ClaudeAutoPM
CLAUDE.md39 starsChanged 7 days agoMITView on GitHub
Think carefully. Implement the most concise solution possible. @include .claude/rules/tdd.enforcement.xml @include .claude/rules/coverage-thresholds.xml @include .claude/rules/agent-mandatory.xml @include .claude/rules/context7.xml @include .claude/rules/github-operations.xml @include .claude/rules/naming-conventions.xml @include .claude/rules/command-pipelines.xml @include .claude/rules/issue-structure.xml @include .claude/commands/pm/pm-commands.md @include .claude/agents/agent-registry.xml Full agent descriptions: .claude/agents/AGENT-REGISTRY.md See .claude/templates/xml-prompts/TEMPLATE_REGISTRY.md for staged workflow templates. Builder: lib/xml-prompt-builder.js — invoke with template path and variables. See .claude/rules/ for reference: Every issue opened here carries type:, area: and size: at creation — including issues filed by gh issue create or the REST API, neither of which picks up issue…
In this repository2 other files it gives its agents.
- .github/copilot-instructions.mdCopilot instructions
- add-framework-assetSkill
The file
We split this CLAUDE.md into the rules it gives an agent
11 rules. About 0 are common, 6 are distinctive, and 5 are 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.
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 41
Every issue opened here carries type:, area: and size: at creation — including issues filed by gh issue create or the REST API, neither of …
- Very specific: line 52
No space after the colon: size:M, never size: M — a cross-repo query against the spaced form returns an empty set, which looks exactly like …
- Very specific: line 64
Path rule: Never hardcode autopm/ in framework files — use .claude/
- Distinctive: line 70
Ask questions rather than guessing intent.
- Distinctive: line 71
No flattery. No compliments unless asked.
- Distinctive: line 93
every required check is green and GitHub reports the PR mergeable.
- Distinctive: line 95
no PR comment asks to hold (for example "DO NOT MERGE").
- Very specific: line 98
Never use --admin to get past a missing approval or a missing check.
- Very specific: line 103
No REST poller faster than every 120 s.
- Distinctive: line 105
Every waiter or watcher has an iteration cap and stops on a 403 / "rate limit exceeded".
- Distinctive: line 106
Never leave a background loop running past its purpose: kill it when the PR merges or the task ends.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.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
CLAUDE.md
Think carefully. Implement the most concise solution possible.
CRITICAL RULES (zero tolerance — read on every task)
@include .claude/rules/tdd.enforcement.xml @include .claude/rules/coverage-thresholds.xml @include .claude/rules/agent-mandatory.xml @include .claude/rules/context7.xml @include .claude/rules/github-operations.xml @include .claude/rules/naming-conventions.xml @include .claude/rules/command-pipelines.xml @include .claude/rules/issue-structure.xml
COMMANDS
@include .claude/commands/pm/pm-commands.md
AGENTS
@include .claude/agents/agent-registry.xml
Full agent descriptions: .claude/agents/AGENT-REGISTRY.md
XML PROMPT TEMPLATES
See .claude/templates/xml-prompts/TEMPLATE_REGISTRY.md for staged workflow templates.
Builder: lib/xml-prompt-builder.js — invoke with template path and variables.
OPERATIONAL GUIDANCE
See .claude/rules/ for reference:
standard-patterns.md— Output formats, datetime handling, error messagesfrontmatter-operations.md— YAML frontmatter read/write/stripgit-strategy.md— Branch naming, commit standards, merge workflow
Labels
Every issue opened here carries type:, area: and size: at creation — including
issues filed by gh issue create or the REST API, neither of which picks up issue forms.
type:—feat,bug,chore,infra,spike,docsarea:—framework(autopm/),lib,cli(bin/),packages,install,scripts,test,docs(docs/ + docs-site/),ci(.github/)size:—Sone sitting,Ma day or so,Lmore than a day (split it first)
An epic also carries tracking, and is never dispatched to a worker.
needs-split, not-code and tracking all mean do not dispatch.
No space after the colon: size:M, never size: M — a cross-repo query against the
spaced form returns an empty set, which looks exactly like a backlog with nothing in it.
These exist because dispatching the wrong issue is expensive: an unlabelled epic once
produced seven issues in one 3,074-line PR, three of them in no sprint at all.
.github/workflows/issue-label-guard.yml applies needs-triage to anything filed without them.
PROJECT
- Language: JavaScript/Node.js
- Testing: Jest (
npm test,npm run test:coverage) - Install:
autopm installcopiesautopm/.claude/to target - Path rule: Never hardcode
autopm/in framework files — use.claude/ - Commits: Semantic format, no Claude attribution signatures
TONE
- Be concise. Be skeptical. Criticism welcome.
- Ask questions rather than guessing intent.
- No flattery. No compliments unless asked.
Approvals and merging (fleet policy, owner decision 2026-10-02)
An approval is an approval, whichever account gives it. An APPROVED review from any of
rlagowski, rafeekpro or Dixter999 counts. Don't hold back a merge because the approval came
from a different account than you expected: not the PR author's, not this session's, and not the
"usual" owner's. Two identity rules, both enforced by GitHub: an author can't approve their own PR,
and (since 2026-10-08, owner rule) the account that pushed the latest commit can't approve it either
(require_last_push_approval: true on every org ruleset; pr-review-gate rulesets/branch-protection.md).
Who may approve (owner rule 2026-10-15). The PR's creator and every account that committed to it
can't approve it; any other of the three accounts can. Usually one account creates the PR and makes every
commit, so the other two can approve; if two accounts committed, the third still can. rlagowski is not a
mandatory approver: rlagowski can create a PR and Dixter999 approve it. So keep one identity per PR:
create, commit and push follow-ups as the same account, and never push to a PR as an account that should
approve it — that turns an approver into a committer. Auto-approve approves as github-actions[bot], which
is never a committer.
A session may merge a PR itself once all of these hold:
- an APPROVED review from one of those three accounts is on the PR's current head commit (a push that changes the diff dismisses it);
- every required check is green and GitHub reports the PR mergeable. The ruleset decides whether that approval satisfies any code-owner rule;
- no PR comment asks to hold (for example "DO NOT MERGE").
Merge with gh pr merge <n> --match-head-commit <approved-sha>, using this repo's merge method.
Never use --admin to get past a missing approval or a missing check.
GitHub API polling (fleet rule, owner 2026-10-15)
REST core is 5000 calls/hour per USER, shared by every token and every session of that user. On 2026-10-15 Dixter999 and rlagowski were exhausted for most of the day.
- No REST poller faster than every 120 s. Prefer GraphQL (
gh pr view --json statusCheckRollup,reviews,mergeStateStatus) overgh run list/view,gh api repos/...and RESTgh pr checks. - For repeated reads, use
gh api --cache 60s. - Every waiter or watcher has an iteration cap and stops on a 403 / "rate limit exceeded". Don't retry; report it.
- Never leave a background loop running past its purpose: kill it when the PR merges or the task ends. A 23.7 h orphan loop was found.
- On DIXTER and dixter-pc,
ghreads go out as rafeekpro via a shim; writes stay on your account. Don't work around it.
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.

