stigmergy: Skill for Claude Code

Validate a Stigmergy deployment

Use this procedure after a deployment or a release spanning more than one subsystem. Read deployment coordinates from the gitignored .env and Fly secrets; never print secret values.

1. Preflight

• Confirm every Fly process group is running the intended image and passes its health checks.
• Confirm Postgres, the private evidence bucket, the GitHub App, Slack credentials, and embedding credentials are configured without displaying them.
• Confirm the deployed knowledge-repository commit contains the target control files and pins the released platform commit in both workflows.
• Inspect recent logs for restart loops, tracebacks, secret-bearing messages, and repeated jobs.

For a clean-cut test release, run the guarded reset before deployment validation. It must leave a fresh target schema, an empty queue/index/object namespace, and the empty target knowledge scaffold.

2. Read path through MCP

After the first full rebuild, exercise all read tools with an authenticated identity:

• search_brain in both corpus languages and inspect lexical/vector ranking arms;
• read_page on a visible result and on an unknown path;
• list_entities and describe_entity when an entity exists;
• ask for one supported and one unsupported answer, checking citations and honest refusal.

Repeat a restricted search/read with identities on both sides of an ACL boundary. Unknown, hidden, and unauthorized pages and entity IDs must have indistinguishable external responses.

3. Unified capture and writer

Submit through the official local bridge:

• exact text;
• a local digital PDF;
• a local scanned PDF;
• a private Drive PDF using local Google OAuth when credentials are available.

Observe brain_submissions reach landed. Each capture must retain exact original bytes, create one neutral sources/YYYY/MM/<capture-id>.md path, and land one Git commit with one Changes record. A retry with the same idempotency identity must not create another commit.