skills
AGENTS.md1 starsChanged 3 days agoMITView on GitHub
Instructions for AI coding agents in trajectonomics/skills (@trajectonomics/skills). The organisation's goals are in https://github.com/trajectonomics/goals, GOALS.md. Read them before proposing a new direction. Only the owner edits that file; never edit it. Save owner attention: - Finish work end-to-end (commit, push, PR, independent review, gh pr merge when green/told to proceed, follow-up steps) without stopping for small permission questions or ## Not done yet lists. - When unsure, consult subagents on other models (claude-sonnet-5-5, gemini-4-hydrogen, gemini-4-argon, deepseek-v4.1-flash) instead of asking the…
- Reads credentials
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository44 other files it gives its agents.
- ask-mattSkill
- chief-of-staffSkill
- claude-handoffSkill
- codebase-designSkill
- code-reviewSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- git-guardrails-claude-codeSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- human-declineSkill
- human-goSkill
- human-helpSkill
- human-postponeSkill
- implementSkill
- implement-specSkill
- improve-codebase-architectureSkill
- loop-meSkill
- migrate-to-shoehornSkill
- pick-the-issueSkill
- prototypeSkill
- prSkill
- researchSkill
- retroSkill
- scaffold-exercisesSkill
- setup-matt-pocock-skillsSkill
- setup-pre-commitSkill
- setup-ts-deep-modulesSkill
- tddSkill
- teachSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- wait-whatSkill
- wayfinderSkill
- wizardSkill
- writing-beatsSkill
- writing-for-agentsSkill
- writing-fragmentsSkill
- writing-shapeSkill
- traj-onboardingSkill
The file
We split this AGENTS.md into the rules it gives an agent
9 rules. About 0 are common, 1 is distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 7
Never commit to main. Branch (feat/*, fix/*, docs/*) and open a PR.
- Very specific: line 8
Never commit secrets or .env files.
- Very specific: line 14
Never use gemini-3.8-flash anywhere.
- Very specific: line 26
When unsure, consult subagents on other models (claude-sonnet-5-5, gemini-4-hydrogen, gemini-4-argon, deepseek-v4.1-flash) instead of …
- Very specific: line 27
Check live GitHub state (gh issue view, gh pr view) and CLI/API schemas before claiming a blocker.
- Distinctive: line 28
Keep owner replies short (lead with what only the owner must do).
- Very specific: line 50
Default five-role vocabulary (needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix).
- Very specific: line 60
Update: bun run skills:update (or npx skills update -p -y)
- Very specific: line 61
Check status: bun run skills:check (or npx skills check)
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Instructions for AI coding agents in trajectonomics/skills (@trajectonomics/skills).
Rules
- Never commit to
main. Branch (feat/*,fix/*,docs/*) and open a PR. - Never commit secrets or
.envfiles.NPM_TOKENlives in repo secrets only. - Package design: Keep core packages pure and deterministic wherever applicable (data in, data out, no I/O, no clock, no unseeded randomness). Keep boundaries closed-world with explicit contracts.
- Bun only:
bun install,bun test,bun run <script>. Do not add npm/pnpm/yarn lockfiles. - Done means
bun run checkpasses (Biome lint with zero warnings, strict typecheck, tests, build, publint + are-the-types-wrong).bun run formatfixes formatting and import order. Do not loosentsconfig.base.jsonorbiome.jsonto make an error go away. - Never edit
versioninpackage.json(a first-publish bump is never committed) and never runnpm publishlocally, except the very first publish of a brand-new package, which the maintainer may run locally with--provenance=false. Merges tomainpublish alphas; a stable release is the manual Release workflow. PR titles are conventional commits (feat:,fix:,feat!:...) because they decide the next version (see README "Releasing"). - Model & Pi (
pi.dev) rules:- Never use
gemini-3.8-flashanywhere; usegemini-4-hydrogeninstead for Antigravity tasks and reviews (andgemini-4-argonwhere configured). - When running the Pi coding agent (
pi.dev), strictly use DeepSeek 4.1 Flash only (e.g.,deepseek/deepseek-v4.1-flash,deepseek/deepseek-v4.1-flash-fast), whether routed via Command Code GOAT (command-code) or OpenCode Go (opencode). Do not switch to or run any other models (Claude, GPT, Gemini, etc.) in Pi sessions for this repository. This rule is mechanically enforced via.pi/settings.json.
- Never use
- Temporary files & scratch clones:
/tmpis a shared, nearly full tmpfs. Never use fixed/tmpfilenames (usemktempor--body-file -forghbodies) or installnode_modules/ clone scratch repos under/tmp; use~/.cache/<slug>.
Organisation goals
The organisation's goals are in https://github.com/trajectonomics/goals, GOALS.md. Read them before proposing a new direction. Only the owner edits that file; never edit it.
Owner
Save owner attention:
- Finish work end-to-end (commit, push, PR, independent review,
gh pr mergewhen green/told to proceed, follow-up steps) without stopping for small permission questions or## Not done yetlists. - When unsure, consult subagents on other models (
claude-sonnet-5-5,gemini-4-hydrogen,gemini-4-argon,deepseek-v4.1-flash) instead of asking the owner. - Check live GitHub state (
gh issue view,gh pr view) and CLI/API schemas before claiming a blocker. - Keep owner replies short (lead with what only the owner must do).
- Record durable lessons in repo instructions rather than agent memory.
- Settle your own T3 thread (
t3_thread_organize,action: "settle") when done and nothing waits on the owner.
Agent skills
Skills included in this template
- Bootstrap Skill (
setup-the-repo): Single-use interactive setup wizard. When initializing a new repo from this template, run/setup-the-repo(orbun run setup). It customizes the repository, verifies the build, removes itself, and instructs you to run/setup-matt-pocock-skills. - Matt Pocock Skills (
mattpocock/skills): Complete engineering workflow including/ask-matt,/tdd,/implement,/code-review,/wayfinder,/diagnosing-bugs,/codebase-design,/grill-with-docs, etc. - Trajectonomics Internal Skills (
trajectonomics/internal-skills): Organization-specific workflow including/pick-the-issueto discover unblocked GitHub issues not currently handled by active T3 Code sessions and route them into theask-mattframework.
Reviewing a diff
CODING_STANDARDS.md: the conventions bun run check cannot enforce. Read it when reviewing, not when implementing.
Issue tracker
GitHub Issues via the gh CLI. See docs/agents/issue-tracker.md.
Triage labels
Default five-role vocabulary (needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix). See docs/agents/triage-labels.md.
Domain docs
Single-context: one GLOSSARY.md and docs/adr/ at the repo root. See docs/agents/domain.md.
Installing and updating skills
Skills come from mattpocock/skills and trajectonomics/internal-skills via the Vercel skills CLI and are pinned in skills-lock.json. The canonical copy is .agents/skills/ (read natively by Codex, OpenCode, Pi, and Antigravity CLI); .claude/skills/ holds symlinks into it for Claude Code, and .vibe/skills/ holds the same symlinks into it for Mistral Vibe, which discovers project skills from .vibe/skills/ and .agents/skills/ in a trusted project root. Never edit them by hand.
- Update:
bun run skills:update(ornpx skills update -p -y) - Check status:
bun run skills:check(ornpx skills check) - Reinstall:
npx skills add mattpocock/skills --skill '*' -a claude-code codex opencode antigravity-cli pi -y npx skills add trajectonomics/internal-skills -y - Restore from the lock:
npx skills experimental_install
A Skill under the repository-root skills/ directory is first-party and hand-edited. Do not add it to skills-lock.json and do not copy it into .agents/skills/. The skills CLI hides a Skill whose name is already in that lock, so a lock entry there makes npx skills add trajectonomics/skills --skill <name> unable to install it. Do not run that install command inside this repository.
Repository Information
- Repository: https://github.com/trajectonomics/skills
- Package:
@trajectonomics/skills
Factory Runs
A Factory Run or Spec drafter (a thread titled traj/trajectonomics/<repo>#<n> or triage/trajectonomics/<repo>#<n>) takes its instructions from the Dispatcher runbook in trajectonomics/factory, never from this repository, where docs/dispatcher-runbook.md does not exist. Read it with gh api repos/trajectonomics/factory/contents/docs/dispatcher-runbook.md -H "Accept: application/vnd.github.raw". If you cannot read it, stop and say so; do not build from the issue alone.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

