L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Développement · SKILL.md

dev-fix

dev-fix

Unified developer workflow for fixing bugs. Analyzes issue-tracker context, cross-checks docs/code, proposes a solution, implements the fix, verifies locally, and delivers a PR/MR.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: dev-fix description: "Unified developer workflow for fixing bugs. Analyzes issue-tracker context, cross-checks docs/code, proposes a solution, implements the fix, verifies locally, and delivers a PR/MR." metadata: internal: true triggers: keywords: - dev fix - workflow

Dev Fix Skill

[!IMPORTANT] Unified developer workflow for fixing bugs. Analyzes issue-tracker context, cross-checks docs/code, proposes a solution, implements the fix, verifies locally, and delivers a PR/MR.

Optional args: slug=, ticket=<id/url>, mode=interactive|autonomous|channel, channel=, auto_continue=true|false, profile=business|hybrid|technical.

Instructions

When the user asks to perform this workflow, execute the following steps:

Dev-Fix — Bug Remediation

Goal: Take a bug ticket from root-cause analysis through a locally-verified PR/MR, enforcing a strict Propose -> Approve -> Verify cycle.

Input

/dev-fix <issue-url-or-key>

Workflow

Step 0: Environment Prep (Turbo)

// turbo

  1. Sync Registry: git pull origin main in the standard repo.

Step 1: Research & Discovery (Implementation Plan Phase)

[!TIP] Sub-Agent Delegation: If your platform supports sub-agents, delegate ticket extraction to specialist-jira-analyst and context lookup to specialist-codebase-scout. If sub-agents are NOT supported, execute these steps yourself.

  1. Analyze Ticket: Use installed Jira/GitHub/GitLab/ADO MCP first; otherwise use exported ticket text. Extract Reproduce steps, Expected Result, and Actual Result.
  2. Cross-Check Context: Use knowledge-base MCP when configured; otherwise use local code search. Locate relevant code.
  3. Create Implementation Plan:
    • Use the Implementation Plan Template below.
    • Initialize project-local docs/prd/prd-plan-[slug].md.
    • Goal: Clear description of the root cause.
    • Proposed Changes: Exact files and logic to be modified.
    • Verification Plan: Detail which QE skill will be used to verify the fix locally before PR.
    • Do not propose code changes until repro steps, expected result, and root cause hypothesis are explicit.
    • SNC: score per common-task-complexity-routing; record snc_tier/model_tier in the plan.
  4. HARD STOP: Request user approval for the implementation plan. Mandatory at snc_tier=high; medium requires self-review first; low may skip approval in autonomous mode only.
  5. Readiness Gate: Run implementation-readiness; code only after READY or approved PARTIAL.

Step 2: Implementation (TDD Phase)

[!TIP] Sub-Agent Delegation: For the actual fix, delegate the TDD loop to specialist-tdd-implementer. If sub-agents are NOT supported, execute the TDD loop yourself using the TDD skill matched in AGENTS.md.

  1. Worktree Branching: Create a new worktree for the fix using git worktree add ../<ticket-key> -b fix/<ticket-key> and cd into it.
  2. Task Tracking:
    • Use the Task Template below.
    • Initialize project-local docs/srs/srs-task-list-[slug].md.
  3. Code: Implement the fix using common-tdd or the @specialist-tdd-implementer sub-agent. Follow common-best-practices and service-specific AGENTS.md rules.
    • Use common-tdd: strict RED first for new behavior; for legacy fixes, characterize only when needed and reproduce the intended change as RED without deleting unrelated implementation.
    • Record the Test Intent Record and run the smallest foreground, single-run target with a project timeout or 120-second fallback before escalating.

Step 3: Local Verification (Enterprise Standard)

Do NOT rely on "it builds" — verify the fix against the issue reproduction steps.

  1. Launch Dev Server: Run the local dev environment for the service.
  2. Execute QE Audit:
    • Web: Load quality-engineering-playwright-cli. Run its preflight, take the first driver rung (CLI, else Playwright MCP). Run the reproduction steps. Capture "After" snapshots.
    • Mobile: Load quality-engineering-appium-mcp. Run its preflight, take the first driver rung (local device, else cloud). Run the reproduction steps.
  3. Final Verdict: Compare results against the issue Expected Result. If any sub-3px regressions exist, fix them now.
    • No success claim without fresh local evidence in docs/srs/srs-walkthrough-[slug].md.

Step 4: Deliver PR

  1. Commit: Generate a commit message using caveman-commit.
  2. PR/MR Details: Draft provider-appropriate PR/MR notes and link the source issue.
  3. Walkthrough:
    • Use the Walkthrough Template below.
    • Create project-local docs/srs/srs-walkthrough-[slug].md with evidence of the local verification.

Runtime Contract

  • Use for bug tickets that need root-cause remediation and a PR/MR.
  • Required inputs: issue URL/key or exported ticket text with reproduce steps.
  • Accepts a test-loop REAL_BUG_DO_NOT_HEAL handoff as the ticket input: the healer's TEST: and EVIDENCE: lines plus the failing AC are the reproduce steps.
  • Return BLOCKED only when repro steps, expected result, or root-cause hypothesis cannot be established.

Handoff Payload

  • slug, operator_profile (carried, not re-inferred), snc_tier, model_tier, implementation plan path, task list path, walkthrough path, PR/MR link, outcome report, next workflow.

Blocking Questions

  • Ask max 3 at a time with a recommended default and 2-3 options.

Artifact Templates

Implementation Plan Template

# Implementation Plan: [Name]

## Goal

## Proposed Changes

## Task Slices

| Slice   | Scope   | Verification   |
| ------- | ------- | -------------- |
| [slice] | [scope] | [verification] |

## Risks

## Verification Plan

## Next Workflow
implementation-readiness

Task Template

# Task: [Name]

## Scope

## Checklist

- [ ] [task]

## Decisions

## Evidence

## Next Workflow
verify-work

Walkthrough Template

# Walkthrough: [Name]

## Scope

## Acceptance Criteria

## Evidence

| Check   | Result              | Evidence   |
| ------- | ------------------- | ---------- |
| [check] | [PASS/FAIL/BLOCKED] | [evidence] |

## Risks

## Outcome Report
{schema_version: 1, run_id: "[run-id]", slug: "[slug]", workflow: dev-fix, feature_status: implemented, started_at: "[timestamp]", completed_at: "[timestamp]", requirement_trace: {brd_objectives: [], requirements: [], acceptance_criteria: [], srs: []}, completed_evidence: [], missing_evidence: [], decision_needed: [], recommended_next_workflow: verify-work, cost: {source: unavailable}, agent: {identity: "[agent-identity]", model: "[model]"}}

## Next Workflow
verify-bug | verify-work

Cost Report

Call get_session_cost(workflow="dev-fix") before final handoff.

Anti-Patterns

  • No Blind Implementation: Never write code before the implementation plan is approved.
  • No Orphan Sessions: Always close browser/appium sessions used during verification.
  • No skipping local verify: "I checked it manually" is not enough. Provide snapshots/logs in the project-local docs/srs/srs-walkthrough-[slug].md.
Skills enregistrées