mohammadhprp/teammate/run-rework
Claude Code and agent skill by mohammadhprp/teammate
- community
Checking that the SKILL.md file is still in the repository.
Turn a failing review into a bounded fix loop that converges or stops: send only the open blocker and major findings back to the same worker, grouped by severity with file/line and the expected behavior, ban scope expansion, require the same report, then monitor and re-review until it passes or hits max_iterations. Use when a review returns a fail verdict with open blocker/major findings, when a rework round comes back still failing or regressed, or when the developer requests changes — not when only minor/nit findings remain.
Install
npx vskill@latest install mohammadhprp/teammate/run-reworkSKILL.md
The same SKILL.md was published earlier as mohammadhprp/teammate/run-rework.
Run rework
A fail verdict is not the end of a task; it is the next instruction. Feed the worker exactly what is wrong, in the least disruptive way, and repeat until the work passes or the budget runs out. Bounding this loop is what stops a review disagreement from quietly consuming the developer's time.
When to use
- A review returns
failwith at least one openblockerormajorfinding. - A rework round came back still failing, or the fix introduced a regression.
- The developer asked for changes (request changes) on an otherwise complete
task.
When not to use
- Only
minor/nitfindings remain: document them in the report and proceed;
they do not block a pass.
- The task is still running:
monitor-agents. - No review has run yet:
review-workfirst — you cannot send findings you do
not have.
Inputs
- The open
blocker/majorfindings recorded on the task
(python3 scripts/tm.py task show <id>), each with severity, category, file, line, and suggestion.
- The same worker that wrote the change, and its task for the current iteration
and max_iterations.
Procedure
- Check the budget. Compare the task's iteration with
max_iterations
(python3 scripts/tm.py task show <id>). At the limit, do not send another round — escalate with the full report and one question via escalate-decision.
- Write a feedback brief of findings only. Pull the open findings from the
ledger (task show <id>); group them by severity, blocker first, and for each give the file and line, what is wrong, and the behavior you expect. Keep out anything already fixed, any praise, and any new requirement. Close with the boundaries: fix only these findings, do not expand scope or touch unrelated files, re-run the project's own checks, and reply in the same handoff-report shape. Narrow, testable feedback is what keeps the next review about the fix rather than about discovering what else changed.
- Send it to the same worker. Prefer continuing the worker that wrote the
change when the harness supports it (for example codex send_input); a fresh subagent call carries no memory, so if you start a new one its brief must be self-contained — the findings and the current state of the tree. Write the brief with tm brief so it lands in state_dir/<project>/briefs/ (the project comes from --task), then pass it to the harness's subagent tool.
``bash python3 scripts/tm.py brief "<name>" --task "<id>" <<'EOF' <feedback brief> EOF # call the harness subagent tool with the brief (or resume the worker where # the harness supports it); then wait for it to return (monitor-agents) ``
Keep the feedback inside the worker's sandbox: inline every fact, and never reference a path outside the project.
- Record the round. This skill owns the iteration increment —
review-work
records only the verdict, so the two do not both increment. python3 scripts/tm.py task update <id> --status rework --iteration <n+1>.
- Monitor, then re-review. Follow
monitor-agentsuntil the worker
returns, then review-work. Collect fresh evidence every round — a diff from a previous iteration proves nothing about the current tree.
- Decide.
pass→ready_for_approval.fail→ back to step 1 if the
budget allows; otherwise escalate.
Output
Either a converged pass or a bounded escalation with the full iteration history.
Failure and escalation
- Non-convergence at
max_iterations: escalate (escalate-decision) with the
iteration history, the open findings, and one clear question.
- The same finding fails twice: the criteria or the brief may be wrong, not the
worker. Escalate rather than repeat an identical round — a loop that cannot converge is the developer's call.
- The worker disputes a finding, or its fix would break scope: do not overrule
it silently; escalate the disagreement with the evidence.
- The worker returned a question or failed: escalate; never impersonate it or
answer its question. If the original worker is gone, re-delegate the findings with delegate-task only when its context can be recovered, otherwise escalate.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingsrc/skills/run-rework/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/mohammadhprp/teammate/run-rework)