feat: webster-weekly-council library skill (route operators to /webster-weekly-council)

[richsak]

Summary

• Convert the 662-line prompts/second-wbs-session.md runbook into a Claude Code library skill at skills/webster-weekly-council/SKILL.md so operators can run the full Webster weekly council via /webster-weekly-council instead of pasting the prompt
• Library shape: slim SKILL.md index + 9 on-demand phase references in references/ + 2 reusable helper scripts in scripts/ (the per-session polling loop, used 7×, and the planner-JSON extractor)
• The production runbook is unchanged — prompts/second-wbs-session.md stays as the locked single-page fallback. Locked by scripts/__tests__/sim-council.test.ts. Both surfaces produce identical artifacts.
• Doc surfaces converted: README, AGENTS, context/ARCHITECTURE.md, context/FEATURES.md, history/AGENTS.md now route operators to the new slash-command path with the prompt listed as fallback

Skill structure

skills/webster-weekly-council/
├── SKILL.md                       66-line index: phase table + 8 invariants + override-default-loop directive
├── references/
│   ├── preflight.md               keychain + GH auth + registration artifacts
│   ├── seed-history.md            phase 1 — 10-week mock analytics seeder
│   ├── branch.md                  phase 2 — shared council/<date> branch
│   ├── planner.md                 phase 3 — context marshal + planner invoke (fail-closed)
│   ├── fan-out.md                 phase 4 — 6 parallel managed-agent sessions
│   ├── verify.md                  phase 5 — findings count gate + bun validate-findings
│   ├── genealogy.md               phase 5.5 — fail-open gap-detection
│   ├── redesign.md                phase 6 — synthesis session
│   ├── publish.md                 phases 7+8 — draft PR + checkpoint
│   └── failure-modes.md           recovery guide
└── scripts/
    ├── run-agent-session.sh       create session → send user.message → poll until idle/timeout (used 7×)
    └── extract-planner-json.py    pull JSON block out of plan.md for new_critic_request handling

Test plan

• bun run validate passes locally (type + lint + format + agent schema + findings + markdownlint + 181/181 tests)
• scripts/__tests__/sim-council.test.ts passes — confirms prompts/second-wbs-session.md git diff is empty (production immutability)
• bash -n on run-agent-session.sh parses cleanly
• extract-planner-json.py smoke-tested — extracts JSON from a fenced ```json block; jq sees new_critic_request
• Live API end-to-end run not yet executed. Skill is structurally equivalent to the prompt by review; full council run against the real Anthropic endpoint is the remaining validation step (~$0.20 + 30–50 min). Operator's call whether to gate merge on that.

Out of scope

• Did not touch the production prompt (locked) or any other prompt
• Did not modify any of the 7 production agent JSONs
• Did not exercise the live API — this PR ships the operator surface, not a re-validation of the council run itself

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

🗂️ Base branches to auto review (2)

• main
• review/empty-base

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: ac2cb3c5-4157-4ea8-af78-82b1c85931e9

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/weekly-council-skill

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}