execution
Skill0 starsChanged 27 days agoMITView on GitHub
Implements code changes from a design, with tests and merge-ready output. Use when the user is building what a design or task describes. Not for investigation (specify), design (design) or one-off scripts.
11 other projects use a version of this.
Adopted by
Most recently taken up by
- mkritter3/codex-paired-superpowerstheir own version, first seen Nov 15, 2026
- jircdev/crew-plugintheir own version, first seen Nov 1, 2026
- ceeinbeta/pm-skillstheir own version, first seen Nov 1, 2026
- CrowdStrike/fusion-skillstheir own version, first seen Oct 30, 2026
- mcveloria/pm-skillstheir own version, first seen Oct 25, 2026
- dynos-fit/dynos-worktheir own version, first seen Oct 13, 2026
Other versions11 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 11
- Towow-ai/Flowness★ 107same name
- CrowdStrike/fusion-skills★ 31same name
- metraton/gaia★ 4same name
- ceeinbeta/pm-skills★ 3same name
- mcveloria/pm-skills★ 3same name
- dynos-fit/dynos-work★ 2same name
- SoFarSoGoodya/mathmodel-weave★ 2same name
- jircdev/crew-plugin★ 1same name
- luisabwk/kraken★ 1same name
- Leostruka/devin-bundle★ 0same name
1 more not listed
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 repository22 other files it gives its agents.
- AGENTS.mdAGENTS.md
- claude-arsenal/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- coverage-gapsSkill
- dep-upgradeSkill
- designSkill
- explore-ideaSkill
- gate-checkSkill
- githubSkill
- initSkill
- litcriticSkill
- mutmut-reportSkill
- pin-checkSkill
- pypi-releaseSkill
- python-bootstrapSkill
- queue-addSkill
- queue-nextSkill
- queue-statusSkill
- reviewSkill
- session-endSkill
- shipSkill
- specifySkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. 0 are common, 8 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 10 edited versions we read keep these.
8
Distinctive
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
- Distinctive: line 17
Keep the notes' Resume section (Decided / Ruled out / Next step) current after RED, after GREEN, and after each decision: it is what the …
- Distinctive: line 31
Where the work lives: Run git status --short and check the branch.
- Distinctive: line 44
Run every later step from the worktree.
- Distinctive: line 53
RED: Before changing production code, write the check that proves the gate and see it fail for the expected reason: a regression test that …
- Distinctive: line 66
Read the target code first and match its style, naming and structure.
- Distinctive: line 72
When the fix is an API-contract mismatch (arity, kwargs, return shape), pin the contract on the interface itself, so the next caller making …
- Very specific: line 76
RECORD: Add the gate evidence to the plan's Evidence log: measured value, exact command, commit SHA and environment provenance (ci, local, …
- Very specific: line 83
Run the host repo's lint and the tests for the affected area once (make lint, make test, pytest <path>, …) and fix what fails.
- Distinctive: line 88
no debug code, commented-out blocks or stray TODOs.
- Distinctive: line 89
no hardcoded secrets, URLs or credentials.
- Very specific: line 95
Before opening the PR, run the pre-PR review: one cold reviewer subagent that sees only the packet.
- Very specific: line 101
Write the description (linking the ticket or design), add reviewers, and see CI pass.
- Very specific: line 115
When arsenal/project/<WORKSPACE>/ exists, read spec.md, plan.md and context.md from there and record gate evidence in that plan.md.
“Common” is measured: kept by at least 6 in 10 of the 10 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
Execution Workflow
CANARY: execution-loaded-2026-05-19-4597aff5bb98dd36
Reads status/specification.md and status/plan.md and updates the plan's task
statuses as work progresses. Per-task notes go in tmp/<task-id>-notes.md
(gitignored); load references/template.md when creating one.
Keep the notes' Resume section (Decided / Ruled out / Next step) current after RED, after GREEN, and after each decision: it is what the session reads back after context compaction.
Work through every task in the plan and on to the PR without pausing between tasks for confirmation; stop only when blocked or before a step that is hard to undo. Deliver the scope the plan asks for, and list pre-existing bugs and unrequested cleanup as follow-ups in the PR description rather than fixing them in this diff.
Steps
Step 1: Prepare
- The design is approved and the branch is created from the default branch.
Where the work lives. Run
git status --shortand check the branch. A clean tree on the default branch (or on a finished, unrelated branch) works in place. A dirty tree, an in-flight branch the user may still be using, or a long-running process pinned to the checkout gets a worktree, because a worktree never destroys state:REPO=$(git rev-parse --show-toplevel) TICKET=<ticket-id> DEFAULT=$(git -C "$REPO" symbolic-ref --short refs/remotes/origin/HEAD 2>/dev/null || echo origin/main) git -C "$REPO" worktree add "../${TICKET}-worktree" -b "${TICKET}-short-description" "$DEFAULT"Run every later step from the worktree; remove it with
git worktree removeonce the PR merges. Ask the user only when the choice is genuinely ambiguous.- List the files and tests to create or change, in dependency order, and read the surrounding code for the patterns to follow (naming, structure, error handling).
Step 2: Implement each task — RED → GREEN → RECORD
Each task's Gate in status/plan.md is its measurable acceptance condition.
RED. Before changing production code, write the check that proves the gate and
see it fail for the expected reason: a regression test that reproduces the bug, a
test that specifies the missing behaviour, or, for a metric gate, the measurement
showing the current value misses the threshold. Cover new branches, endpoints and
edge cases (null/empty, boundaries, error paths). Name tests
test_<what>_<condition>_<expected_result>. Put a new test in a file that is not
already the suite's slowest, since the longest file sets the parallel suite's wall
clock; load claude-arsenal:core:init § references/performance-tuning.md (§ The
long pole) when a test file needs splitting. With <!-- test-discipline: test-after --> in the host CLAUDE.md, write
the test alongside the change instead.
GREEN. Implement until the check passes:
- Read the target code first and match its style, naming and structure.
- One concern per commit; type hints or strict types; handle errors explicitly; configuration through environment variables or constants.
- Comments: match the file's density, default to none, and add one only for a why the code cannot show (a hidden invariant, a surprising edge case). Leave task and PR references out of the source.
- When the fix is an API-contract mismatch (arity, kwargs, return shape), pin the contract on the interface itself, so the next caller making the same mistake is caught too.
RECORD. Add the gate evidence to the plan's Evidence log: measured value,
exact command, commit SHA and environment provenance (ci, local, or a project
tag). The gate-check skill's run_gate.py --id <task> <measured> reports PASS/FAIL. A task is done when its gate passes and the row is recorded; a
plan with no Gate column predates the convention, so skip the record there.
Step 3: Check
Run the host repo's lint and the tests for the affected area once
(make lint, make test, pytest <path>, …) and fix what fails; this is the real
check behind reporting the code done. When a check cannot run here, say which and
why. Refactor with the green suite as the safety net, then confirm:
- no debug code, commented-out blocks or stray TODOs;
- no hardcoded secrets, URLs or credentials;
- the diff matches the design scope;
- the PR description says what changed, why, and how to test it.
Step 4: Independent review
Before opening the PR, run the pre-PR review: one cold reviewer subagent that sees
only the packet. Protocol, exits and round cap:
claude-arsenal:core:init § references/pre-pr-review.md.
Step 5: Create the PR
Write the description (linking the ticket or design), add reviewers, and see CI
pass. For several PRs that merge in sequence, follow the github skill's stacking
reference.
Abbreviation
Abbreviated execution = Steps 2, 3 and 4, where the host repo's CLAUDE.md
allows it. Tests, gate evidence and the independent review stay, because
abbreviating skips ceremony, not the checks the author cannot do on their own work.
Workspace-aware paths
When arsenal/project/<WORKSPACE>/ exists, read spec.md, plan.md and
context.md from there and record gate evidence in that plan.md. Under the task
queue, the claimed task's payload at arsenal/tasks/<id>.md carries the acceptance
gate, which the queue's gate runner executes before the task is released.
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.

