nix-skills
AGENTS.md9 starsChanged 55 days agoMITView on GitHub
Read CONTRIBUTING.md before changing the collection. - Start with an issue and a task branch (feat/<issue>-<slug> or equivalent). Use Conventional Commits; never commit directly to main. - Tasks tracked as issues or touching multiple files require three artifacts named YYYY-MM-DD-<issue>-<slug>.md: intent/ (why), spec/ (what), then plan/ (how). Each starts with status: draft frontmatter and links its predecessor. Commit each stage and stop for human review before the next. - Never self-approve. Record each human approval in its own commit, changing…
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 32,671 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 repository10 other files it gives its agents.
- devenv-projectSkill
- home-managerSkill
- microvm-nixSkill
- nix-darwinSkill
- nix-languageSkill
- nixos-coding-agentsSkill
- nixos-operationsSkill
- nixos-wikiSkill
- nixpkgs-developmentSkill
- nix-workflowSkill
The file
We split this AGENTS.md into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
3
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
- Distinctive: line 3
Read CONTRIBUTING.md before changing the collection.
- Very specific: line 7
Start with an issue and a task branch (feat/<issue>-<slug> or equivalent).
- Distinctive: line 8
Use Conventional Commits; never commit directly to main.
- Very specific: line 13
Never self-approve. Record each human approval in its own commit, changing only that stage to status: approved.
- Very specific: line 19
Read before editing, check Git status, make small reversible changes, run relevant tests, and report actual evidence and limitations.
- Distinctive: line 46
Do not broaden publication allowlists to make a failing update pass.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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
Working on nix-skills
Read CONTRIBUTING.md before changing the collection.
Workflow
- Start with an issue and a task branch (
feat/<issue>-<slug>or equivalent). Use Conventional Commits; never commit directly to main. - Tasks tracked as issues or touching multiple files require three artifacts
named
YYYY-MM-DD-<issue>-<slug>.md:intent/(why),spec/(what), thenplan/(how). Each starts withstatus: draftfrontmatter and links its predecessor. Commit each stage and stop for human review before the next. - Never self-approve. Record each human approval in its own commit, changing
only that stage to
status: approved. Implement only an approved plan. Cite its current step; update deviations with the implementation commit. - Typos, lock-only bumps and one-line configuration changes are exempt. Recurring generated-only updates follow their already-approved provider contract; changes to that contract require review.
- Read before editing, check Git status, make small reversible changes, run
relevant tests, and report actual evidence and limitations. Open a PR linking
all artifacts with
Closes #<issue>. Do not merge untested or without approval.
Layout and boundaries
skills/<name>/: portable complete packages. Keep instructions focused; link supporting resources and retain source attribution and licenses.skills.json: sorted unique skill names, not commands or updater settings.scripts/check_collection.py: offline generic metadata/resource validation.scripts/check.py,update.py, provider modules,artifact.py: explicit maintained-source validation and tightly scoped automated publication.flake.nix,nix/: data packages, the documentation site and opt-in Home Manager installation.demo/: separate flake (ownflake.lock) for the disposable demo VM and its offline VM test;.github/workflows/demo.ymlruns it and is not a required check.templates/: flake templates:agentic-nixosuses the demo flake'snixosModules.agentic;projectuses thedevenv/module.demo.ymlchecks both.devenv/: devenv module (imports: [ nix-skills/devenv ]) linking skill folders into a project; unrelated to the rootdevenv.nix, which is this repository's own environment.docs/: mdBook site sources;nix build .#docsgenerates the catalog, options and update schedule pages and checks links and Nix style. Never edit generated pages.- Root
devenv.*: contributor tools;tests/devenv/is a separate provider fixture. The environment and distribution locks are updated independently.
Registering a skill must not enroll an updater. Never execute contributed
helpers during metadata validation. PR jobs are read-only, have no secrets,
and cannot publish or install. Do not introduce pull_request_target execution.
Do not broaden publication allowlists to make a failing update pass.
Checks
devenv shell check-fast
nix flake check
nix build .#nix-skills --no-link
nix eval .#packages.aarch64-linux.nix-skills.drvPath --raw
devenv shell check-providerscheck-fast runs collection validation, Python unit tests and actionlint.
Provider regeneration checks require upstream network/cache access; failures
are failures, not skipped tests. See CONTRIBUTING for direct commands.
Stage new Nix files before Git-flake evaluation. Never activate a test home,
rebuild the host, install skills or grant devenv allow as part of validation.
Consumers using Home Manager as a NixOS module rebuild through NixOS, never
home-manager switch.
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.

