Discovery

Discovery Skill

Find ALL relevant code before planning. No blind implementation.

When to Invoke

Planning agent invokes this skill at the start of any Standard/Complex task.

Process

1. Extract Search Terms

From task input, identify:

• Technical keywords (OAuth, React, API, etc.)
• Error messages or stack traces
• Feature names or component references

2. Search for Relevant Code

# Find files by keyword
grep -rl "keyword" --include="*.{py,ts,js,go}" src/

# Find by function/class name
grep -rn "def function_name\|class ClassName" .

# Find related tests
find . -name "*test*" -type f | xargs grep -l "keyword"

3. Analyze Dependencies

• Check imports in relevant files
• Identify API calls between services
• Map data flow

4. Output Format

{
  "relevant_files": [
    {"path": "src/auth/login.py", "reason": "Contains login logic", "relevance": 0.95}
  ],
  "dependencies": ["src/models/user.py", "src/utils/hash.py"],
  "test_files": ["tests/test_login.py"],
  "complexity": "medium",
  "estimated_scope": "3-5 files"
}

Quality Criteria

• Return top 10 most relevant files
• Each file has clear relevance reason
• Include both source and test files
• Identify cross-file dependencies

Hand-off to Planning

After discovery, planning agent uses output to:

1. Define accurate scope in PLAN.md
2. List affected files per sub-task
3. Identify verification commands