build-brief
Skill1 starsChanged 2 months agoApache-2.0View on GitHub
Execute a dispatch brief on the GitHub Actions runner that `claude-build.yml` starts -- the standard builder operational block, but the workflow's checkout IS the worktree. Invoked as `/build-brief <brief body>` by that workflow's automation-mode prompt; never invoke it by hand against a checkout the orchestrator or another agent owns.
- Commits and pushes
3 other projects use a version of this.
Adopted by
Most recently taken up by
- arnonsang/skillstheir own version, first seen Nov 1, 2026
- devexpert-io/harness-startertheir own version, first seen Sep 26, 2026
- SanketSapkal/founder-skillstheir own version, first seen Apr 30, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- SanketSapkal/founder-skills★ 10same name
- arnonsang/skills★ 4same name
- devexpert-io/harness-starter★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- userdocs/llms-full.txtllms.txt
- userdocs/llms.txtllms.txt
- briefSkill
- drive-installed-appSkill
- manual-checksSkill
- run-millSkill
- mill-devSkill
- mill-useSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
0
Distinctive
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 14
Never a second review or merge path -- the reviewer subagent, the PR's ## Review section, and CI's ci-gate stay the authority exactly as …
- Very specific: line 37
Load the rules: Read every file under .claude/rules/*.md and .claude/skills/brief/divergences.md before touching code -- same requirement …
- Very specific: line 62
Add a Claude-Session: trailer only if this run's own session URL is available in the environment.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Running a brief on a runner
This skill runs the SAME contract a local worktree builder follows
(CLAUDE.md's orchestrator/builder model, .claude/rules/*.md,
.claude/skills/brief/divergences.md) with one substitution: there is
no separate worktree to create, because claude-build.yml's checkout
already sits on the brief's fresh branch (git checkout -b "$BRANCH",
run before this skill starts) inside a runner nothing else is using.
Never a second review or merge path -- the reviewer subagent, the PR's
## Review section, and CI's ci-gate stay the authority exactly as
they are for a local builder.
Input
The brief's BODY text arrives as this skill's argument (the text after
/build-brief in the prompt) -- never a docs/goals/briefs/<n>.md
path. docs/goals/ is git-ignored and local-only (it carries the
owner's verbatim words and work context); a runner checkout never has
it, so a path argument here is always a defect in the dispatch, not
something to work around by inventing content.
Two more values ride in the job's environment, set by the workflow
from its own inputs (read them with the Bash tool, e.g. echo
"$BRANCH" -- they are not part of the prompt text):
$BRANCH-- already checked out; never create or switch to a different branch.$ARM_AUTO_MERGE--"true"or"false"; gates the auto-merge step below.
Steps
- Load the rules. Read every file under
.claude/rules/*.mdand.claude/skills/brief/divergences.mdbefore touching code -- same requirement as any builder brief. - Ground-truth the brief. For every file:line the brief cites, read that file at that location before trusting the citation; a brief's claim about the code is a hypothesis, not a fact.
- Implement exactly the brief's design contract. A design question the brief does not answer is reported in this run's final message, never decided here -- stop short of the undecided part and finish whatever the brief DID fully specify.
- Run the brief's own gates. Lefthook is not installed on this
runner -- run the exact
scripts/check-*.shcommands,go test,npm run test(Vitest), andnpx playwright testinvocations the brief names, directly, the same way lefthook would have run them locally. A gate this runner cannot execute (a manual-only check) is named as skipped in the final report, never silently omitted. CI'sci-gateon the opened PR is still the authority these local runs are rehearsing, not replacing. - Dispatch the
reviewersubagent (theAgenttool, same as a local builder) ongit diff origin/main...HEADagainst the brief text. Fix every Important finding before opening the PR. - Commit with the standard trailers:
Add aCo-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>Claude-Session:trailer only if this run's own session URL is available in the environment; omit the line rather than fabricate one. - Push
$BRANCHand open the PR (gh pr create,$GH_TOKENis already set) with:- Title from the brief's
## PRsection. - Body ending in the standard footer:
🤖 Generated with [Claude Code](https://claude.com/claude-code) - A
## Reviewsection holding the reviewer's report in the gate's required shape: aNit —/Important —/Nothing to reportline per finding,Contract match: yes, and the exact lineImportant findings open: 0.scripts/check-review-report.shenforces this shape as CI'sreview-reportjob; get it right before opening, not after a rejection.
- Title from the brief's
- Arm auto-merge only when
$ARM_AUTO_MERGEis"true":gh pr merge --auto --squash <n>. When it is"false", leave the PR open and say so in the final report -- never merge and never silently skip stating which happened. - Report as this run's final message (there is no orchestrator
session reading files back from this runner): the PR URL and
gh pr view <n> --json number,stateoutput, the gate outputs from step 4, the reviewer's findings, and the auto-merge decision from step 8. This message is the only artifact the dispatcher sees.
Dispatching this workflow
gh workflow run claude-build.yml --repo millhq/mill \
-f branch=goal/<n>-<slug> \
-f brief="$(cat brief.txt)" \
-f arm_auto_merge=truebrief.txt holds the brief BODY only. It must never contain
owner-verbatim or internal work-context text -- docs/goals/ is
local-only precisely because that material never leaves this machine;
a brief bound for this workflow is written clean of it from the start,
not redacted after the fact.
workflow_dispatch requires claude-build.yml to already exist on the
ref it runs against; with no --ref given it dispatches against the
repository's default branch. Add --ref <branch> to run it from a
branch that has not merged yet (how this workflow proves itself before
its own PR merges to main):
gh workflow run claude-build.yml --repo millhq/mill \
--ref <branch-with-this-file> \
-f branch=goal/<n>-<slug> -f brief="$(cat brief.txt)"Poll the result with gh run list --repo millhq/mill --workflow
claude-build.yml -L 1, not by watching -- runs take minutes, not
seconds.
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.

