coreci-executor: add per-agent onboarding checklist

This commit is contained in:
2026-10-04 16:50:51 +00:00
parent a208eb36d3
commit 9dbe005e21
+50
View File
@@ -0,0 +1,50 @@
# coreci-executor — Agent Onboarding Checklist
> Use this checklist when adding a new agent (`coreci-<role>`) or when
> an existing agent loses its tool binding mid-session.
## Pre-flight (one-time, per agent)
- [ ] Gitea user `coreci-<role>` exists (admin creates via the `coreci` org).
- [ ] KB file set in the agent's credential KB:
- `git.coreci.dev.txt` → username `coreci-<role>`
- `executor.coreci.dev.txt` → `EXECUTOR_API_KEY=<shared key from .env>`
- one `<cap>.txt` per scoped capability (e.g. `proxmox.txt`, `openwebui.txt`)
- [ ] Allowlist entry exists in `allowlist.yaml` with
`auto_run_targets: [coreci-<role>/home]` (PR to `executor-config`,
CEO approval).
- [ ] Home repo `coreci-<role>/home` exists and is initialized
(`README.md`, `.gitkeep`).
## Per-bind (every re-bind, including recovery from tool loss)
- [ ] Fetch `coreci/executor-config/tools/coreci-executor/SPEC.md` and
note version.
- [ ] Instantiate the `coreci-executor` tool locally bound to the SPEC.
- [ ] Handshake: `GET /health` → `200`.
- [ ] Smoke-test pack: 1-line `echo HELLO_FROM_<role>_<utc_ts>` →
receive `request_id` → record in `ONBOARDING.md`.
- [ ] Append row to `ONBOARDING.md`:
| utc_ts | SPEC version | request_id | result |
|---|---|---|---|
## Failure recovery
- Tool missing in palette → re-fetch SPEC, re-instantiate, re-handshake.
- `/submit_request` returns `NEEDS_APPROVAL` for a target that should
be `AUTO_RUN` → check `allowlist.yaml`; entry may be missing or
agent name may not match (see SPEC §4 naming convention).
- All smoke-test failures → escalate to CEO; do not run real work
before `coreci-executor` is verified bound.
## References
- `coreci/executor-config/tools/coreci-executor/SPEC.md` (canonical
tool definition; v0.1 as of 2026-09-26)
- `coreci/company/standards/communications.md` §"Mailbox" (mailbox
convention; per-home outbox + inbox)
- `coreci/executor-config/allowlist.yaml` (AUTO_RUN classification
per agent; v2 naming convention; v4 will add `read_targets` once
the executor parser change lands — see D6 in CAIRO's task brief)