corvus-extras - SKILL.md Agent Skill

name: corvus-extras description: Extra utilities - subagent reference, todo patterns, error handling

SUBAGENT REFERENCE

Phase	Subagent	Purpose	Approval Needed
0a	requirements-analyst	Initial request analysis	No
0b	requirements-analyst	Post-discovery analysis	No
1a	researcher	External docs, best practices	No
1b	code-explorer	Codebase analysis	No
2	task-planner	Create master plan + task files	No (creates plan)
3.5	plan-reviewer	High accuracy plan review (if user chose)	No (optional)
3.5 fail	task-planner	Fix plan based on review feedback	No
4a	code-implementer	Implementation + lint/typecheck/build	No (delegated mode)
4b	code-quality	Run tests, verify acceptance criteria	No
4b fail	task-planner	FAILURE_ANALYSIS (learning mode)	No
4c	(Corvus)	Update plan directly	N/A (no subagent)
5a	code-quality	Final comprehensive test run	No
5b	ux-dx-quality	Subjective validation (if required)	No (if required)
6a	task-planner	SUCCESS_EXTRACTION (once, entire feature)	No

Invoking Subagents

Use the Task tool with subagent_type parameter:

task(
  subagent_type: "code-explorer",
  description: "Analyze auth module",
  prompt: "**TASK**: Analyze codebase..."
)

Parallel Invocation

When subagents are independent (Phase 1), invoke them in the same message:

// These run in parallel
task(subagent_type: "researcher", description: "Research JWT", prompt: "...")
task(subagent_type: "code-explorer", description: "Explore auth", prompt: "...")

TODO TRACKING

Use TodoWrite to track progress at the phase level:

Phase 0a/0b

todowrite([
  { id: "clarify-initial", content: "Initial requirements analysis", status: "in_progress", priority: "high" }
])

After requirements-analyst returns:

// If QUESTIONS_NEEDED
todowrite([
  { id: "clarify-initial", content: "Initial requirements analysis - awaiting user answers", status: "pending", priority: "high" }
])

// If REQUIREMENTS_CLEAR (skip to Phase 2)
todowrite([
  { id: "clarify-initial", content: "Initial requirements analysis", status: "completed", priority: "high" },
  { id: "planning", content: "Create master plan", status: "in_progress", priority: "high" }
])

// If DISCOVERY_NEEDED
todowrite([
  { id: "clarify-initial", content: "Initial requirements analysis", status: "completed", priority: "high" },
  { id: "discovery", content: "Discovery phase", status: "in_progress", priority: "high" }
])

Phase 1

todowrite([
  { id: "research", content: "Research [topic]", status: "in_progress", priority: "high" },
  { id: "explore", content: "Explore codebase", status: "in_progress", priority: "high" },
  { id: "plan", content: "Create master plan", status: "pending", priority: "high" }
])

Phase 4 (Phase-Level Tracking)

During Phase 4, track at PHASE level, not step level:

Starting a New Phase:

todowrite([
  // Previous completed phases...
  { id: "phase-1", content: "Phase 1: Foundation (Tasks 01-02)", status: "completed", priority: "high" },
  { id: "phase-2", content: "Phase 2: Core Implementation (Tasks 03-07)", status: "in_progress", priority: "high" },
  { id: "phase-3", content: "Phase 3: Integration (Tasks 08-10)", status: "pending", priority: "high" },
])

After Phase Completes:

todowrite([
  { id: "phase-1", content: "Phase 1: Foundation (Tasks 01-02)", status: "completed", priority: "high" },
  { id: "phase-2", content: "Phase 2: Core Implementation (Tasks 03-07)", status: "completed", priority: "high" },
  { id: "phase-3", content: "Phase 3: Integration (Tasks 08-10)", status: "in_progress", priority: "high" },
])

Phase 5 (Final Validation)

todowrite([
  // All implementation phases completed...
  { id: "phase-5a", content: "Phase 5a: Final objective validation", status: "in_progress", priority: "high" },
  { id: "phase-5b", content: "Phase 5b: Final UX/DX review", status: "pending", priority: "high" },
])

Why Phase-Level Tracking:

Reduces todo noise
Matches the per-phase execution model
User sees meaningful progress (phases, not steps)
Prevents "what phase am I on?" errors

ERROR HANDLING

Implementation Errors

When code-implementer reports an error:

Analyze - Understand the error and its cause
Propose fix - Determine the correction needed
Send fix request - Instruct code-implementer to fix
Re-validate - Run code-quality again
Iterate - Max 3 attempts, then escalate to user

Validation Failures

When code-quality reports failures:

Categorize - Is it a test failure, build error, or criteria miss?
Create specific fix - Target the exact issue
Track iterations - Don't loop infinitely
Escalate if stuck - After 3 attempts, ask user

Fundamental Issues

If the entire approach is wrong:

Stop implementation immediately
Report to user with clear explanation
Propose alternatives (2-3 options)
Wait for guidance before continuing

## Approach Issue Detected

**Problem**: [Clear description of the fundamental issue]

**Why This Matters**: [Impact if we continue]

**Options**:
1. [Alternative approach 1] - [tradeoffs]
2. [Alternative approach 2] - [tradeoffs]  
3. [Abort and start fresh]

**My Recommendation**: [Which option and why]

**Awaiting your guidance.**