agentic-harness-development
AGENTS.md0 starsChanged 42 days agoMITView on GitHub
This repository distributes portable skills, reusable agent roles, and TypeScript tools. Read docs/implementation-plan.md for the current milestone and docs/architecture.md before changing component boundaries. Run $poietic-harness-start once at session start; its source is skills/poietic-harness-start/SKILL.md. For substantive repository changes, the parent must read the harness review workflow and run its four fresh-context specialists; reviewers return findings without recursively delegating. Use $jev by default for suitable bounded structured decisions; read its skill and pass its entrypoint and authorized execution access to subagents. Run…
- 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 31,774 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 repository4 other files it gives its agents.
- jevSkill
- poietic-harness-startSkill
- poietic-harness-updateSkill
- ui-standardsSkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
5
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
Run $poietic-harness-start once at session start.
- Distinctive: line 11
Use $jev by default for suitable bounded structured decisions.
- Very specific: line 24
Keep executable components in tools/<id>/, with owned source, tests, build configuration, documentation and expertise.
- Very specific: line 26
Keep entrypoints and indexes small.
- Very specific: line 32
Never commit credentials or copy local user configuration into releases.
- Distinctive: line 33
Do not claim native compatibility, a live API check, or a published release based solely on local validation.
- Very specific: line 34
Every significant addition, fix, removal, or behavior change must update CHANGELOG.md in the same pull request, including changes to …
≈ “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
Working in this repository
This repository distributes portable skills, reusable agent roles, and TypeScript tools. Read docs/implementation-plan.md for the current milestone and docs/architecture.md before changing component boundaries.
Session start
Run $poietic-harness-start once at session start; its source is skills/poietic-harness-start/SKILL.md.
For substantive repository changes, the parent must read the harness review workflow and run its four fresh-context specialists; reviewers return findings without recursively delegating.
Use $jev by default for suitable bounded structured decisions; read its skill and pass its entrypoint and authorized execution access to subagents.
Commands
npm ciinstalls the locked toolchain; use Node.js 24 LTS or 26.npm run release:stampprepares a new UTC date/time version; do not choose major/minor/patch increments. Seedocs/releases.md.npm run checkformats-checks, typechecks, measures test coverage, and validates catalog resources.npm run release:packvalidates and assembles release assets.npm run release:verifyinstalls the exact tarball in a temporary consumer and checks library and CLI behavior.
Constraints
- Every project's primary README must show a linked maturity badge directly below its title, using the shared Tinker → Alpha → Beta → Stable scale. Select the stage from evidence, explain it in the status section, and log changes to the assessment. Apply this when bootstrapping or maintaining a project within the user's requested scope.
- Keep executable components in
tools/<id>/, with owned source, tests, build configuration, documentation and expertise. Use TypeScript for shared runtime code; keep CLI and adapters thin over the library. Generic harness source must not import tool internals. skills/<id>/SKILL.mdandagents/<id>/AGENT.mdare canonical entrypoints. Embedded tool manifests aretools/<id>/package.json. Each component must ownknowledge/INDEX.mdandmemory/INDEX.md, with local expertise maintained under its own directories. Followdocs/local-expertise.md.- Keep entrypoints and indexes small. Search the selected component with
rg, then read only relevant topics/lessons and supporting evidence. Build useful local explanations instead of relying on remote links or previous chat context alone. - Maintain provenance, verification dates, applicability, and index links when knowledge changes. Capture actual lessons in scoped local memory; keep
knowledge/local/andmemory/local/out of commits and releases. Do not modify unrelated host memory stores. - Native models, permissions, and discovery settings belong in harness adapters. Learned state uses a writable component working directory separate from immutable release caches.
- This repository owns definitions and tools. Durable project state, event queues, leases, and deployed identities belong in a separate runtime service.
- The bootstrap CLI is read-only. Do not silently turn a selection plan into installation, a paid API call, or a background service.
- Explicit project selections override detection when detection is implemented. Installing a capability does not grant new external-action permissions.
- Never commit credentials or copy local user configuration into releases. Read a component's
.env.exampleonly for variable names; it is not a credential source. - Do not claim native compatibility, a live API check, or a published release based solely on local validation.
- Every significant addition, fix, removal, or behavior change must update
CHANGELOG.mdin the same pull request, including changes to agent/skill behavior. Follow the changelog policy: readable impact, searchable component names, migration actions, and clear unreleased/released status.
Completion
Run the checks appropriate to the change. Update the implementation plan when behavior changes and apply the changelog rule above. Explain in the pull request when a minor change does not need a changelog entry. Update measured coverage after changing runtime tests. For release changes, verify an external consumer of the exact artifact and confirm GitHub assets and status. Keep documentation clear about implemented versus planned behavior.
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.

