fix(atlas): gate final-wave approval on real plan state

Ignore nested plan checkboxes and track parallel final-wave approvals so Atlas only pauses for user approval when the real top-level review wave is complete.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

src/agents/atlas/atlas-prompt.test.ts

@@ -130,4 +130,17 @@ describe("Atlas prompts plan path consistency", () => {
       expect(prompt).toMatch(/read[\s\S]*?\.sisyphus\/plans\//)
     }
   })
+
+  test("all variants should distinguish top-level plan tasks from nested checkboxes", () => {
+    // given
+    const prompts = [ATLAS_SYSTEM_PROMPT, ATLAS_GPT_SYSTEM_PROMPT, ATLAS_GEMINI_SYSTEM_PROMPT]
+
+    // when / then
+    for (const prompt of prompts) {
+      const lowerPrompt = prompt.toLowerCase()
+      expect(lowerPrompt).toMatch(/top-level.*checkbox/)
+      expect(lowerPrompt).toMatch(/ignore nested.*checkbox/)
+      expect(lowerPrompt).toMatch(/final verification wave/)
+    }
+  })
 })

src/agents/atlas/default.ts

@@ -140,7 +140,8 @@ TodoWrite([
 ## Step 1: Analyze Plan
 
 1. Read the todo list file
-2. Parse incomplete checkboxes \`- [ ]\`
+2. Parse actionable **top-level** task checkboxes in \`## TODOs\` and \`## Final Verification Wave\`
+   - Ignore nested checkboxes under Acceptance Criteria, Evidence, Definition of Done, and Final Checklist sections.
 3. Extract parallelizability info from each task
 4. Build parallelization map:
    - Which tasks can run simultaneously?
@@ -242,7 +243,7 @@ After verification, READ the plan file directly — every time, no exceptions:
 \`\`\`
 Read(".sisyphus/plans/{plan-name}.md")
 \`\`\`
-Count remaining \`- [ ]\` tasks. This is your ground truth for what comes next.
+Count remaining **top-level task** checkboxes. Ignore nested verification/evidence checkboxes. This is your ground truth for what comes next.
 
 **Checklist (ALL must be checked):**
 \`\`\`
@@ -296,6 +297,7 @@ Repeat Step 3 until all implementation tasks complete. Then proceed to Step 4.
 
 The plan's Final Wave tasks (F1-F4) are APPROVAL GATES — not regular tasks.
 Each reviewer produces a VERDICT: APPROVE or REJECT.
+Final-wave reviewers can finish in parallel before you update the plan file, so do NOT rely on raw unchecked-count alone.
 
 1. Execute all Final Wave tasks in parallel
 2. If ANY verdict is REJECT:

src/agents/atlas/gemini.ts

@@ -157,7 +157,8 @@ TodoWrite([
 ## Step 1: Analyze Plan
 
 1. Read the todo list file
-2. Parse incomplete checkboxes \`- [ ]\`
+2. Parse actionable **top-level** task checkboxes in \`## TODOs\` and \`## Final Verification Wave\`
+   - Ignore nested checkboxes under Acceptance Criteria, Evidence, Definition of Done, and Final Checklist sections.
 3. Build parallelization map
 
 Output format:
@@ -263,7 +264,7 @@ ALL three must be YES. "Probably" = NO. "I think so" = NO.
 \`\`\`
 Read(".sisyphus/plans/{plan-name}.md")
 \`\`\`
-Count remaining \`- [ ]\` tasks.
+Count remaining **top-level task** checkboxes. Ignore nested verification/evidence checkboxes.
 
 ### 3.5 Handle Failures
 
@@ -284,6 +285,7 @@ Repeat Step 3 until all implementation tasks complete. Then proceed to Step 4.
 
 The plan's Final Wave tasks (F1-F4) are APPROVAL GATES — not regular tasks.
 Each reviewer produces a VERDICT: APPROVE or REJECT.
+Final-wave reviewers can finish in parallel before you update the plan file, so do NOT rely on raw unchecked-count alone.
 
 1. Execute all Final Wave tasks in parallel
 2. If ANY verdict is REJECT:

src/agents/atlas/gpt.ts

@@ -167,7 +167,8 @@ TodoWrite([
 ## Step 1: Analyze Plan
 
 1. Read the todo list file
-2. Parse incomplete checkboxes \`- [ ]\`
+2. Parse actionable **top-level** task checkboxes in \`## TODOs\` and \`## Final Verification Wave\`
+   - Ignore nested checkboxes under Acceptance Criteria, Evidence, Definition of Done, and Final Checklist sections.
 3. Build parallelization map
 
 Output format:
@@ -268,7 +269,7 @@ Before moving to the next task, answer these THREE questions honestly:
 \`\`\`
 Read(".sisyphus/plans/{plan-name}.md")
 \`\`\`
-Count remaining \`- [ ]\` tasks. This is your ground truth.
+Count remaining **top-level task** checkboxes. Ignore nested verification/evidence checkboxes. This is your ground truth.
 
 ### 3.5 Handle Failures
 
@@ -289,6 +290,7 @@ Repeat Step 3 until all implementation tasks complete. Then proceed to Step 4.
 
 The plan's Final Wave tasks (F1-F4) are APPROVAL GATES — not regular tasks.
 Each reviewer produces a VERDICT: APPROVE or REJECT.
+Final-wave reviewers can finish in parallel before you update the plan file, so do NOT rely on raw unchecked-count alone.
 
 1. Execute all Final Wave tasks in parallel
 2. If ANY verdict is REJECT: