---

name: dod-validation-cycle description: HAIOS DoD Validation Cycle for validating Definition of Done before closure. Use before DONE phase. Guides CHECK->VALIDATE->APPROVE workflow. generated: '2025-12-25' last_updated: '2025-12-28T21:12:55'

DoD Validation Cycle (Bridge Skill)

This is a Validation Skill (bridge) that validates Definition of Done criteria before work item closure. It acts as a quality gate at the entry to close-work-cycle.

When to Use

Invoked automatically by close-work-cycle before VALIDATE phase. Manual invocation: Skill(skill="dod-validation-cycle") before closing a work item.

---

The Cycle

CopyCHECK --> VALIDATE --> APPROVE

1. CHECK Phase

Goal: Verify DoD prerequisites exist.

Prerequisites to Check:

• Work item has associated plan(s)
• Tests were run (test results available)
• Memory refs captured (WHY documented)
• Docs reviewed (CLAUDE.md, READMEs)

Actions:

1. Read work file to find associated documents
2. Check plan status(es)
3. Verify memory_refs field is populated
4. Report missing prerequisites

Exit Criteria:

• Associated plans found
• Test results mentioned or N/A documented
• memory_refs populated or WHY captured

Tools: Read, Glob, Grep

---

2. VALIDATE Phase

Goal: Validate each DoD criterion per ADR-033.

DoD Criteria (ADR-033):

Actions:

1. For each criterion, verify status
2. Flag any failing criteria
3. Report validation status

Ground Truth Verification (Plan-Specific):

If associated plans exist (from work item's documents.plans field):

1. For each plan:

   • Read plan file from docs/work/active/{id}/plans/PLAN.md (or legacy path)
   • Parse Ground Truth Verification table (looking for ## Ground Truth Verification section)
   • Use parse_ground_truth_table() from .claude/lib/validate.py

2. For each verification item:

   • Classify type based on file_path pattern (via classify_verification_type())

   • Execute machine-checkable items:

     Type	Tool	Success Criteria
     file-check	Read(path)	File exists AND expected_state pattern found
     grep-check	Grep(pattern)	Match count matches expectation (0 or >0)
     test-run	Bash(pytest)	Exit code 0
     json-verify	Read + JSON parse	File exists AND expected fields present
     human-judgment	-	Flag for manual confirmation

   • Flag human-judgment items for manual confirmation

3. Report results:

   • "X of Y machine checks passed"
   • "Z items require human confirmation" (list them)

4. Gate decision:

   • BLOCK if any machine-check FAILS (list specific failures)
   • WARN if unchecked items remain (items with [ ] checkbox)
   • PASS if all machine-checks pass and human-judgment items confirmed

Exit Criteria:

• Tests MUST pass (or N/A with rationale)
• WHY MUST be captured
• Docs MUST be current
• Associated plans MUST be complete
• Ground Truth Verification: all machine-checks MUST pass (or N/A)
• Ground Truth Verification: human-judgment items MUST be confirmed

Tools: Read, Grep, Bash

---

3. APPROVE Phase

Goal: Confirm work item is ready for closure.

Actions:

1. If all criteria pass -> APPROVED
2. If any fail -> Report blockers
3. Return to calling cycle immediately

On PASS: Return to close-work-cycle - it will proceed to VALIDATE phase. On FAIL: Return to implementation for fixes.

MUST: Do not pause for acknowledgment - return to calling cycle immediately.

Exit Criteria:

• All DoD criteria validated
• Approval status reported
• Returned to calling cycle (no pause)

Tools: -

---

Composition Map

---

Quick Reference

---

Key Design Decisions

---

Related

• plan-validation-cycle skill: Pre-DO validation (parallel pattern)
• design-review-validation skill: During-DO validation (parallel pattern)
• close-work-cycle skill: Invokes this skill as MUST gate
• ADR-033: Work Item Lifecycle Governance (DoD definition)
• INV-042: Machine-Checked DoD Gates investigation (design source)
• E2-219: Ground Truth Verification Parser (parse_ground_truth_table(), classify_verification_type())
• validate.py: .claude/lib/validate.py contains parser implementation