confidence-scoring（置信度评分）

Confidence Scoring for PRPs and Work-Orders

This skill provides systematic evaluation of PRPs (Product Requirement Prompts) and work-orders to determine their readiness for execution or delegation.

When to Use This Skill

Activate this skill when:

• Creating a new PRP (/prp:create)
• Generating a work-order (/blueprint:work-order)
• Deciding whether to execute or refine a PRP
• Evaluating whether a task is ready for subagent delegation
• Reviewing PRPs/work-orders for quality

Scoring Dimensions

1. Context Completeness (1-10)

Evaluates whether all necessary context is explicitly provided.

Checklist:

• File paths are absolute or clearly relative to project root
• Code snippets include actual line numbers (e.g., src/auth.py:45-60)
• Library versions are specified
• Integration points are documented
• Patterns from codebase are shown with examples

2. Implementation Clarity (1-10)

Evaluates how clear the implementation approach is.

Checklist:

• Task breakdown is explicit
• Pseudocode is provided for complex logic
• Implementation order is specified
• Edge cases are identified
• Error handling approach is documented

3. Gotchas Documented (1-10)

Evaluates whether known pitfalls are documented with mitigations.

Checklist:

• Library-specific gotchas documented
• Version-specific behaviors noted
• Common mistakes identified
• Each gotcha has a mitigation
• Race conditions/concurrency issues addressed

4. Validation Coverage (1-10)

Evaluates whether executable validation commands are provided.

Checklist:

• Linting command provided and executable
• Type checking command provided (if applicable)
• Unit test command with specific test files
• Integration test command (if applicable)
• Coverage check command with threshold
• Security scan command (if applicable)
• All commands include expected outcomes

5. Test Coverage (1-10) - Work-Orders Only

Evaluates whether test cases are specified.

Checklist:

• Each test case has code template
• Assertions are explicit
• Happy path tested
• Error cases tested
• Edge cases tested

Calculating Overall Score

For PRPs

Overall = (Context + Implementation + Gotchas + Validation) / 4

For Work-Orders

Overall = (Context + Gotchas + TestCoverage + Validation) / 4

Score Thresholds

Response Templates

High Confidence (7+)

## Confidence Score: X.X/10

| Dimension | Score | Notes |
|-----------|-------|-------|
| Context Completeness | X/10 | [specific observation] |
| Implementation Clarity | X/10 | [specific observation] |
| Gotchas Documented | X/10 | [specific observation] |
| Validation Coverage | X/10 | [specific observation] |
| **Overall** | **X.X/10** | |

**Assessment:** Ready for execution

**Strengths:**
- [Key strength 1]
- [Key strength 2]

**Recommendations (optional):**
- [Minor improvement 1]

Low Confidence (<7)

## Confidence Score: X.X/10

| Dimension | Score | Notes |
|-----------|-------|-------|
| Context Completeness | X/10 | [specific gap] |
| Implementation Clarity | X/10 | [specific gap] |
| Gotchas Documented | X/10 | [specific gap] |
| Validation Coverage | X/10 | [specific gap] |
| **Overall** | **X.X/10** | |

**Assessment:** Needs refinement before execution

**Gaps to Address:**
- [ ] [Gap 1 with suggested action]
- [ ] [Gap 2 with suggested action]
- [ ] [Gap 3 with suggested action]

**Next Steps:**
1. [Specific research action]
2. [Specific documentation action]
3. [Specific validation action]

Examples

Example 1: Well-Prepared PRP

## Confidence Score: 8.5/10

| Dimension | Score | Notes |
|-----------|-------|-------|
| Context Completeness | 9/10 | All files explicit, code snippets with line refs |
| Implementation Clarity | 8/10 | Pseudocode covers main path, one edge case unclear |
| Gotchas Documented | 8/10 | Redis connection pool, JWT format issues covered |
| Validation Coverage | 9/10 | All gates have commands, outcomes specified |
| **Overall** | **8.5/10** | |

**Assessment:** Ready for execution

**Strengths:**
- Comprehensive codebase intelligence with actual code snippets
- Validation gates are copy-pasteable
- Known library gotchas well-documented

**Recommendations:**
- Consider documenting concurrent token refresh edge case

Example 2: Needs Work

## Confidence Score: 5.0/10

| Dimension | Score | Notes |
|-----------|-------|-------|
| Context Completeness | 4/10 | File paths vague ("somewhere in auth/") |
| Implementation Clarity | 6/10 | High-level approach clear, no pseudocode |
| Gotchas Documented | 3/10 | No library-specific gotchas |
| Validation Coverage | 7/10 | Test command present, missing lint/type check |
| **Overall** | **5.0/10** | |

**Assessment:** Needs refinement before execution

**Gaps to Address:**
- [ ] Add explicit file paths (use `grep` to find them)
- [ ] Add pseudocode for token generation logic
- [ ] Research jsonwebtoken gotchas (check GitHub issues)
- [ ] Add linting and type checking commands

**Next Steps:**
1. Run `/prp:curate-docs jsonwebtoken` to create ai_docs entry
2. Use Explore agent to find exact file locations
3. Add validation gate commands from project's package.json

Integration with Blueprint Development

This skill is automatically applied when:

• /prp:create generates a new PRP
• /blueprint:work-order generates a work-order
• Reviewing existing PRPs for execution readiness

The confidence score determines:

• 9+: Proceed with subagent delegation
• 7-8: Proceed with direct execution
• < 7: Refine before execution

Tips for Improving Scores

Context Completeness

• Use grep to find exact file locations
• Include actual line numbers in code snippets
• Reference ai_docs entries for library patterns

Implementation Clarity

• Write pseudocode before describing approach
• Enumerate edge cases explicitly
• Define error handling strategy

Gotchas Documented

• Search GitHub issues for library gotchas
• Check Stack Overflow for common problems
• Document team experience from past projects

Validation Coverage

• Copy commands from project's config (package.json, pyproject.toml)
• Include specific file paths in test commands
• Specify expected outcomes for each gate