Skip to content

AGENTS.md

Latest commit

 

History

History
51 lines (40 loc) · 2.51 KB

AGENTS.md

File metadata and controls

51 lines (40 loc) · 2.51 KB
Raw

PyAutoScientist — Agent Guidance

This file is for AI coding agents (Claude Code, Codex, Cursor, etc.) and humans discovering this repository. It is the agent-agnostic source of truth; Claude Code loads it via the @AGENTS.md import in CLAUDE.md.

What this repo is

PyAutoScientist is the umbrella landing repo for the PyAuto organism — a human-led AI software-development system. It presents the organism (what it is, how to fork it) through the docs and links in README.md; it is not a framework you install, and it holds no organism code of its own. The organs that do the work are peer repositories — see the body map in PyAutoMind/repos.yaml and the canonical boundaries in PyAutoBrain/ORGANISM.md.

Editing rules

  • The organ table in README.md is generated from PyAutoMind/repos.yaml (between the repos_sync:organs markers). Do not hand-edit it — change repos.yaml, then run python3 PyAutoMind/scripts/repos_sync.py --write.
  • The RTD documentation source does not live here yet. The published docs at https://pyautoscientist.readthedocs.io are built from PyAutoBrain/docs/; edit the docs there.

Future intent (not now — Phase 3, demand-gated)

Migrating the RTD docs source into this repo — so PyAutoScientist becomes the real documentation centre rather than a landing page pointing at Brain's docs/ — is a recorded future step. It is deliberately deferred until there is demand; do not start it as part of routine work.

Never rewrite history

Never rewrite pushed history on any repo with a remote — no git init over a tracked repo, no force-push to main, no fresh-start "Initial commit", no filter-repo / filter-branch / rebase -i on pushed branches. To get a clean tree: git fetch origin && git reset --hard origin/main && git clean -fd.

Sessions end at their deliverable

A session ends when it reports its deliverable — never arm anything that outlives the turn to wait for CI, a review or a merge: no send_later, no subscribe_pr_activity, no CronCreate, no ScheduleWakeup, no /loop, no RemoteTrigger create/update/run. Judge once, report, stop; the human re-runs /prm (or the batch review) when it is green. Measured: five batch members armed hourly check-ins on 2026-08-31, and a mobile /prm re-armed a 60-minute send_later hourly all night on 2026-09-03 with no task active, draining usage.