Eval Harness Skill

Eval Harness Skill

A formal evaluation framework for Claude Code sessions, implementing eval-driven development (EDD) principles.

Philosophy

Eval-Driven Development treats evals as the "unit tests of AI development":

• Define expected behavior BEFORE implementation
• Run evals continuously during development
• Track regressions with each change
• Use pass@k metrics for reliability measurement

Sudocode alignment: EDD's "define before implement" maps to sudocode's "spec before issue" philosophy. Evals can be part of specs, and eval results become feedback.

Eval Types

Capability Evals

Test if Claude can do something it couldn't before:

[CAPABILITY EVAL: feature-name]
Task: Description of what Claude should accomplish
Success Criteria:
  - [ ] Criterion 1
  - [ ] Criterion 2
  - [ ] Criterion 3
Expected Output: Description of expected result

Regression Evals

Ensure changes don't break existing functionality:

[REGRESSION EVAL: feature-name]
Baseline: SHA or checkpoint name
Tests:
  - existing-test-1: PASS/FAIL
  - existing-test-2: PASS/FAIL
  - existing-test-3: PASS/FAIL
Result: X/Y passed (previously Y/Y)

Grader Types

1. Code-Based Grader

Deterministic checks using code:

# Check if file contains expected pattern
grep -q "export function handleAuth" src/auth.ts && echo "PASS" || echo "FAIL"

# Check if tests pass
npm test -- --testPathPattern="auth" && echo "PASS" || echo "FAIL"

# Check if build succeeds
npm run build && echo "PASS" || echo "FAIL"

2. Model-Based Grader

Use Claude to evaluate open-ended outputs:

[MODEL GRADER PROMPT]
Evaluate the following code change:
1. Does it solve the stated problem?
2. Is it well-structured?
3. Are edge cases handled?
4. Is error handling appropriate?

Score: 1-5 (1=poor, 5=excellent)
Reasoning: [explanation]

3. Human Grader

Flag for manual review:

[HUMAN REVIEW REQUIRED]
Change: Description of what changed
Reason: Why human review is needed
Risk Level: LOW/MEDIUM/HIGH

Metrics

pass@k

"At least one success in k attempts"

• pass@1: First attempt success rate
• pass@3: Success within 3 attempts
• Typical target: pass@3 > 90%

pass^k

"All k trials succeed"

• Higher bar for reliability
• pass^3: 3 consecutive successes
• Use for critical paths

Eval Workflow

1. Define (Before Coding)

## EVAL DEFINITION: feature-xyz

### Capability Evals
1. Can create new user account
2. Can validate email format
3. Can hash password securely

### Regression Evals
1. Existing login still works
2. Session management unchanged
3. Logout flow intact

### Success Metrics
- pass@3 > 90% for capability evals
- pass^3 = 100% for regression evals

2. Implement

Write code to pass the defined evals.

3. Evaluate

# Run capability evals
[Run each capability eval, record PASS/FAIL]

# Run regression evals
npm test -- --testPathPattern="existing"

# Generate report

4. Report

EVAL REPORT: feature-xyz
========================

Capability Evals:
  create-user:     PASS (pass@1)
  validate-email:  PASS (pass@2)
  hash-password:   PASS (pass@1)
  Overall:         3/3 passed

Regression Evals:
  login-flow:      PASS
  session-mgmt:    PASS
  logout-flow:     PASS
  Overall:         3/3 passed

Metrics:
  pass@1: 67% (2/3)
  pass@3: 100% (3/3)

Status: READY FOR REVIEW

Integration Patterns

Pre-Implementation

/eval define feature-name

Creates eval definition file at .claude/evals/feature-name.md

During Implementation

/eval check feature-name

Runs current evals and reports status

Post-Implementation

/eval report feature-name

Generates full eval report

Eval Storage

Store evals in project:

.claude/
  evals/
    feature-xyz.md      # Eval definition
    feature-xyz.log     # Eval run history
    baseline.json       # Regression baselines

Best Practices

1. Define evals BEFORE coding - Forces clear thinking about success criteria
2. Run evals frequently - Catch regressions early
3. Track pass@k over time - Monitor reliability trends
4. Use code graders when possible - Deterministic > probabilistic
5. Human review for security - Never fully automate security checks
6. Keep evals fast - Slow evals don't get run
7. Version evals with code - Evals are first-class artifacts

Example: Adding Authentication

## EVAL: add-authentication

### Phase 1: Define (10 min)
Capability Evals:
- [ ] User can register with email/password
- [ ] User can login with valid credentials
- [ ] Invalid credentials rejected with proper error
- [ ] Sessions persist across page reloads
- [ ] Logout clears session

Regression Evals:
- [ ] Public routes still accessible
- [ ] API responses unchanged
- [ ] Database schema compatible

### Phase 2: Implement (varies)
[Write code]

### Phase 3: Evaluate
Run: /eval check add-authentication

### Phase 4: Report
EVAL REPORT: add-authentication
==============================
Capability: 5/5 passed (pass@3: 100%)
Regression: 3/3 passed (pass^3: 100%)
Status: SHIP IT

---

Sudocode Integration

Evals in Specs

Include eval definitions directly in specs as success criteria:

# Feature Spec: Authentication

## Success Criteria (Evals)

### Capability Evals
- [ ] User can register with email/password
- [ ] User can login with valid credentials
- [ ] Invalid credentials rejected with proper error

### Regression Evals
- [ ] Public routes still accessible
- [ ] API responses unchanged

### Metrics
- pass@3 > 90% for capability evals
- pass^3 = 100% for regression evals

Eval Results as Feedback

When closing an issue, include eval results in feedback:

add_feedback to_id=s-auth type=comment content="
## Eval Report

Capability Evals:
  register:        PASS (pass@1)
  login:           PASS (pass@2)
  error-handling:  PASS (pass@1)
  Overall:         3/3 passed

Regression Evals:
  public-routes:   PASS
  api-unchanged:   PASS
  Overall:         2/2 passed

Metrics:
  pass@1: 67% (2/3)
  pass@3: 100% (3/3)

Status: All evals passing, ready to close.
"

Workflow: Eval-Driven Sudocode

1. Create spec with eval criteria
2. Create issue implementing spec
3. Define evals (/eval define)
4. Implement to pass evals
5. Run evals (/eval check)
6. Include eval report in feedback
7. Close issue

Evals → Issue Acceptance Criteria

When planning issues from specs, convert eval criteria to acceptance criteria:

Spec eval: "User can register with email/password"
↓
Issue acceptance:
- POST /api/register accepts email/password
- Returns user object on success
- Returns validation errors on failure
- Sends confirmation email

Failed Evals → Discovered Issues

If evals fail during implementation:

1. Document failure in issue
2. Create new issue for fix: discovered-from current issue
3. Add blocks relationship if blocking
4. Continue with other evals or fix blocker