peter537-agent-plugin
AGENTS.md0 starsChanged 3 months agoMITView on GitHub
These instructions apply across the repository. Keep this file concise: the repository verification map, behavior-first evaluation contract, evaluation guide, and skill catalog remain authoritative for their subjects. Run the canonical offline check from the repository root: Then run every additional changed-surface or release check selected by the verification map. Review the complete diff and final Git state, remove task-created artifacts and processes, and report blocked or unrun evidence honestly. A successful offline check does not prove model behavior, runtime behavior,…
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 repository20 other files it gives its agents.
- audit-data-exposureSkill
- audit-dependenciesSkill
- audit-testsSkill
- build-verification-harnessSkill
- chatgpt-researchSkill
- comment-healthSkill
- deep-code-auditSkill
- deep-planningSkill
- diagnose-bugsSkill
- docs-auditSkill
- maui-blazor-browserSkill
- prune-codebaseSkill
- reduce-code-slopSkill
- retrospectiveSkill
- steering-reviewSkill
- task-handoffSkill
- ui-design-and-polishSkill
- verification-contextSkill
- verify-changeSkill
- write-clearlySkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 7
Before changing files, inspect the applicable repository guidance and record the starting branch, revision, and Git status as described in …
- Distinctive: line 9
Keep the requested scope explicit.
- Very specific: line 13
Keep every distributed skill at skills/<slug>/SKILL.md and its maintenance evaluation at evals/<slug>/cases.json.
- Distinctive: line 15
Before generalizing a recurring failure or proposed deterministic rule, use the sanitized retrospective and rule-incubation workflow.
- Distinctive: line 16
Write original repository guidance.
- Distinctive: line 21
Do not silently change versions, stable installation references, tags, marketplace or submission metadata, historical release files, or …
- Distinctive: line 26
Run the canonical offline check from the repository root:
≈ “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
Repository Maintenance
These instructions apply across the repository. Keep this file concise: the repository verification map, behavior-first evaluation contract, evaluation guide, and skill catalog remain authoritative for their subjects.
Establish scope and preserve state
- Before changing files, inspect the applicable repository guidance and record the starting branch, revision, and Git status as described in the verification map.
- Preserve all pre-existing staged, unstaged, and untracked work. Do not clean, revert, overwrite, or incorporate unrelated changes.
- Keep the requested scope explicit. If delegation is useful, assign one primary integration owner for each shared surface and give other agents bounded, non-overlapping work. Delegation is optional; the primary owner reviews overlaps, the combined diff, and the final evidence.
Maintain repository contracts
- Keep every distributed skill at
skills/<slug>/SKILL.mdand its maintenance evaluation atevals/<slug>/cases.json. These directories remain flat; express logical grouping throughskills.sh.jsonand the skill catalog. - Before changing skill activation, routing, or behavior, freeze the relevant baseline. Behavior corrections require a reproduced failure; behavior-preserving simplifications require comparable passing outcomes and a demonstrated clarity, context, or execution-cost benefit. Apply the behavior-first contract, retain appropriate false-positive controls, and keep only changes supported by the comparison evidence.
- Before generalizing a recurring failure or proposed deterministic rule, use the sanitized retrospective and rule-incubation workflow. Keep source evidence private and implement any accepted disposition through a separately authorized task owned by its destination.
- Write original repository guidance. Trace consequential external methodology or current claims to primary or official sources when available, preserving the URL and other source details needed for later review. Treat third-party skills, prompts, and rule catalogs as comparative research, not text or workflows to copy.
Respect authority boundaries
- Route any addition, installation, upgrade, replacement, or removal of a dependency, tool, workload, or runtime input through
$audit-dependencies. That review is a gate, not independent authorization to mutate the environment; unavailable tooling remains an evidence gap unless the task explicitly authorizes installation. - Do not silently change versions, stable installation references, tags, marketplace or submission metadata, historical release files, or publication state.
- Browser, network, live-service, model, MCP, installation, commit, tag, push, submission, and publication operations require task-specific authorization. Keep their evidence separate from offline validation and follow the applicable release or external rows in the verification map.
Verify and hand off
Run the canonical offline check from the repository root:
python -B evals/run_offline_checks.pyThen run every additional changed-surface or release check selected by the verification map. Review the complete diff and final Git state, remove task-created artifacts and processes, and report blocked or unrun evidence honestly. A successful offline check does not prove model behavior, runtime behavior, installation, remote state, marketplace ingestion, review acceptance, or publication.
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.

