Everything below is plain Markdown, derived from the actual code — on any conflict, the code and tests win. That's deliberate: these docs are written to be read by your agent, raw and unstyled, exactly as they live in the repo.
https://openpouch.dev/llms.txt — the agent-facing entry point with everything it needs before deploying. Humans: pick a doc below; your browser will show the raw Markdown.INDEX.mdThe machine-readable documentation map (this page, for agents)
llms.txtAgent-facing entry point — the facts an agent needs before deploying
WORKFLOWS.mdAgent workflows: happy-path, self-repair, claim-to-keep, upgrade, resume
CLI.mdCommand by command — flags, --json shapes, exit codes
MCP.mdMCP server, tool by tool (14 tools, no approve) + one-shell smoke recipe
SCHEMAS.mdMachine-readable JSON Schemas for every core result + error shape
HARNESSES.mdHarness-neutrality: CLI / MCP / repo-files integration matrix
ACCOUNTS.mdAccounts, API keys, tiers & quotas, upgrade / billing boundary
INSTANT-LANE.mdThe instant lane: static + dynamic runtime, persistence, routing, ops
DATA-MODEL.mdThe three open file formats (manifest / policy / evidence)
API.mdPublic API of @openpouch/core
ARCHITECTURE.mdMonorepo layout, design principles, adapters, delivery lanes