Skip to content

AGENTS.md

Latest commit

 

History

History
84 lines (63 loc) · 3.68 KB

AGENTS.md

File metadata and controls

84 lines (63 loc) · 3.68 KB
Raw

Slow-powers

Slow-powers is a set of software development methodology skills and meta skill-writing guidance.

What lives here

This repo ships Slow-powers across four harnesses:

  • skills/ — Skills, assets, and cross-cutting tests
  • .claude-plugin/ — Claude Code plugin
  • .codex-plugin/ — OpenAI Codex plugin
  • opencode/ — OpenCode plugin (@slowdini/slow-powers-opencode)
  • cline/ — Cline plugin (CLI/SDK/Kanban only; declared via the cline field in package.json, with skills/ auto-discovered from the package root)

Cline-specific setup for working on this repo also lives at root:

  • .clinerules/ — Cline rules (the Memory Bank custom instructions), tracked
  • memory-bank/ — Cline Memory Bank contents. Gitignored: each developer keeps their own local bank; only the rule in .clinerules/ is shared

Editing the right files

Two file-confusion traps are common in this repo. Avoid both:

  • Write memory-file changes to AGENTS.md, the only real file. CLAUDE.md (and any future harness-specific memory files) are symlinks to AGENTS.md. Edit AGENTS.md directly; the symlinks reflect it automatically. Never try to "fix up" the other names — there's nothing to fix.
  • Read and edit only files inside this repository, never the installed slow-powers plugin. Your environment almost always has the slow-powers plugin installed while this repo is the bleeding-edge dev source. The two trees diverge, so an installed skill file is not the file you want. Every read and edit belongs to a path under this project directory.

Cross-Harness Compatibility

All skills MUST use cross-harness vocabulary, as described in slow-powers/writing-skills.

Pull Request Requirements

  • Read existing skills before proposing changes to skill content. Skill prose has been tuned over many iterations upstream and downstream; changes to behavior-shaping content (Red Flags tables, rationalization lists, "human partner" language) need evidence the change is an improvement.
  • Test your change on at least one harness and note which one in the PR description. If your change touches harness-specific infrastructure, test that harness.

Skill Changes

If you modify skill content:

  • Use slow-powers:writing-skills to develop and test changes.
  • Run adversarial pressure testing across multiple sessions, not just the happy path.
  • For behavior-shaping changes, show before/after eval results in the PR description. For deterministic changes (instruction-following the agent reliably does anyway), state the decision and reasoning to skip the eval instead — see "Choosing to test with evals" in slow-powers:evaluating-skills.
  • Ensure skills are cross-harness compatible: avoid harness-specific tool or feature names.
  • Our discipline-enforcing skills should carry at least one seeded eval case — one that embeds a short prior transcript so the skill is met mid-session under a competing attractor — because their real-world failures happen in-flight and a cold prompt alone under-measures them (see "Seeding conversation context" in slow-powers:evaluating-skills). hardening-plans is the reference example.

Local development

bun install   # also activates git hooks via the `prepare` script
bun test
bun run check

bun install runs the prepare script, which installs the git hooks (pre-commit runs typecheck + lint-staged; pre-push runs the test suite).

bun scripts/bump-version.ts <version> updates every manifest in lockstep.

To test the Cline integration live: cline plugin install ./ --cwd <scratch-dir>, run a Cline session in the scratch dir, then cline plugin uninstall slow-powers --cwd <scratch-dir> when done.