pakem
AGENTS.md0 starsChanged 8 days agoMITView on GitHub
Rules for agents editing this repository. The kernel itself is kernel/KERNEL.md; it is installed into each host by tools/pakem install, not loaded from here. - This repository holds the universal kernel, skills, standards, templates, and tooling. It must stay free of organization- and client-specific content and of the maintainer's own preferences: those belong in profile repositories (ADR 0001, ADR 0012). - Project business rules never belong here. - Tests: python3 -m unittest discover -s tests -v - Lint: tools/pakem lint…
In this repository14 other files it gives its agents.
- backend-engineeringSkill
- change-planningSkill
- change-reviewSkill
- device-healthSkill
- genai-engineeringSkill
- instruction-maintenanceSkill
- pakemSkill
- pakem-updateSkill
- project-bootstrapSkill
- release-deliverySkill
- root-cause-debuggingSkill
- sensitive-operationsSkill
- state-and-handoffSkill
- ui-ux-qualitySkill
The file
Files
Pakem
Rules for agents editing this repository. The kernel itself is kernel/KERNEL.md;
it is installed into each host by tools/pakem install, not loaded from here.
Purpose and boundaries
- This repository holds the universal kernel, skills, standards, templates, and tooling. It must stay free of organization- and client-specific content and of the maintainer's own preferences: those belong in profile repositories (ADR 0001, ADR 0012).
- Project business rules never belong here.
Commands
- Tests:
python3 -m unittest discover -s tests -v - Lint:
tools/pakem lint(must report 0 errors before any commit) - Install preview:
tools/pakem install(dry run;--applyonly with owner approval) - CI:
.github/workflows/ci.ymlruns lint, withprofiles/example, and the unit tests on Python 3.11 and 3.14 for every pull request andmain, on GitHub-hosted runners; sandbox tests stay opt-in. - Python 3.11+ standard library only for
tools/,enforcement/, skill scripts. Hooks are POSIX sh and must work with macOS/bin/sh.
Rules for changing the framework
- Kernel budget: at most 9,500 characters and 150 lines. Every line must be universal, not discoverable from code, not mechanically enforceable, and tied to a real or plausible failure. List rules removed or considered when adding one.
- Skills follow the Agent Skills format: trigger-first description of at most 350 characters, body under 500 lines, references one level deep, each reference starting with a status header.
- Instruction text drafted by an agent needs owner review before it is merged.
- The framework is stable (ADR 0007, amended 2026-10-22). Change rules only for a
problem a real project hit, through
pakem propose: each accepted proposal is its own PR. Thepakemtool releases whenever an area is proven; kernel and skill text waits for the next release, batched at most once a month, and never triggers one alone (ADR 0011). Build no newpakemmechanism unless a project needs it. - ADRs only for decisions with real trade-offs. Record each change as a fragment
changes/<type>-<kebab-topic>.md(changes/README.md); never editCHANGELOG.mdon a branch. A release runstools/pakem changelog --release X.Y.Z --apply, which writesCHANGELOG.mdandVERSION. MAJOR only when a floor changes or an existing install or project would break; MINOR for new or changed rules; PATCH for fixes. - Compliance evals are optional: run them by hand after a large host or model
change, not as a release gate (
evals/README.md).
Overrides
default:openspec: framework changes use ADRs, CHANGELOG, and semver instead of OpenSpec (ADR 0005).
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.

