feat(atlas): enforce mandatory manual code review and direct boulder state checks

- VERIFICATION_REMINDER: add Step 2 manual code review (non-negotiable)
  - Require Read of EVERY changed file line by line
  - Cross-check subagent claims vs actual code
  - Verify logic correctness, completeness, edge cases, patterns
- Add Step 5: direct boulder state check via Read plan file
  - Count remaining tasks directly, no cached state
- BOULDER_CONTINUATION_PROMPT: add first rule to read plan file immediately
- verification-reminders.ts: restructure steps 5-8 for boulder/todo checks
- Atlas default.ts (Claude): enhance 3.4 QA with A/B/C/D sections
  - A: Automated verification
  - B: Manual code review (non-negotiable)
  - C: Hands-on QA (if applicable)
  - D: Check boulder state directly
- Atlas gpt.ts (GPT-5.2): apply same QA enhancements with GPT-optimized structure
- verification_rules: update both Claude and GPT versions with manual review requirements

Addresses issue where Atlas would skip manual code inspection after delegation,
leading to rubber-stamping of broken or incomplete work.

src/hooks/atlas/system-reminder-templates.ts

@@ -33,6 +33,7 @@ export const BOULDER_CONTINUATION_PROMPT = `${createSystemDirective(SystemDirect
 You have an active work plan with incomplete tasks. Continue working.
 
 RULES:
+- **FIRST**: Read the plan file NOW to check exact current progress — count remaining \`- [ ]\` tasks
 - Proceed without asking for permission
 - Change \`- [ ]\` to \`- [x]\` in the plan file when done
 - Use the notepad at .sisyphus/notepads/{PLAN_NAME}/ to record learnings
@@ -48,15 +49,36 @@ Tests FAILING, code has ERRORS, implementation INCOMPLETE - but they say "done".
 
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
-**STEP 1: VERIFY WITH YOUR OWN TOOL CALLS (DO THIS NOW)**
+**STEP 1: AUTOMATED VERIFICATION (DO THIS FIRST)**
 
 Run these commands YOURSELF - do NOT trust agent's claims:
 1. \`lsp_diagnostics\` on changed files → Must be CLEAN
 2. \`bash\` to run tests → Must PASS
 3. \`bash\` to run build/typecheck → Must succeed
-4. \`Read\` the actual code → Must match requirements
 
-**STEP 2: DETERMINE IF HANDS-ON QA IS NEEDED**
+**STEP 2: MANUAL CODE REVIEW (NON-NEGOTIABLE — DO NOT SKIP)**
+
+Automated checks are NECESSARY but INSUFFICIENT. You MUST read the actual code.
+
+**RIGHT NOW — \`Read\` EVERY file the subagent touched. No exceptions.**
+
+For EACH changed file, verify:
+1. Does the implementation logic ACTUALLY match the task requirements?
+2. Are there incomplete stubs (TODO comments, placeholder code, hardcoded values)?
+3. Are there logic errors, off-by-one bugs, or missing edge cases?
+4. Does it follow existing codebase patterns and conventions?
+5. Are imports correct? No unused or missing imports?
+6. Is error handling present where needed?
+
+**Cross-check the subagent's claims against reality:**
+- Subagent said "Updated X" → READ X. Is it actually updated?
+- Subagent said "Added tests" → READ tests. Do they test the RIGHT behavior?
+- Subagent said "Follows patterns" → COMPARE with reference. Does it actually?
+
+**If you cannot explain what the changed code does, you have not reviewed it.**
+**If you skip this step, you are rubber-stamping broken work.**
+
+**STEP 3: DETERMINE IF HANDS-ON QA IS NEEDED**
 
 | Deliverable Type | QA Method | Tool |
 |------------------|-----------|------|
@@ -66,7 +88,7 @@ Run these commands YOURSELF - do NOT trust agent's claims:
 
 Static analysis CANNOT catch: visual bugs, animation issues, user flow breakages.
 
-**STEP 3: IF QA IS NEEDED - ADD TO TODO IMMEDIATELY**
+**STEP 4: IF QA IS NEEDED - ADD TO TODO IMMEDIATELY**
 
 \`\`\`
 todowrite([
@@ -76,7 +98,8 @@ todowrite([
 
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
-**BLOCKING: DO NOT proceed to Step 4 until Steps 1-3 are VERIFIED.**`
+**BLOCKING: DO NOT proceed until Steps 1-4 are ALL completed.**
+**Skipping Step 2 (manual code review) = unverified work = FAILURE.**`
 
 export const ORCHESTRATOR_DELEGATION_REQUIRED = `
 

src/hooks/atlas/verification-reminders.ts

@@ -26,7 +26,16 @@ export function buildOrchestratorReminder(
 
 ${buildVerificationReminder(sessionId)}
 
-**STEP 4: MARK COMPLETION IN PLAN FILE (IMMEDIATELY)**
+**STEP 5: CHECK BOULDER STATE DIRECTLY (EVERY TIME — NO EXCEPTIONS)**
+
+Do NOT rely on cached progress. Read the plan file NOW:
+\`\`\`
+Read(".sisyphus/tasks/${planName}.yaml")
+\`\`\`
+Count exactly: how many \`- [ ]\` remain? How many \`- [x]\` completed?
+This is YOUR ground truth. Use it to decide what comes next.
+
+**STEP 6: MARK COMPLETION IN PLAN FILE (IMMEDIATELY)**
 
 RIGHT NOW - Do not delay. Verification passed → Mark IMMEDIATELY.
 
@@ -36,14 +45,14 @@ Update the plan file \`.sisyphus/tasks/${planName}.yaml\`:
 
 **DO THIS BEFORE ANYTHING ELSE. Unmarked = Untracked = Lost progress.**
 
-**STEP 5: COMMIT ATOMIC UNIT**
+**STEP 7: COMMIT ATOMIC UNIT**
 
 - Stage ONLY the verified changes
 - Commit with clear message describing what was done
 
-**STEP 6: PROCEED TO NEXT TASK**
+**STEP 8: PROCEED TO NEXT TASK**
 
-- Read the plan file to identify the next \`- [ ]\` task
+- Read the plan file AGAIN to identify the next \`- [ ]\` task
 - Start immediately - DO NOT STOP
 
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
@@ -57,7 +66,12 @@ export function buildStandaloneVerificationReminder(sessionId: string): string {
 
 ${buildVerificationReminder(sessionId)}
 
-**STEP 4: UPDATE TODO STATUS (IMMEDIATELY)**
+**STEP 5: CHECK YOUR PROGRESS DIRECTLY (EVERY TIME — NO EXCEPTIONS)**
+
+Do NOT rely on memory or cached state. Run \`todoread\` NOW to see exact current state.
+Count pending vs completed tasks. This is your ground truth for what comes next.
+
+**STEP 6: UPDATE TODO STATUS (IMMEDIATELY)**
 
 RIGHT NOW - Do not delay. Verification passed → Mark IMMEDIATELY.
 
@@ -66,15 +80,15 @@ RIGHT NOW - Do not delay. Verification passed → Mark IMMEDIATELY.
 
 **DO THIS BEFORE ANYTHING ELSE. Unmarked = Untracked = Lost progress.**
 
-**STEP 5: EXECUTE QA TASKS (IF ANY)**
+**STEP 7: EXECUTE QA TASKS (IF ANY)**
 
 If QA tasks exist in your todo list:
 - Execute them BEFORE proceeding
 - Mark each QA task complete after successful verification
 
-**STEP 6: PROCEED TO NEXT PENDING TASK**
+**STEP 8: PROCEED TO NEXT PENDING TASK**
 
-- Identify the next \`pending\` task from your todo list
+- Run \`todoread\` AGAIN to identify the next \`pending\` task
 - Start immediately - DO NOT STOP
 
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━