr free plan, cost preflighted 0.15cr, z_image single shot passed every checklist line after the standard 80px purple retouch — budget deviation from batch-then-select recorded, not hidden); cwc-engineering integrated (band crop 2048x590 → 78KB JPEG → data URI so worker AND artifact carry it identically), re-verified layout via preview-layout-verify (slide 12 still exactly one viewport, image loaded, zero new invariant violations), deployed 2ae765e8 + artifact v11. Also: higgsfield-cli added to durable-mac-toolchain manifest (doctor now 4/7 PRESENT); project-scoped .agents/skills/higgsfield-* from npx skills add are the CLI's companions. Total Higgsfield spend this session: ~0.45 of 10 free credits; ~9.55 remain.
Toolchain 7/7 + CH-01 coworker critters on deck slide 7 (v12)
DONE 2026-07-03. (1) Operator granted install permission: all 3 escalated durable-mac-toolchain entries installed verbatim from the manifest (google-chrome cask, zsh NO_NOMATCH+EQUALS, RUSTC_WRAPPER bypass) — doctor now 7/7 PRESENT; chrome-devtools MCP unblocked for future verification. (2) CH-01: operator's spin on Anthropic's Fable-5 pixel-critter Instagram announcement — 14 coworker critters, one per plugin in its accent color. Legal guardrail applied: distinct ORIGINAL character (upright body, signal antenna, terminal-prompt chest mark — not Anthropic's eared quadruped), inspiration credited in the deck foot, pixels not traced. Fully typed+deterministic: pixel grid as string constants, closed name→hex map (no purple), zero RNG. Verified: 14 render at 24px compact, visual check confirms family reads; slide 7 scrolls 11% (honest note), no overflow. Worker 16432378 + artifact v12, BRIEFS.md CH-01 recorded.
Critters v2 (buddy-inspired) + team-portrait poster: deck v13 + plugins-page hero
DONE 2026-07-03. Redesign sourced from vendors/anthropics/claude-desktop-buddy ASCII species (expression/feature-variant pattern): rounded body, dark pixel eyes, pixel smile, two feet, closed variant enum antenna/ears/sprout/crest cycling the 14 — typed string-constant grids, zero RNG. Deck v13 slide 7 carries v2 (crisp SVG, tooltips). Team-portrait poster composited in-medium (interleaved 2-row group, per-critter floor reflections, rim-light vignette, rsvg-rasterized 1680px → 101KB JPEG) now the hero on subagentcoworkers.com/platform/plugins (asset /coworkers-team.jpg, 200). HONEST plan-gate finding: Higgsfield reference-image models (nano_banana_2_lite, recraft_v4_1) 403 job_minimum_basic_plan_required on free tier — text-only z_image is the free surface; plan upgrade = finance spend_commit decision, recorded in BRIEFS.md. Worker f988dcb9, artifact v13, all curl-verified.
Higgsfield free-tier map probed + deck visual wrap-up (favicon, durable builder)
DONE 2026-07-03. Free-tier inventory established by real create-gate probes (cost quotes lie — several models quote then 403): WORKS FREE = z_image 0.15cr, flux_2 1cr (quality headline), bytedance_image_upscale 2cr; PLAN-GATED = soul_location/nano_banana v1+lite/recraft (all reference-image + environment paths); video 4.8-7.5cr untested (one try = half balance, VL-01 stays operator call). Upscaler probed ON the team poster: crisp edges but erased pixel-grid texture + noised reflections → original rsvg render kept, upscaler verdict = photographic assets only. Durable record: plugins/cwc-marketing/skills/creative-production/FREE-TIER.md (probe dates + balance + ≥5cr reserve rule). Deck wrap-up: cyan critter SVG favicon on /deck (in-medium, 0cr) + scripts/deck-build.py checked in as THE single deck builder (head og/favicon can no longer fork from session-inline wrappers). Balance 6.55cr. Workers ec4f7ddb→0b69853f, locks recorded.
cwc-agent-resources (15th coworker) + cwc-operations incident-runbook layered in
DONE 2026-07-03. (1) cwc-agent-resources: NEW plugin adapting anthropics human-resources to a coworker workforce — coworker-lifecycle skill (recruit=when a domain earns a plugin w/ real precedents cwc-deploy birth + cwc-founder dissolution; onboard=bootstrap checklist; evaluate=rubric rows on subagentevaluations.com; grow=scars-not-wishes + gates.toml grants as comp; dissolve=founder precedent). Dogfooded: its first lifecycle action was onboarding ITSELF — plugin files, marketplace (15 registered), durable identity id_0143a85e-387 on subagentidentities.com, critter #15 (bright teal #5eead4, ears) via the NOW-CHECKED-IN typed generator scripts/critters.py, roster 14→15 across deck (v14: lede, slide 7, caption, foot) + plugins.md + regenerated 15-critter team poster. (2) cwc-operations incident-runbook: upstream runbook/change-request/status-report grounded in the fleet real incident log (disk-full, silent wrangler exit, Python-UA 403, stale-edge rule, gen-hashes gotcha, sccache). Worker 56b36afd, artifact v14, use-case page live, all curl-verified.
Scrapy docs mirror-keeper: bloom+sha256 CDC crawler, live-proven, shipped as skill
DONE 2026-07-03. crawlers/ package on SitemapSpider: md-first per URL (probed live: platform.claude.com serves real .md → source:md; claude.com/blog|customers|connectors|skills|plugins + anthropic.com/news + support 404 the variant → deterministic stdlib HTML distiller fallback). Skip logic exactly as requested: persisted deterministic bloom filter (sha256-derived k-hashes, 512KB bitset, pure stdlib vs awesome-scrapy Redis variants) = cross-run URL skip; sha256 manifest = write skip (unchanged pages never rewrite, git stays clean). PROVEN live: 10 blog pages written → rerun bloom-skipped all 10 with zero fetches → recrawl=1 re-fetched and SKIP-SHA d all 10 → platform target wrote 5 real-markdown docs. Typed TARGETS table (new mirror = one dict entry). Two real bugs found+fixed live: 404 .md responses never reached parse (handle_httpstatus_list) and the bloom filtered its own sitemap request on rerun (discovery layer now exempt — new posts stay discoverable). Skill+command: /scrapy-docs-crawler in cwc-engineering. ALSO: disk-full incident #2 mid-build (Chrome cache 4.3GB 1h after cask install; Desktop Commander proved to be the zero-disk break-glass shell) — logged in incident-runbook. scrapy 2.13.4 recorded in durable-mac-toolchain (7→8 entries).
Crawler CommonMark refactor: converting distiller + markdown-it-py parse gate
DONE 2026-07-03 per operator. Distiller rewritten (crawlers/claude_docs/md.py): piecewise inline converter — <a>→[text](url) with absolutized+percent-encoded hrefs, strong/em/code→**/*/backticks — and spec backslash-escaping on every plain-text run (adversarially tested: ](\[evil](x) injection neutralized). New cm.py gate: markdown-it-py parses every DISTILLED doc; any token outside CommonMark core (esp. html_block/html_inline) → COMMONMARK-REJECT, write refused. Vendor markdown (source:md) written as-is with beyond-core tokens logged only. Design-system rationale encoded: core-only subset renders 1:1 in Chrome 151 canary AND swift-markdown/swift-markdown-ui (iPhone 16 Pro path). Live-proven: 10 blog pages recrawled → 10 WROTE, 0 rejects; link conversion surfaced previously-invisible nav junk → measurement fix (URLs excluded from length floor) + noise entries; platform vendor-md logged html_block/image, sha-skipped unchanged. markdown-it-py recorded in toolchain (9 entries).
Full 4-sitemap crawl + 3-tier RAG cache + mined cowork directives applied
DONE 2026-07-03. CRAWL: 1,101 pages written (claude 272 / anthropic 234 / support 300 / platform 295), manifest 1,116 URLs, 25+5 bloom skips, 5 CommonMark rejects — all one real distiller bug (bare < in prose parsed as html_inline), fixed (real-tag-only stripper + < escaping, unit-proven) and refetched via new crawlers/refetch.py (targeted single-URL path through the same gate+CDC). RAG: crawlers/ragcache.py three-tier cache per operator spec — L1 functools.lru_cache (deserialized) / L2 redis serialized TTL 1h (server started, round-trip proven 0.2ms fresh-process) / L3 sqlite FTS5 bm25 — 7,409 docs indexed, sha-incremental (+795, 6,614 skipped). docs-rag skill + /docs-rag command: query-before-read token-thrift directive. MINED + APPLIED: services-track-partner-hub → agent-gtm second-channel section (Select tier honestly 1/10 certifiable today; their Partner Hub MCP connector = our pattern as prior art) + coworker-lifecycle credential-recency (90-day active rule) + deck v16 PwC proof point (Claude Code + Cowork toward hundreds of thousands, live both surfaces, worker 27159541). NEXT candidates surfaced by first queries: anthropics-finance-team blog post (cwc-finance directive), Cowork computer-use support articles. Toolchain 11/11.
Deck v17: final six coworker layers via capped structured subagent dispatch
DONE 2026-07-03. The six coworkers with no deck layer (operations, deploy, analytics-engineer, customer-support, human-resources, science) each dispatched as an atomic capped subagent per operator spec: sonnet-5 ×4 + fable-5 ×2, ≤8 tool calls each, actuals 103-113k subagent tokens, 3-5 tool calls, every proposal REQUIRED to ground via ragcache queries (docs-rag dogfood — all six cited index-returned paths; science closed the full loop: crawl→index→query→deck line citing the mirrored Claude Science doc). Structured JSON contracts (anchor_quote/insertion/grounding/rationale); central review applied all six with two editorial changes recorded: deploy proposed-new-chip trimmed to a sentence in the leverage chip (slide-8 height), support 403 zinger status corrected to pending-operator-call. Fable self-truth-checking observed: support agent verified its zinger against the incident log and adjusted phrasing unprompted. Layout re-verified in-browser: slide 7 accretion (1019px) fixed via chip pairing @short-heights (998→807); final talls 782-878 = graceful ≤15%, zero clip/overflow. All six lines curl-verified live (worker b247878a, artifact v17). The deck now carries a contribution from every one of the fifteen coworkers.
IDENTITY.json + BRAND.json in all 15 cwc-* plugins — typed, claude-tag-grounded, deck-integrated
DONE 2026-07-03. plugins/_schema/: two closed JSON Schemas + generate.py (typed table; critter accent/variant IMPORTED from scripts/critters.py; gates PARSED from each gates.toml — both fleet formats after the validator caught the array-of-tables misparse; deck_contribution = each coworker real v17 line) + stdlib validate.py (schema checks WITHOUT jsonschema dep + cross-file invariants: IDENTITY accent == BRAND accent == critters.py, grounding paths must exist on disk, 15/15 coverage). IDENTITY grounds claude-tag primitives from the mirror: attribution_mode service_account (agent-identity.md), directive_layers mapping attach-to-scope three layers onto SKILL.md-rules/durable-rows/dispatch-prompts. BRAND = design system as data: fixed palette consts, shared voice, per-coworker terminology, example_good = the REAL deck line (voice proven in production, never invented). Validator caught 2 real issues before ship: reports/ missing from grounding roots, and the gates format misparse. 30 files, exit-0 coherent.
Coworker schema v2: skills-as-core + tri-language mirrors (Pydantic/Zod4/serde)
DONE 2026-07-03 per operator refactor. IDENTITY v2 centers Anthropic's Agent Skills primitive from the crawled mirrors — verbatim: a Skill packages instructions, metadata, and optional resources (scripts, templates); progressive disclosure; scripts provide deterministic operations without consuming context (the operator's thesis in Anthropic's words). skills[] now DISCOVERED from each plugin's filesystem (SKILL.md frontmatter incl. YAML block scalars after a real parse bug, bundled scripts, reference files) + skill_model grounding block. Tri-language mirrors per the fleet pattern: scripts/coworkers/schema.py (Pydantic v2, extra=forbid, validates 30/30), schema.ts (Zod 4 strictObject, tsc --noEmit + runtime parse 30/30, local package), crates/schema/src/coworker.rs (serde deny_unknown_fields + schemars + cargo test deserializing all 30 REAL files, 15/15 assert) — plus the stdlib validator for zero-dep environments. One writer (generate.py), four gates; drift precedent (agent.rs/schema.ts) cited in every mirror header. pydantic recorded in toolchain (12 entries).
python-build-standalone migration + multilspy LSP + code-mode/PTC doctrine
DONE 2026-07-03. (1) astral-sh/python-build-standalone 3.13.14 installed to ~/.local/pbs, PATH-fronted in zshrc — THE python; full package set (scrapy/markdown-it/redis/certifi/pydantic/multilspy/jedi) self-contained in its prefix, ending the 3-interpreter drift (CLT 3.9 vs Framework 3.13 vs /usr/local shim) that broke imports mid-crawl. Framework/shim removal needs operator sudo (documented; PATH-shadowed meanwhile); CLT 3.9 stays (Xcode tooling). Toolchain: pbs entry (#1) + all python_pkg installs de---user-ed. (2) multilspy installed; scripts/lsp_sweep.py — per-scope LSP workspaces after the repo-root workspace made jedi index 6,600 mirrors per query (7min timeout, real lesson). (3) Four docs mirrored via our own refetch pipeline incl. PTC notebook ipynb→md converter; code-mode-ptc skill encodes the doctrine (CLIs over resident plugins — the LSP plugins point; PTC batching; fable orchestrates/sonnet executes; ragcache first). Sonnet verification matrix dispatched (all gates under pbs + bounded sweep rerun).
e2m loop envelopes + bounded LSP sweep complete (4s, was 46min silent)
DONE 2026-07-03. scripts/e2m.py: envelope=typed shape (id/agent/loop_type per the Claude Code loops taxonomy turn|goal|time|proactive/state/trigger/stop_criteria/budget/counters/ts), mailbox=durable version-controlled JSONL at reports/mailbox/<agent>.jsonl; `e2m.py open` alerts on runs with no terminal envelope — silence IS the alert. Loops post mirrored via refetch. lsp_sweep rewritten clean (4 sequential patches had frankenlayered two main paths — patch archaeology stopped, honest rewrite): outermost SIGALRM bounds STARTUP too (a jedi server hang outlived the loop budget by 40min), per-query alarms, scope heartbeats to mailbox + per-file .progress. Result: 84 symbols in 4 SECONDS under pbs, envelope lifecycle started→heartbeat×3→completed in the mailbox. 3 zero-ref candidates = all verified false positives (string-wired framework refs: DUPEFILTER_CLASS, scrapy spider name) — no deletions; the delegates-must-verify rule proven. Runbook gains the e2m section.
coworkers-console: localhost HTML entry point to the workforce (dogfooded live)
DONE 2026-07-03. apps/coworkers-console/ per the html-effectiveness pattern (post crawled+mirrored first): single standalone index.html (inline CSS/JS, terminal-noir tokens) + stdlib serve.py (binds 127.0.0.1; /api/coworkers reads all 15 IDENTITY+BRAND, critter SVGs imported from scripts/critters.py single-source; /api/mailbox + /api/alerts = e2m envelopes with silence-alert logic; /api/reports = newest WBR/territory/sweep/toolchain). Dogfooded sequentially: built → preview_start → measured (15 cards, 5 envelopes, 4 report sections, 0 errors, critters rendering) → modal verified (directives/gates/skills/deck-line) → screenshot confirmed. Read-only v1; gated POST /api/dispatch is the recorded v2 candidate (operator decision). launch.json committed as the durable launch config. Skill in cwc-agent-resources.
Per-coworker filesystem memory with dream-only CRUD — spec-driven, all criteria pass
DONE 2026-07-03. Operator prompt rewritten as a structured prompt per mirrored best practices (role/grounding/tasks/constraints/success-criteria/non-goals) at plugins/_schema/MEMORY-SPEC.md — which doubles as the spec. scripts/coworker_memory.py: SQLAlchemy+sqlite index & FTS5 search & lru reads over filesystem-truth stores plugins/cwc-*/memory/v<N>/; CRUD ONLY through dreams (v(N+1) new dir, inputs never mutated per dreams.md; 100KB/2000 caps per memory.md as hard errors; e2m envelopes per dream = office-agents audit telemetry; claude-tag channel-memory layer made literal). Seeded 15 coworkers via 15 real dreams. ALL FIVE success criteria verified live: v1+current ×15, v1 byte-identical after finance v2, 100001-byte put rejected with byte count, warm search 0.00ms, mailbox all-terminal incl. the failed cap-test envelope. LSP sweep re-covers the module. Follow-up recorded: sync local dreams → subagentmemories.com API when write secret provisioned; console memory rendering = v2 candidate.
Refactor T1: one shared HTTP fetch util (3 hand-rolled copies)
EXECUTED 2026-07-03 (delegate stalled mid-stream at 7 read-only calls — clean tree confirmed, completed inline). Shared modules crawlers/claude_docs/fetch.py (UA/certifi/timeout/max_bytes; 3 copies eliminated) and fm.py (frontmatter incl. block scalars; 2 copies eliminated, the >- bug documented in the docstring). Gates all green: refetch smoke no-traceback, ragcache build 7412 unchanged, generate→validate→pydantic OK, IDENTITY regen byte-level no-op (parser parity proven), cm unit true, max_bytes fetch path 200. Net: ~90 duplicated lines removed, ~85 shared added once.
Refactor T2: one frontmatter parser (2 hand-rolled copies)
EXECUTED 2026-07-03 (delegate stalled mid-stream at 7 read-only calls — clean tree confirmed, completed inline). Shared modules crawlers/claude_docs/fetch.py (UA/certifi/timeout/max_bytes; 3 copies eliminated) and fm.py (frontmatter incl. block scalars; 2 copies eliminated, the >- bug documented in the docstring). Gates all green: refetch smoke no-traceback, ragcache build 7412 unchanged, generate→validate→pydantic OK, IDENTITY regen byte-level no-op (parser parity proven), cm unit true, max_bytes fetch path 200. Net: ~90 duplicated lines removed, ~85 shared added once.
Self-steered reduction pass: −401 lines, −5 files on PR 21
DONE 2026-07-03. (1) Schema surfaces 5→3+1: deleted coworker-identity/brand .schema.json + the ~100-line hand-rolled JSON-Schema walker; validate.py now orchestrates Pydantic (canonical) + cross-file invariants only (critter parity, accent coherence, grounding-paths-exist). Zod4 + serde mirrors unchanged; all gates green incl. cargo test over the 30 real files and a regen no-op. (2) Run artifacts out of git: lsp-sweep.json/.progress + memories.sqlite gitignored+uncached (all rebuildable; dream audit trail lives in committed mailbox JSONL). Combined with the T1/T2 dedup pass: session reduction ≈ −490 lines, −5 files, 5 hand-rolled copies eliminated.
Extend agent-schema-drift to cover D1 primitive schemas, not just AgentConfig
agent_schema_drift.py currently only cross-checks agent.rs/schema.ts/schema.py/schema.java for the AgentModel/AgentConfig/EffortLevel enums (confirmed clean this session: 5 enums, 27 plugin agent frontmatter files all in sync). It has no check for drift between a D1 primitive's actual table schema (e.g. workers/subagentroles/migrations/0001_init.sql) and the TypeScript types the worker's own src/index.ts assumes. Add a second check mode that diffs a migration's CREATE TABLE columns against the fields referenced in that worker's api handler.
Wire agent-schema-drift into a scheduled fleet-wide check instead of on-demand only
The script is currently run manually. Adopt this as a weekly scheduled task so drift is caught within a week instead of whenever someone remembers to run it — CLAUDE.md documents at least 2 historical drift incidents caught manually, after the fact.
Add the 5th (Swift) mirror to agent-schema-drift's checked set
CLAUDE.md documents the Swift mirror as deliberately NOT yet part of the automated drift check. Extend agent_schema_drift.py's mirror list to include it once scripts/agents/schema.swift exists.
Add a snapshot-age check to the analytics-playbook skill
subagentcache.com's cache_snapshots table has a source column to avoid claiming API-fetched numbers without a token. Refactor analytics-playbook to warn when the most recent manual_dashboard_paste snapshot is more than 14 days old.
Extend cache_snapshots-style provenance tagging to the flywheel-metrics.sh output
scripts/flywheel-metrics.sh produces real numbers but never persists them anywhere durable — confirmed by rerunning it this session, it just prints to stdout. Adopt a new snapshot type (script_run) so flywheel metrics accumulate a real trend line.
Verify Cloudflare AI Crawl Control is enabled on all zones with a live public site
Execute the actual per-zone verification across the ~70 live subagent*.com zones and record which are/aren't enabled as a real cache_snapshots row.
Update support-playbook's known-issue list with the confirmed-stable metrics baseline
Refactor it to document the confirmed-stable flywheel baseline (70/71 discoverability, 25/71 MCP, 49-entry catalog, re-verified twice with zero drift) so future triage can tell known-state from genuine regression.
Extend fleet-inbox-digest to flag workers still missing an MCP endpoint
Adopt a new check: cross-reference the live MCP handshake list (25/71) against the full worker roster and flag any D1-backed primitive that still lacks one.
File a real task for the one domain (subagentdatamodels.com) that fails all discoverability checks
scripts/flywheel-metrics.sh consistently shows 70/71 — the 1 miss is subagentdatamodels.com, root-caused as an unregistered domain. File this as a correctly-scoped triage task.
Refactor the warehouse-analyst skill's schema-gotchas section for the 2 latest primitives
cwc-data's semantic-layer-first SQL discipline is scoped to subagentjobs-dwh only. Refactor the schema-gotchas doc to explicitly note subagentmemories/subagentmodels are OUT of scope by design, so a future session doesn't assume DW coverage that was deliberately excluded.
Adopt the fact_repo_files code index (914 rows) as a queryable source in the analyst skill
Migration 010 added fact_repo_files (FTS over workers/plugins/crates/scripts, confirmed live via indexer this session). cwc-data's analyst skill doesn't yet mention this table as a valid query target — adopt it into the semantic layer docs.
Run a real query confirming fact_orgs/fact_repositories/dim_packages seed counts still match CLAUDE.md's documented numbers
CLAUDE.md states fact_orgs=6, fact_repositories=24, dim_packages=54 as of migration application. Execute the actual count query against the live Postgres DB and confirm no drift, or file a correction if it has drifted.
Refactor session-telemetry-erd to include subagenttasks.com's tasks/task_dependencies/sessions tables
The ERD skill models the Coworkers board's own data model but has not been extended to the now-live subagenttasks.com D1 schema, which is the durable backing store for exactly this kind of telemetry. Refactor to add it as a documented sibling schema.
Adopt subagentcache.com's dual value+label column pattern as a documented DW convention
cache_snapshots stores both a numeric value and its original display string (e.g. cached_requests_count + cached_requests_label). This is a real, useful pattern for any future dashboard-paste ingestion — adopt it into cwc-data-engineering's own conventions doc rather than leaving it implicit in one worker.
Verify migration 006's dim_packages last_updated_at/is_archived/is_deprecated fields are populated, not null, for all 54 rows
Execute a real query against fact_orgs/dim_packages in the live Postgres DB checking for null-heavy columns that would silently degrade any dashboard built on top of this table.
Refactor the deploy-playbook's troubleshooting table with the account_id pin gotcha
CLAUDE.md documents 4 workers that silently failed non-interactive deploys due to a missing account_id (fixed 2026-07-01). Confirm cwc-deploy's playbook already covers this (it references scripts/cf-zone-id.sh) and add the WRANGLER_LOG=debug diagnostic step explicitly if missing.
Adopt scripts/cf-zone-id.sh --all as a required pre-deploy step for any NEW domain
The script already resolves zone IDs via a fixed fallback order (env var, both wrangler config paths, Docker MCP OAuth). Adopt --all as a standing pre-flight check in the deploy playbook so a missing zone ID is caught before a deploy attempt, not after.
Redeploy the 25 MCP-enabled workers to ship the readOnlyHint/destructiveHint annotation fix
workers/_shared/kit/mcp.ts's toolDef() now sets annotations:{readOnlyHint:true,destructiveHint:false} (edited this session, confirmed exactly 2 internal call sites, no external consumers). This is a real, already-written, not-yet-deployed source change. Execute the actual Desktop Commander + wrangler deploy sequence for all 25 workers currently serving a live /mcp endpoint, and curl-verify the annotations field appears in a fresh tools/list response for at least 3 of them.
Refactor the design-critique skill's hard-gate criteria to include the grid-overflow class of bug
scripts/design-system-audit.sh added a 3rd hard-gate criterion (min-width:0 + overflow-wrap co-occurrence) after a real user-reported bug on subagentdomains.com. Confirm cwc-design's own critique skill references this specific gate, not just the older P3/reduced-motion checks.
Adopt the --teal/--slate/--gold/--rose/--lime/--cobalt/--coral/--sage/--blush/--stone/--brass/--copper/--rust/--azure/--mint/--violet/--indigo/--jade/--peach accent palette as a documented registry
18+ accent colors have been assigned ad hoc, one per new site, across this session's history. Adopt a single documented accent-color registry (in the design system skill) so the next new site picks an unused color deliberately instead of by grep-checking every prior CLAUDE.md entry.
Run design-system-audit.sh fleet-wide and report current pass/fail state
CLAUDE.md documents this passing clean (exit 0) across all 36 sites as of the grid-overflow incident fix. Execute a fresh run now that the fleet has grown to ~71 sites and report whether it's still exit 0 or has drifted.
Refactor docs-rag's SKILL.md to document the search_index MCP tool's --corpus flag
CLAUDE.md references 'plugins/cwc-engineering/skills/docs-rag/SKILL.md § Code index (fact_repo_files)' for the --corpus-scoped indexer query. Confirm that section is current and refactor if the corpus list (workers/plugins/crates/scripts) has changed.
Adopt a documented list of which workers still import wellknown.ts vs. have it hand-rolled inline
The Content-Signal/llms.txt/robots.txt routes were added via 6 parallel dispatched agents touching all 36 sites at the time — some may have received hand-rolled inline versions rather than importing the shared kit. Adopt a definitive list (via grep) so future shared-kit edits know their real blast radius.
Grep the full workers/ tree for any remaining bare-hex CSS values outside the documented token system
CLAUDE.md documents a full retrofit of subagentjobs.com/subagentdb.com/subagentworkers.com off bespoke hardcoded-hex CSS, with 2 known cleanup spots (.bcard:hover, .d-apply). Execute a fresh grep for hex-color literals across all worker src files and report any new stragglers among the ~71 total sites.
Refactor the cost-tracking skill to record Cloudflare Workers paid-plan usage, not just external vendor spend
cwc-finance currently tracks third-party vendor spend. Refactor to add a line item for the account's own Cloudflare Workers Paid plan (needed once the ~71-site fleet exceeds free-tier request/CPU limits), since this is a real, currently-untracked recurring cost.
Adopt subagentcache.com's live analytics numbers as an input to a real cost-per-request estimate
subagentcache.com already has real cached-bandwidth/cached-requests numbers (49.24k requests, 432.31 MB bandwidth) from a manual dashboard paste. Adopt these as a documented cost baseline in cwc-finance rather than treating fleet cost as unknown.
Check whether any of the ~71 D1 databases are approaching Cloudflare's free-tier row/storage limits
Execute a real per-database size check across the fleet's D1 databases via wrangler d1 info and flag any approaching the free-tier ceiling before it becomes an outage.
Refactor cwc-hardware-engineering's scope note to explicitly state it has no current workers/ connection
Checked directly this session: no worker in workers/ references hardware/device-driver concepts distinct from subagentdevices.com's CSS/viewport progressive-enhancement catalog, which is a design-system primitive, not a hardware-engineering one. Refactor the plugin's README to say so plainly rather than implying overlap that doesn't exist.
Adopt subagentdevices.com's grep-verified adoption-matrix pattern as a template for any future hardware-integration site
subagentdevices.com's /adoption page is built by grepping actual marker strings per site rather than asserted from memory. If cwc-hardware-engineering ever gets a real workers/ target, adopt this exact verification discipline rather than a hand-written claim.
Confirm via a fresh grep that no workers/*/src/index.ts references GPIO, serial, USB, or BLE APIs
Execute this specific negative-confirmation grep once, to close the ambiguity definitively rather than leave 'probably no connection' as a standing assumption. Record the result (even if it's 0 matches) as the actual evidence.
Refactor the org-planning skill to reflect the current 20-plugin coworker roster, not the historically undercounted 8-9 agent figure
CLAUDE.md documents a real historical undercount (design-coworker missing from an agent list, later corrected to 9). Refactor cwc-human-resources' own roster references to cross-check against subagentroles.com's live /api/roles rather than a hardcoded number in a skill file.
Adopt subagentroles.com's role/agent-definition cross-reference pattern as the canonical org chart source
subagentroles.com already cross-references its 9 domain-coworker roles against the 9 real agent definitions by id. Adopt this live API as cwc-human-resources' org-chart data source instead of a static doc that can drift.
Query subagentroles.com and subagentcode.com's live APIs and reconcile any mismatch between role count and agent-definition count
Execute the actual GET /api/roles and GET /api/agent-definitions calls and diff the results — CLAUDE.md's own agent-schema-drift check only covers model enums, not this specific role-vs-agent-count reconciliation.
Refactor the swift-conduit skill's versions.json references now that Swift-on-Linux is proven
CLAUDE.md documents the conduit's Swift package proving Swift-on-Linux via a Lima VM build. Refactor swift-conduit's SKILL.md to move this from 'in progress' framing to 'proven, see reports/conduit/' framing, since the fact has changed.
Adopt subagentswift.com's conduit_runs table as the durable log target for future OODA iterations
subagentswift.com's migrations/0003_conduit_runs.sql already exists and is live (first row is the real W3.4 Linux proof). Adopt this as the standing durable sink for cwc-ios-engineering's OODA cycle output instead of only reports/conduit/ooda.jsonl.
POST a fresh conduit_runs row confirming the Lima VM Swift toolchain still answers swift_toolchain_info correctly
Execute a live re-check of the Swift-on-Linux proof (the same binary, same MCP tool call) and record the result as a new dated row via subagentswift.com's gated POST /api/conduit-runs, rather than relying on the one historical proof row staying representative forever.
Refactor cwc-legal's contract-triage skill to reference subagentcontracts.com's real done_kind taxonomy
subagentcontracts.com's Contract primitive uses a real 4-value done_kind enum (objective_check/human_approval/human_choice/unclosable) grounded in Claude Tag's own definition-of-done table. Refactor cwc-legal's own contract/NDA triage language to use these exact terms instead of inventing separate legal-specific status labels.
Adopt workers/subagentcowork's /integrity drift-detection pattern for legal doc-mirror staleness checks
subagentcowork.com's sha256:16 content-hash table detects drift against upstream Anthropic docs it mirrors. Adopt the identical mechanism for any legal-relevant mirrored doc (e.g. Software Directory Policy, Terms) cwc-legal references, so a policy change upstream is caught automatically.
Diff docs/docs/claude.com's mirrored Software Directory Policy/Terms pages against the live pages
Execute a real fetch-and-diff of the two legal reference docs already read this session (review-criteria.md, what-to-build.md) against their live claude.com sources to confirm no drift before relying on them for any future connector submission.
Refactor agent-traffic-flywheel's stage-0 fitness table now that the readOnlyHint fix is deployed
This session's own SKILL.md edit marked the annotations fix as 'FIXED IN SOURCE, NOT YET DEPLOYED.' Once cwc-deploy's execution task above ships it fleet-wide, refactor this line to 'FIXED AND DEPLOYED' with the actual deploy date and curl evidence, per the flywheel's own honesty-rules discipline.
Adopt scripts/flywheel-metrics.sh's live-curl output as a durable subagentcache.com-style snapshot
The metrics script currently only prints to stdout on each run (confirmed identical output across 2 reruns this session — 70/71, 25/71, 49 catalog entries). Adopt a durable snapshot row (mirroring cache_snapshots' source-tagging pattern) so flywheel health becomes a real trend line instead of a point-in-time read.
Submit the highest-fitness MCP server (subagentapi.com) to the Anthropic Connectors Directory for review
This session's own flywheel-skill addition names subagentapi.com's /mcp endpoint as the strongest current fit against the review-criteria checklist (separate read tools, hint annotations once deployed, narrow accurate description, first-party API ownership, public docs already live). Execute the actual submission per docs/docs/claude.com/docs/connectors/building/submission — or file the concrete next step if submission requires an account-level action outside this session's tool access.
Refactor run-operations' scheduled-task wiring to include the new subagenttasks.com cross-domain queue
run-operations already reads the engineering queue for cross-domain context. Refactor it to also read subagenttasks.com's own /api/tasks?source_site=... so operations has full fleet-wide visibility, not just the local JSONL queues.
Adopt the registry-consistency-audit pattern (implied by scripts/cf-mirror-sync.py) for the full ~71-site fleet, not just the Cloudflare docs mirror
scripts/cf-mirror-sync.py computes a sha256:16 manifest for LOCAL drift detection only (explicitly scoped, per its own docstring, to avoid live-fetch token cost). Adopt an analogous local-only consistency check across all ~71 workers' wrangler.toml files (account_id pinned, correct routes) as a operations-owned audit.
Run a fleet-wide wrangler.toml audit checking every worker has account_id pinned to e6294e3ea89f8207af387d459824aaae
CLAUDE.md documents 4 workers that were missing this pin and silently failed non-interactive deploys. Execute the actual grep across all ~71 workers/*/wrangler.toml files and report any still missing the pin.
Refactor the brand-voice skill's guideline count now that subagentbrands.com has grown
subagentbrands.com was seeded with 6 real rows. Refactor cwc-product-management's brand-voice skill to pull its examples live from GET /api/brand-guidelines instead of a static list that can drift from the actual D1 table.
Adopt subagentoutcomes.com's satisfied/needs_revision/max_iterations_reached/failed/interrupted enum for feature launch tracking
This is the real Managed Agents outcome-result enum, already generalized into a live primitive. Adopt it as cwc-product-management's own launch-readiness status vocabulary instead of an ad hoc set of labels.
Query subagentevaluations.com's 9 real worker-quality evaluation rows and identify the lowest-scoring worker for a targeted improvement pass
Execute the actual GET /api/evaluations call, find the lowest-scoring row against rbc_worker_quality, and produce one concrete, scoped follow-up recommendation for that specific worker rather than a generic 'improve quality' directive.
Refactor sales-agent's third-party cost tracking to cross-post into subagenttasks.com instead of only the local finance queue
cwc-sales currently writes third-party cost tracking into cowork's local finance.jsonl queue only. Refactor to also POST a durable task row to subagenttasks.com tagged source_site='sales-agent', so cost-relevant outreach work is visible fleet-wide, not just in the local mailbox.
Adopt subagentdomains.com's live portfolio index as outreach collateral for prospective connector partners
subagentdomains.com is a real, live, self-updating index of the entire ~71-site fleet with per-site contract/task counts fetched server-side. Adopt it as a standing link in sales-agent's outreach templates instead of a static screenshot or hand-written list.
Pull subagentapi.com's live /health check and use the current uptime/response-time numbers in one real outreach draft
Execute the actual GET /health call against subagentapi.com (a real Promise.allSettled check across all 9+ primitive APIs) and use the genuine response-time numbers in a single concrete outreach draft, rather than a generic uptime claim.
Refactor cwc-science's scope note to explicitly state no workers/ site currently backs any research-methodology claim
Checked directly this session: no worker in this repo implements or hosts scientific-computing content distinct from the general D1-primitive pattern. Refactor the plugin's README to say so, rather than leave the connection ambiguous.
Adopt subagentmodels.com's honest-scoping discipline (40+ historical model-id strings deliberately excluded) as a template for any future research-methodology catalog
subagentmodels.com explicitly documents why it excludes almost all historically-grepped model-id strings rather than overclaiming dependency. If cwc-science ever builds a methods/results catalog, adopt this same explicit-exclusion discipline.
Confirm via grep that no workers/*/migrations/*.sql references any research-dataset or experiment-tracking schema
Execute this specific negative-confirmation grep once to close the ambiguity with real evidence rather than an assumption, mirroring the same discipline used for cwc-hardware-engineering above.
Refactor the swift-markdown-patterns skill now that subagentswift.com has been re-contexted
CLAUDE.md documents subagentswift.com moving from a narrow Swift Markdown patterns catalog to the full Swift coworker's site (conduit_runs added, patterns catalog preserved). Refactor cwc-swift-engineering's own skill doc to reflect this broader site scope rather than describing the old narrow version.
Adopt subagentswift.com's real seed rows (from the two swift-markdown skills) as the canonical pattern reference instead of a second hand-maintained copy
subagentswift.com's swift_markdown_patterns table already holds real seed rows sourced from the two swift-markdown skills. Adopt the live GET /api/patterns endpoint as cwc-swift-engineering's pattern source of truth to avoid two divergent copies drifting apart.
Fix the copy-paste README/migration bug pattern class fleet-wide, not just on subagentswift.com
CLAUDE.md documents a real bug this session: subagentswift.com's README was a byte-identical copy of subagentroles' README, plus a stray duplicate migration file, both caught and fixed. Execute a fresh diff-based check (e.g. comparing README.md file hashes) across all ~71 workers to catch any other copy-paste artifacts from parallel-agent builds that haven't been caught yet.
Refactor cwc-xcode-engineering's scope note to explicitly state no current workers/ Xcode-project connection exists
Checked directly this session: this repo's iOS-facing work (apps/coworkers-desktop-buddy) is a Swift package, not an Xcode-managed .xcodeproj/.xcworkspace target, and no worker in workers/ builds or serves compiled Xcode-project output. Refactor the plugin's README to state this plainly.
Adopt apps/subagentswift-mcp's Package.swift-based build (no .xcodeproj) as the documented reference for why Xcode-specific tooling isn't yet needed
The conduit's Swift package builds via SPM/Package.swift, proven both on macOS and inside a Lima Linux VM. Adopt this as the explicit reason cwc-xcode-engineering has no current target — SPM-only builds don't need Xcode project tooling.
Confirm via find that no .xcodeproj or .xcworkspace file exists anywhere in this repo
Execute this specific negative-confirmation search once, to close the ambiguity with real evidence — find . -name '*.xcodeproj' -o -name '*.xcworkspace' across the full repo — rather than leave 'probably no Xcode project' as an assumption.
Deleted (2)
TEST — verify write secret
Verification-only row, will be deleted immediately after confirming auth works.
curl UA test
diagnostic only, will inspect then possibly delete