agent-devkit
AGENTS.md4 starsChanged 2 months agoMITView on GitHub
The sections below apply only when developing the agent-devkit source repository. When this file is loaded as a plugin rule in another project, follow that project's own AGENTS.md and use using-devkit to choose the workflow; this file imposes no repository-specific policies on that project. This repository contains dependency-free Markdown skills for keeping coding-agent context and an LLM-facing codebase wiki in sync. - Read this file. Read docs/llm/INDEX.md only in target projects where the skills have created that wiki. - Inspect…
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 repository15 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- writing-skillsSkill
- brainstorm-featureSkill
- context-handoffSkill
- document-wikiSkill
- estimate-featureSkill
- implement-taskSkill
- lean-auditSkill
- plan-featureSkill
- read-codebase-contextSkill
- review-and-verifySkill
- setup-codebaseSkill
- setup-openezSkill
- systematic-debuggingSkill
- using-devkitSkill
The file
We split this AGENTS.md into the rules it gives an agent
12 rules. About 1 is common, 8 are distinctive, and 3 are very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈8
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
- Very specific: line 15
Read this file. Read docs/llm/INDEX.md only in target projects where the skills have created that wiki.
- Distinctive: line 24
Keep changes focused and minimal.
- Distinctive: line 26
Follow the team's one-branch/PR-per-task workflow.
- Very specific: line 33
When resolving conflicts in docs/agent-devkit/INDEX.md or docs/llm/INDEX.md, preserve links from every task and verify all targets after …
- Distinctive: line 36
Treat generated LLM documentation as a product artifact: update it when the documented code or behavior changes.
- Distinctive: line 38
Do not add TypeScript or a build system unless the task explicitly asks for it.
- Distinctive: line 42
Never hide errors that could leave code or documentation stale.
- Common: line 43
Do not commit secrets, generated dependency directories, or temporary files.
- Very specific: line 44
Do not create commits during implementation.
- Distinctive: line 50
Run the smallest relevant check after every non-trivial change.
- Distinctive: line 57
Prefer explicit instructions and the fewest files possible.
- Distinctive: line 67
Only then, write the minimum clear new code that works.
≈ “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
Coding Agent Contract
Scope
The sections below apply only when developing the agent-devkit source
repository. When this file is loaded as a plugin rule in another project,
follow that project's own AGENTS.md and use using-devkit to choose the
workflow; this file imposes no repository-specific policies on that project.
This repository contains dependency-free Markdown skills for keeping coding-agent context and an LLM-facing codebase wiki in sync.
Before changing files
- Read this file. Read
docs/llm/INDEX.mdonly in target projects where the skills have created that wiki. - Inspect the existing implementation and its callers before editing relevant files.
- Reuse the current Markdown skill patterns; do not add scripts, dependencies, or abstractions unless prompt-driven instructions are insufficient.
Working rules
- Keep changes focused and minimal.
- Preserve user changes and do not rewrite unrelated files.
- Follow the team's one-branch/PR-per-task workflow; worktrees are optional. This does not override existing commit or push approval rules.
- For new change folders and for new spec, plan, decision, estimate, or handoff
files, use
YYYY-MM-DD-<issue-id>-<slug>before any existing type suffix when an issue ID appears in the task or a related artifact, including its filename. Without one, keep the existing format; never infer an ID or rename an existing artifact. - When resolving conflicts in
docs/agent-devkit/INDEX.mdordocs/llm/INDEX.md, preserve links from every task and verify all targets after the merge. - Treat generated LLM documentation as a product artifact: update it when the documented code or behavior changes.
- Do not add TypeScript or a build system unless the task explicitly asks for it.
read-codebase-contextandreview-and-verifyuse no-edit subagent briefs. These prompts do not restrict host permissions; parallel execution or subagents assigned file edits need an explicit task.- Never hide errors that could leave code or documentation stale.
- Do not commit secrets, generated dependency directories, or temporary files.
- Do not create commits during implementation. Leave the working tree for the
user to review; commit only after final
review-and-verifypasses and the user explicitly requests it.
Verification
- Run the smallest relevant check after every non-trivial change.
- For workflow changes, exercise the affected skill on a temporary sample repository when its behavior cannot be verified by inspection alone.
- A change is complete only when its behavior and documentation agree.
Change shape
Prefer explicit instructions and the fewest files possible. If deterministic behavior eventually requires automation, prefer one simple script over a framework. After understanding the affected flow, use this ladder and stop at the first option that satisfies the required behavior:
- Does this need to exist at all? Skip speculative work (YAGNI).
- Does it already exist in this codebase? Reuse it.
- Does the standard library do it? Use it.
- Does a native platform feature cover it? Use it.
- Does an already-installed dependency solve it? Use it.
- Only then, write the minimum clear new code that works.
The ladder never removes explicit requirements, validation at trust boundaries, security, accessibility, or error handling that prevents data loss. For bug fixes, the minimum change is the smallest root-cause fix, not a symptom patch.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.

