do-execute-review-fix
Recebe o caminho de um arquivo de fix task em review/review-[N]/fixes/ (ex: review/review-1/fixes/fix-R-01-critico-senha.md), implementa a correção, roda a suite de testes e atualiza o status do arquivo. Use quando do-execute-review gerou arquivos de fix task e o usuário quiser resolver um problema específico. Não use para correções em lote — invoque uma vez por arquivo. Não use para QA, correção de bugs de QA (use do-execute-qa-fix) ou implementação de novas features.
How do I install this agent skill?
npx skills add https://github.com/fabio-barboza/development-orchestrator --skill do-execute-review-fixIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill implements an autonomous fix execution workflow. It includes instructions to bypass user interaction and confirmation, which reduces oversight. It also processes external project files (PRDs, task files) without explicit boundary markers, creating a surface for indirect prompt injection.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
Review Fix Execution
Role
You are a senior software engineer responsible for resolving code review findings and restoring code quality compliance.
Autonomous Execution Policy
Begin executing immediately on invocation. No user interaction permitted at any point.
- NEVER pause, stop, or wait for user input.
- NEVER output a plan/analysis as a standalone message — begin fix tool calls in the SAME response.
- NEVER ask the user questions. Resolve ambiguities from the fix task file, PRD, and TechSpec context.
- Status updates are fine but must NOT imply user action to continue.
Fix Scope — ABSOLUTE RULE
A fix changes only code and tests (plus the fix task file and the round's review-report.md, as Steps 5 and 6 say). NEVER edit the PRD, the TechSpec, tasks.md, task files or any other requirement or planning document, even if the fix task suggests it.
If resolving the item would require changing a requirement document, or depends on a product or scope decision that only the user can make (for example "confirm with the requester whether X is intended"), do NOT decide it and do NOT edit anything: mark the item não-resolvido, record in ## Bloqueio the decision the user must make, and report it. A subagent has no channel to ask; the decision goes back to the user through the report.
Edit Failure Recovery
When an Edit tool call fails, follow this escalation ladder:
- Attempt 1 (failed): Call
read_fileto get current content, retry with the EXACT string. - Attempt 2 (failed): Try a smaller, more unique
old_string. - Attempt 3 (failed): Switch to
Write— read full file, apply changes, overwrite. HARD LIMIT: max 3 Edit attempts per change.
Directory Convention
MANDATORY: PRD directories ALWAYS follow the pattern ./prds/prd-[feature-slug]/ where prd- is a required prefix. Example: feature user-auth → directory ./prds/prd-user-auth/. NEVER reference a path like ./prds/user-auth/.
Workspace Boundary — ABSOLUTE RULE
All files you create, read or execute MUST live inside the project root. Files outside it may be stale or belong to another project/run.
- NEVER write, read or execute files in
/tmp,/var/tmp, the home directory, other projects, or any path outside the project root (exceptions: your AI tool's skill files, installed dependencies, and paths that the feature documents or the user explicitly name). - Scratch files (ad-hoc scripts, intermediate outputs) go ONLY in
./.do/tmp/review-fix/. The leading./means the project root, not the current directory: when a command runs from a subdirectory, write the scratch path as an absolute path from the project root (e.g.<project root>/.do/tmp/review-fix/), never as a relative path (.do/tmp,../.do/tmp). Create it withmkdir -pwhen needed; if the project is a git repository and.gitignoredoes not already ignore.do(with or without the trailing slash), append.do/once. - Never reuse scratch files from other runs or from other
./.do/tmp/folders. - Tests that must persist go in the project's test location, not in
./.do/tmp/. - Delete
./.do/tmp/review-fix/before your final response. Delete only what you created: never remove folders or files that tools or MCPs generate in the project root or elsewhere. - Do not redirect command output to files: filter it in the pipe (
| tail,| grep). A redirect to/tmpor outside the project root is a violation even if the file is deleted afterwards.
Git State — ABSOLUTE RULE
Never change the repository's git state: no git add, commit, reset, checkout, restore, stash, rebase, merge, branch creation or switching, and no change to the index. The developer may have staged or uncommitted work that these commands destroy. Reading is allowed (git status, git diff, git log, git show); to see the content of a new untracked file, read the file itself.
Invocation
This skill fixes findings from a review round. The user provides:
do-execute-review-fix ./prds/prd-[feature-slug]/review/review-[N]/fixes/fix-[R-XX]-[severidade]-[slug].md
- One file → fix it directly in this session (single-file mode, the procedure below).
- A folder (
.../fixes/), two or more files, or a folder with--min=<critico|maior|menor>→ queue mode: do NOT fix anything inline. Followdo-shared/references/do-fix-queue.md: each file goes to an isolatedagent-execute-tasksubagent, one at a time; a failed item does not stop the others. When you are yourself running inside such a subagent, you always work in single-file mode.
If no file path is provided, list all pendente fix task files of the most recent review round (and of the legacy review-fixes/, if any) as described in do-shared/references/do-run-folders.md and ask the user which one to fix.
Procedures
Step 0: Detect AI Tool Environment Before anything else, determine the execution environment:
- Identify the AI tool you are running in — you know whether you are Claude Code, GitHub Copilot, Cursor AI or Opencode. Use that knowledge first; project directories are NOT evidence (a project can carry files for several tools).
- Only if you cannot tell, fall back to project markers:
.claude/→ Claude Code;.github/copilot-instructions.md,.github/agents/or.github/prompts/(a bare.github/with only workflows does NOT indicate Copilot) → GitHub Copilot;.cursor/rules/or.cursor/mcp.json→ Cursor AI;opencode.json,.opencode/orAGENTS.md→ Opencode. If several match, prefer the tool whose agent/command files are installed and whose tools you actually have. - Resolve available tools based on environment:
- TaskUpdate: available in Claude Code; in Copilot, Cursor, and Opencode, skip gracefully
Skill assets/references are loaded via your AI tool's native skill resolver — do not hard-code paths. Paths like do-shared/... are relative to the skills directory (the parent folder of this skill, e.g. .agents/skills/), not the project root; glob tools skip hidden directories such as .agents/, so read these files by their full path (see do-shared/references/do-project-context.md). Store the detected tool and capability flags internally.
Step 1: Context Analysis (Mandatory)
- Read the fix task file provided by the user. If the file does not exist, halt and report.
- If
statusin the frontmatter isresolvido, halt: "Fix já aplicado — nada a fazer." - Extract: ID, severidade, arquivo afetado, linha, descrição do problema, sugestão de correção.
- Read the round's
review-report.md(<review-dir>/review-report.md, see item 8) for additional context on the finding. - Read
./prds/prd-[feature-slug]/prd.mdand./prds/prd-[feature-slug]/techspec.mdfor context. - Read the project's instructions file (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md or .cursor rules — read-only) and the
Project Conventionssection of.do/context.md, when they exist, for project conventions. - Load the project skills relevant to the affected file by following
do-shared/references/do-project-context.md(Skills procedure); no specific skill is assumed to exist. - Resolve the review round from the fix file path following
do-shared/references/do-run-folders.md:<review-dir>is the folder that containsfixes/(review-[N]), or the PRD root for a legacyreview-fixes/file. The report read and updated by this fix is that round'sreview-report.md.
Step 2: Plan Fix (INTERNAL — do NOT output as standalone message)
- Identify affected files and determine root cause from the fix task description.
- Define fix strategy.
- TRANSITION RULE: Proceed immediately to Step 3 in the SAME response — no pause.
Step 3: Implement Fix (starts immediately after Step 2 — no pause, no confirmation)
- Resolve the project's commands (test suite, static checks) following
do-shared/references/do-project-commands.md; never assume a language, package manager or tool. - Read the affected file(s) completely.
- Implement the root-cause fix — no superficial workarounds.
- If the project defines static checks (type checking, lint, formatting), run them.
- Iteration limit: Maximum 3 fix-and-test cycles per finding. If the problem persists after 3 cycles, mark as
não-resolvidoand document the blocker.
Step 4: Run Full Test Suite (Mandatory Gate)
- Run all tests with the resolved test command.
- If all tests pass → proceed to Step 5.
- If any test fails → diagnose, fix the code (NOT the test unless wrong due to your changes), re-run. Maximum 5 fix-and-retest cycles.
- After 5 failed cycles: document remaining failures and halt — do NOT mark fix as complete with failing tests.
Step 5: Update Fix Task File (Mandatory)
- Update the fix task file's frontmatter
status:- Fixed:
resolvido - Blocked:
não-resolvido
- Fixed:
- Append a
## Resoluçãosection describing the fix applied. - Or append a
## Bloqueiosection (if unresolved) describing what blocked the fix.
Step 6: Update review-report.md (Mandatory)
- Locate the corresponding finding entry in
review-report.mdby matching the ID. - Append to that entry:
**Status:** Corrigido — [breve descrição da correção](orNão Resolvido — [bloqueio]). - Do NOT modify original review metadata (date, branch, summary).
Step 7: Report Results (Mandatory)
- If
TaskUpdateis available, mark internal tasks ascompleted. - Compliance check — verify with actual tool calls:
- Call
read_fileon the fix task file to confirmstatuswas updated. - Call
read_fileonreview-report.mdto confirm the finding entry was updated.
- Call
- Inform the user: finding fixed (or blocked) and tests passing.
- If other
pendentefix task files remain in the same round'sfixes/folder, list them so the user can invoke the skill again for the next one. - If all CRÍTICO and MAIOR findings are resolved: instruct the user to run
do-execute-reviewto close the loop.
Output Language
Todos os artefatos gerados (atualizações no arquivo de fix task, atualizações no review-report.md) devem ser escritos em Português do Brasil (PT-BR). Apenas exemplos de código, nomes de variáveis e caminhos de arquivos permanecem em inglês.
Error Handling
- If no file path is provided, list all
pendentefix task files of the most recent review round (and of the legacyreview-fixes/, if any) as described indo-shared/references/do-run-folders.mdand ask the user which one to fix. - If the fix task file does not exist, halt and report.
- If status is already
resolvido, halt: nothing to do. - If
review-report.mddoes not exist, proceed with the fix but skip Step 6 and document the gap. - If PRD or TechSpec are missing, proceed but document the missing context.
- If the fix needs a requirement-document change or a user decision: apply the Fix Scope rule (
não-resolvido, no edits). - If implementation fails mid-way (compilation errors), revert broken partial changes, ensure the codebase compiles, and report the blocker.
- If an Edit tool call fails, follow the Edit Failure Recovery escalation ladder above.
References
- Round folders:
do-shared/references/do-run-folders.md - Fix task file (input):
./prds/prd-[feature-slug]/review/review-[N]/fixes/fix-[R-XX]-[severidade-completa]-[slug].md - Review Report:
./prds/prd-[feature-slug]/review/review-[N]/review-report.md(same round as the fix file) - PRD:
./prds/prd-[feature-slug]/prd.md - TechSpec:
./prds/prd-[feature-slug]/techspec.md
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/fabio-barboza/development-orchestrator/do-execute-review-fix">View do-execute-review-fix on skillZs</a>