agentleFS
rcdelfinrcdelfin/agentkit/skills/planning/tweak-workflow/SKILL.md

tweak-workflow

Skill2 starsChanged 3 months agoMITView on GitHub

Use for end-to-end project changes: understand, route, plan, implement, verify, and ship.

The file

Show its rules and text

We split this skill into the rules it gives an agent

17 rules. About 0 are common, 12 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    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:

    Use when the user wants a bug fix, feature, or hotfix carried through the repository workflow.

    line 16
  2. Distinctive:

    If destination or safe path remains unclear, stop and ask.

    line 29
  3. Very specific:

    Branch safely: Use git-actions. Follow repository branch conventions, confirm the base when it is ambiguous, and never assume a protected …

    line 31
  4. Very specific:

    Route and plan: Use skill-orchestration to load only skills required by touched code.

    line 35
  5. Distinctive:

    Implement and verify: Follow the plan, then run the strongest applicable tests, typechecks, linters, and builds.

    line 46
  6. Very specific:

    Ship only when requested: Commit and push through git-actions.

    line 50
  7. Distinctive:

    Handle review feedback: If review feedback exists after shipping, use the applicable review workflow to fix, verify, reply, and resolve …

    line 53
  8. Distinctive:

    Never resolve a thread before its fix is verified.

    line 55
  9. Distinctive:

    Do not treat a ticket or request description as proof of root cause.

    line 59
  10. Distinctive:

    Do not continue without a clear destination or resolved blocking decisions.

    line 60
  11. Distinctive:

    Do not claim multi-session work complete without a handoff checkpoint.

    line 61
  12. Distinctive:

    Do not pre-slice unknown work into speculative tasks.

    line 62
  13. Distinctive:

    Do not load every domain skill.

    line 63
  14. Distinctive:

    Do not push unverified code or open an MR/PR without explicit approval.

    line 64
  15. Distinctive:

    Do not resolve review feedback before implementing and verifying the fix.

    line 65
  16. Distinctive:

    No unexplained unstaged changes remain before shipping.

    line 74
  17. Very specific:

    Commit/push/MR/PR actions match explicit user intent.

    line 75

≈ “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

Tweak Workflow

Take an implementable project change end to end: Understand → Route → Plan → Implement → Verify → Ship.

When to Use

Use when the user wants a bug fix, feature, or hotfix carried through the repository workflow. Do not use for plan-only work (tweak), review-only work, or requests to summarize an existing change.

Steps

  1. Set destination and boundary. State the one-sentence done condition, explicit out-of-scope work, and any blocking unknowns. Read repository guidance and relevant code, then classify the change:

    • Bugfix / hotfix → use investigate before editing.
    • Feature / cross-boundary change → use systems-thinking before editing.
    • Trivial one-file change → skip deep investigation and planning.

    If destination or safe path remains unclear, stop and ask; do not invent tasks.

  2. Branch safely. Use git-actions. Follow repository branch conventions, confirm the base when it is ambiguous, and never assume a protected branch is safe to modify directly.
  3. Route and plan. Use skill-orchestration to load only skills required by touched code. For non-trivial work, use tweak to produce proposal, specs, design, and tasks before implementation. Keep unresolved decisions explicit; plan only what recon supports. If blocking decisions remain, stop and ask to clarify or split. If work exceeds one session but its destination is clear, complete the plan and leave an explicit handoff checkpoint in its tasks or handoff note; do not pretend to finish or execute speculative scope. Use scrutinize before implementation when the change is risky, public-contract, security-sensitive, cross-boundary, or has multiple reasonable approaches.
  4. Implement and verify. Follow the plan, then run the strongest applicable tests, typechecks, linters, and builds. Fix failures before any push. Keep the diff focused; do not add speculative scope.
  5. Ship only when requested. Commit and push through git-actions. Create an MR/PR only when explicitly requested. Never force-push or skip hooks.
  6. Handle review feedback. If review feedback exists after shipping, use the applicable review workflow to fix, verify, reply, and resolve each thread. Never resolve a thread before its fix is verified.

Pitfalls

  • Do not treat a ticket or request description as proof of root cause.
  • Do not continue without a clear destination or resolved blocking decisions.
  • Do not claim multi-session work complete without a handoff checkpoint.
  • Do not pre-slice unknown work into speculative tasks.
  • Do not load every domain skill; route from the files and behavior actually touched.
  • Do not push unverified code or open an MR/PR without explicit approval.
  • Do not resolve review feedback before implementing and verifying the fix.

Verification

  • Destination, boundary, and blocking unknowns recorded before implementation.
  • Multi-session work has a plan and explicit handoff checkpoint.
  • Repository guidance read before implementation.
  • Required skills loaded from exact discovered paths.
  • Tests, typechecks, lint, or strongest applicable checks pass.
  • No unexplained unstaged changes remain before shipping.
  • Commit/push/MR/PR actions match explicit user intent.

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.

Reviews can't be read right now.

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.