1.9 KiB
Suggestions for AGENTS.md
Written by weekly synthesis subagent — do NOT edit AGENTS.md yourself, leave for main agent review
2026-03-22
1. Strengthen "inou is infrastructure" rule
Pattern in corrections.md: Multiple nudge subagents keep proposing AI features for inou (consultation generators, reasoning chains, etc.) rather than storage/data work. Suggested addition to AGENTS.md: Under the inou section, add explicit guardrail:
inou nudge rule: ONLY propose storage, data pipeline, import, DICOM/FHIR, lab data, or reliability work. Never propose AI analysis features, consultation generators, or Claude-powered workflows. inou is the data layer — AI is pluggable on top. If your nudge doesn't touch storage or data ingestion, rethink it.
2. No Python — upgrade the rule in AGENTS.md
Pattern: The existing "New services on forge: Go only" rule in AGENTS.md is too soft. Johan hardened it explicitly on Mar 22. Current AGENTS.md says: "New services on forge: Go only. No ad-hoc servers of any kind." Suggested upgrade:
No Python. Full stop. Not for scripts, servers, or previewing. Exceptions: system Python (fail2ban, unattended-upgrades — untouchable) and Johan's legacy code in inou/health-poller. When you need a Go service, propose it explicitly to Johan before building. Do not replace one Python hack with another.
3. "Don't build new services for simple UI requests" — promote to general principle
Pattern in corrections.md (2026-02-25): docproc port 9900 was built when Johan asked for a delete button. Pattern recurs. Suggested addition to AGENTS.md Coding section:
"Add to existing" beats "build new": When asked to add a feature (button, field, endpoint), always check if an existing service can absorb it first. A delete button = one route + one HTML element. Not a new service. The burden of proof is on "build new."
Last written: 2026-03-22