agentleFS
sssemilsssemil/SKILLS/skills/brutal/brutal-worker/SKILL.md

brutal-worker

Skill2 starsChanged 4 months agoMITView on GitHub

Own one exact BRUTAL.md task or review finding through claim, isolated implementation, verification, stacked pull request publication, and an uncapped material-correctness review/fix loop. Managed workers execute one phase at a time with fresh context.

The file

We split this skill 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
  1. Very specific:

    Always require the stable task reference and read applicable repository rules.

    line 12
  2. Distinctive:

    Read only the supplied phase context file.

    line 26
  3. Distinctive:

    Never merge, approve, request changes, force-push, delete a remote branch, or create a second pull request for the item.

    line 33
  4. Very specific:

    Keep the item in_progress until a fresh review has no CRITICAL or MAJOR.

    line 36
  5. Distinctive:

    When the prompt contains a managed phase snapshot, do only that phase and return the supervisor’s exact result.

    line 41
  6. Distinctive:

    Read the full item, history, parent/blockers, assignment, and existing PR link.

    line 46
  7. Very specific:

    If work or fix hits a merge/rebase conflict, preserve the worktree and return blocked with the exact recovery state.

    line 67
  8. Very specific:

    After a matching PR is verified merged, record acceptance, move the item to done, and return terminal clean with completion_kind: merged.

    line 82
  9. Distinctive:

    Read and claim the exact item.

    line 94
  10. Distinctive:

    Commit and push normally. Find the PR by exact marker and head before creating one against the handed-off base:

    line 97
  11. Very specific:

    Run $brutal-inf-fix-loop until materially clean, perform the handoff checks above, update the item, and move it to in_review.

    line 108

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Brutal Worker

Complete one exact work item and its pull request. Never select a replacement.

Required Context

Always require the stable task reference and read applicable repository rules. Read ../brutal-shared/integration-resolver.md and ../brutal-shared/support/contracts.md. Reuse a managed assignment’s validated work-store/code-host identities, canonical local root, worktree, branch, and base; resolve only missing data needed by the current phase.

For standalone work, also read ../brutal-swarm/references/parallel-execution.md, create or resume the isolated worktree with ../brutal-swarm/scripts/worktree_manager.py, and read ../brutal-inf-fix-loop/SKILL.md plus its source skills.

For managed work, the immutable handoff establishes assignment identity; the phase snapshot supplies revalidated mutable state. Runtime names, logs, and old snapshots are observational and grant no provider authority. Read only the supplied phase context file.

Hard Rules

  • Own only the handed-off item, branch, worktree, and pull request.
  • Select only task or review_finding; never implement a plan or investigation.
  • Never merge, approve, request changes, force-push, delete a remote branch, or create a second pull request for the item.
  • Preserve unrelated changes and edit only the isolated worktree.
  • Keep the item in_progress until a fresh review has no CRITICAL or MAJOR. Then it may move to in_review; move it to done only after its PR merged.

Managed Phase Protocol

When the prompt contains a managed phase snapshot, do only that phase and return the supervisor’s exact result. Do not continue into the next phase.

work

Read the full item, history, parent/blockers, assignment, and existing PR link. Prove the exact claim from the expected state before changing code. Validate worktree, branch, base branch/SHA, and branch task marker. Implement, test, self-review, commit, push, and find-or-create the single marked PR. Return a checkpoint with current base/head, PR, verification, and zeroed review fields.

review

Read ../brutal-inf-fix-loop/SKILL.md and ../brutal-pr-review/SKILL.md. Use only the exact PR/base/head snapshot and relevant repository rules. Run one fresh material_convergence pass. Return counts for all severities, review id, queued/unhandled counts, summary status, and residual MINOR/NIT findings. Exit after the checkpoint.

fix

Read ../brutal-inf-fix-loop/SKILL.md and ../brutal-pr-finding-fixer/SKILL.md. Drain the full generated all-severity queue for the supplied review occurrence, verify, push, re-read base/head, and return a checkpoint. Do not start a review.

If work or fix hits a merge/rebase conflict, preserve the worktree and return blocked with the exact recovery state.

handoff

This phase is reached only after a fresh review has zero CRITICAL and zero MAJOR. Revalidate PR identity, base/head, task ownership, queue state, and required checks. Reuse same-head verification only when the complete base/head snapshot still matches and checks still pass; otherwise rerun it. Record changed files, commands/results, commits, PR/base, review id, completion kind, and residual MINOR/NIT findings on the task. Move it to in_review and return terminal clean with zero_findings or materially_clean.

complete

After a matching PR is verified merged, record acceptance, move the item to done, and return terminal clean with completion_kind: merged. This is the only preassigned phase that replaces the scheduler’s post-merge finalize action; never use finalize as a managed worker phase.

The supervisor alone derives legal phase transitions. A fresh phase uses fresh execution context. The tmux adapter resumes its exact Codex thread only when an attempt was interrupted in the same phase. Return blocked, canceled, claim_lost, or failed with exact recovery state on hard stops.

Standalone Lifecycle

  1. Read and claim the exact item. Resume a matching open PR/branch; complete a matching merged PR; block on closed-unmerged or identity mismatch.
  2. Confirm the isolated branch/base and implement the smallest verified change.
  3. Commit and push normally. Find the PR by exact marker and head before creating one against the handed-off base:

    <!-- brutal-worker:v1:<stable-task-ref> -->
    Source: brutal-worker
    Work item: <stable ref or URL>
    Parent: <stable ref or none>
    Blocked by: <stable refs or none>
    Stack base: <branch and blocker ref or root>
    Verification: <commands and results>
  4. Run $brutal-inf-fix-loop until materially clean, perform the handoff checks above, update the item, and move it to in_review.

Redirect verbose builds, tests, diffs, and provider output to run-local temporary logs. Report command status and duration. On failure, include only the last 200 lines or 16 KiB, whichever is smaller.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.