Playbooks

Step-by-step recipes for authoring hooks, tools, adapters, and stacks.

• Playbook — Authoring or Modifying an Adapter
• Playbook — Turning On Agent SupervisionThe operator recipe for routing each semantic role to its own adapter, model, and reasoning effort — including the single-adapter case, capacity cooldowns, and how to back the feature out.
• DB Reset Playbook — Wipe & Rebuild coding-os Data
• Doctor Checks — Reference
• Playbook — Authoring a Hook in src/core/hooks/
• Playbook — Authoring a cos MCP Tool
• Playbook — Multi-Agent Git: Layered Defense & Real-World Use CasesOne place that answers "is the git base safe for many agents on one repo, and which mode do I pick for *my* project?" It captures the **layered-defense model** (why client-side hooks are fast-feedback, not the wall), the **pre-publish GitHub setup**, and a **validated matrix of 10+ real-world project archetypes** with the recommended mode and the breakpoints each one hits.
• Polyglot Extractor Roadmap — Python-Grade Coverage for Every Language
• Playbook — PR-Mode CI Economics (consumer cost guidance + reference workflow)pr-mode `autonomy_level=auto_merge`/`autonomous` + a GitHub merge queue multiplies CI runs (each PR runs CI, then each `merge_group` runs it again), so on a free **private** repo with a slow suite the 2000-min/mo Actions budget burns fast. This playbook gives the cost model, the fast-gate / full-suite split that keeps the autonomous loop affordable, and a copy-paste consumer CI workflow.
• Playbook — PR-Mode Multi-Agent Git Workflow (Consumer-Only)The operational contract for `COS_GIT_WORKFLOW=pr` — per-task git worktree → PR → required-CI → auto-merge → auto-cleanup, so 5+ agents work one consumer repo concurrently without collision and broken code never reaches the integration branch.
• Playbook — Security Review (Per-Change Overlay)
• Playbook — Authoring a Rich Skill in src/core/skills/ and src/templates/<stack>/skills/
• Playbook — Authoring a Stack Template