project-docs-kit / templates
AGENTS.md1 starsChanged 4 days agoMITView on GitHub
{{SUMMARY: same 2–3 sentence summary as docs/PROJECT.md}} Full detail: docs/PROJECT.md 1. docs/PROJECT.md — goals, scope, non-goals, success criteria 2. docs/ROADMAP.md — the phase with - **Status:** active is the current phase 3. docs/features/ — the features of the active phase (grep -l "phase: <id>" docs/features/*.md); approved and in-progress ones are the current scope 4. docs/tasks/ — tasks with status: in-progress first, then status: todo (lowest number first) 5. docs/issues/ — issues with status: open or investigating may block tasks 6.…
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 repository5 other files it gives its agents.
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 32
All project documents are written in English.
- Very specific: line 35
Before a task, load every row of "Project skills" and every row of "Recommended tools" whose kind includes skill or plugin and whose "Use …
- Very specific: line 41
Never edit a document with status: approved without asking the human first.
≈ “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
{{PROJECT_NAME}}
{{SUMMARY: same 2–3 sentence summary as docs/PROJECT.md}}
Full detail: docs/PROJECT.md
Start here (read in order)
docs/PROJECT.md— goals, scope, non-goals, success criteriadocs/ROADMAP.md— the phase with- **Status:** activeis the current phasedocs/features/— the features of the active phase (grep -l "phase: <id>" docs/features/*.md);approvedandin-progressones are the current scopedocs/tasks/— tasks withstatus: in-progressfirst, thenstatus: todo(lowest number first)docs/issues/— issues withstatus: openorinvestigatingmay block tasksdocs/DECISIONS.md— the decision index; read it, then open only the decision files indocs/decisions/that touch the work, before proposing an alternative to something already decideddocs/{{TYPE_DOC: STACK.md, METHODS.md, STYLE.md or RESOURCES.md}}— {{TYPE_DOC_PURPOSE: purpose line from references/project-types.md}} {{DESIGN_DOCS: "8.docs/design/ARCHITECTURE.md— how the system fits together" if it exists; then "<next number>.DESIGN.md— the UI look and feel; build screens only from its approved mockups" if it exists; then "<next number>.docs/design/— other design notes:<file>,<file>" if docs/design/ holds other files; delete this line if none of these}}
Current state
This file deliberately does not record progress — it would go stale. Derive it:
grep -l "status: in-progress" docs/tasks/*.md # work in progress
grep -l "status: proposed" docs/features/*.md # features waiting for the human
grep -l "status: in-progress" docs/features/*.md # features being built
grep -l "status: todo" docs/tasks/*.md # next up
grep -l "status: open" docs/issues/*.md # open issuesWorking rules
- All project documents are written in English.
- Creating or editing anything in
docs/→ follow.claude/skills/project-docs/SKILL.md. - A request for a new feature or more scope → follow
.claude/skills/project-docs/adding-a-feature.md. - Before a task, load every row of "Project skills" and every row of "Recommended tools" whose
kind includes
skillorpluginand whose "Use when" matches the work. A project skill that your tool cannot load as a skill is plain markdown: open the file and follow it. A recommended tool that is not available is not installed or is turned off for this user: skip it and say so in your report. - Record every significant choice, and what was rejected, as a new file in
docs/decisions/with a row in the indexdocs/DECISIONS.md. - Never edit a document with
status: approvedwithout asking the human first. - Work may start while core docs are
draft; only the human (project owner) setsapproved. - Values marked
Undecidedare open questions — ask the human; do not assume. {{TYPE_RULES: the "Working rules" bullets for this type from references/project-types.md, one per line}}
Project skills
| Skill | Use when | Path |
| project-docs | creating or updating tasks, issues, reports, or any status | .claude/skills/project-docs/SKILL.md |
| {{SKILL_ROWS: one table row per selected project skill; delete this line if none were selected}} |
Skills are plain markdown. Agents without native skill support should open the file and follow it.
Recommended tools (Claude Code)
Installed per user — not part of this repo. Tools that are not installed are skipped (see Working rules).
| Tool | Kind | Use when | Source / install |
| {{TOOL_ROWS: one row per tool the user chose, per references/stack-tools.md §9; if none were chosen, delete the whole section including its heading}} |
Status reference
| Kind | Flow |
| Feature | proposed → approved → in-progress → done (also rejected, cancelled) |
| Task | todo → in-progress → review → done (also blocked, cancelled) |
| Issue | open → investigating → resolved (also wont-fix) |
| Core doc | draft → approved → outdated |
| Phase | planned → active → done |
| Report | draft → final |
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.

