This is agent-sync, a multi-agent coordination platform at https://sync.parc.land/.

The README.md is served as the root response (GET /) and doubles as the remote SKILL.md for Claude's skill system. It is fetched at module load time and served as text/plain.

This means README.md must:

• Follow SKILL.md formatting conventions (frontmatter with name and description, concise workflow-first structure)
• Be self-contained enough for an LLM to use the API without reading reference docs
• Stay under ~4K tokens so it fits comfortably in a context window alongside other skills
• Link to reference/api.md, reference/cel.md, reference/examples.md for deeper detail

When editing README.md, you are editing the skill definition that other Claude instances will read to learn how to use this platform. Write for that audience.

Two operations: read context, invoke actions.

• GET /context — returns everything (state, views, agents, actions with defs, messages with bodies)
• POST /actions/:id/invoke — the only write endpoint (builtin and custom actions)
• GET /wait — blocks until condition, returns full context

10 total endpoints. Every write flows through action invocation. Every invocation is logged to _audit scope.

There are NO direct write endpoints for state, actions, views, or messages. All writes go through built-in actions (_set_state, _register_action, etc.) or custom actions.

main.ts          — HTTP router, all endpoint handlers, built-in actions, audit logging
auth.ts          — Token generation, hashing, scope authority checks
cel.ts           — CEL context builder, expression evaluation, view context
schema.ts        — SQLite schema and v4→v5 migration
timers.ts        — Wall-clock and logical-clock timer lifecycle
dashboard.ts     — Browser dashboard (auth-gated, single HTML response)
README.md        — Skill definition (served at root)
reference/       — Detailed docs (api.md, cel.md, examples.md)

Agents, State, Messages, Actions, Views, Audit, CEL Console. The Audit tab shows every action invocation with success/failure status.