feat(prometheus): integrate hierarchical plan generation

[rentiansheng]

• Update system-prompt.ts to include PROMETHEUS_PLAN_TEMPLATE_V2
• Prometheus now generates v2 hierarchical plans for complex tasks
• Update plan-generation.ts to conditionally use task_create/task_update when available
• Fall back to TodoWrite when task_system disabled
• Add step-by-step guidance for both tooling paths
• Maintain V2 hierarchical TODO format requirement
• Resolve conflicts with tasks-todowrite-disabler hook expectations

Summary

Changes

Screenshots

Before	After

Testing

bun run typecheck
bun test

Related Issues

Closes #1785

---

Summary by cubic

Adds hierarchical plan generation (V2) to Prometheus with numbered task trees, markdown rendering, and a new task_tree tool. Backward compatible (v1 by default) and addresses #1785.

• New Features

  • Plan Template V2 added to system prompt; plans marked with .
  • plan-generation prefers task_create/task_update (task_list/get) with step-by-step guidance; falls back to TodoWrite when task system is disabled; examples updated for metadata priority and ID capture; includes hierarchical parentID and TodoWrite nested numbering examples.
  • task_tree tool renders a numbered hierarchy with progress; collapses beyond depth 6.
  • Plan format detector (v1 vs v2) and markdown renderer for hierarchical tasks; tree utils with numbering and validation (parent exists, cycles, max depth); broad tests.

• Bug Fixes

  • Fixed tool API examples (priority now in metadata, proper ID capture); documented flat and hierarchical workflows.
  • Fixed tree-numbering issues (deleted root lookup, duplicate sibling indices) with regression tests.
  • Optimized markdown renderer to avoid O(N^2) tree reconstruction.

^{Written for commit 0f7761b. Summary will update on new commits.}

[github-actions]

All contributors have signed the CLA. Thank you! ✅
_{Posted by the CLA Assistant Lite bot.}

[cubic-dev-ai]

5 issues found across 27 files

Confidence score: 3/5

• Several medium-high severity issues in src/agents/prometheus/plan-generation.ts could cause tool misuse: prompt examples reference a non-existent priority arg and promote a flawed task-creation workflow that may yield race conditions or hallucinated IDs.
• src/features/claude-tasks/tree-numbering.ts has concrete numbering bugs (duplicate sibling indexes and deleted roots never numbered), which can affect visible task ordering for users.
• Given multiple user-facing logic issues (severity ~6–7/10), there is some regression risk; this PR feels slightly risky to merge without fixes.
• Pay close attention to src/agents/prometheus/plan-generation.ts and src/features/claude-tasks/tree-numbering.ts - prompt/tool mismatches and numbering logic errors.

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/agents/prometheus/plan-generation.ts">

<violation number="1" location="src/agents/prometheus/plan-generation.ts:41">
P1: Rule violated: **Opencode Compatibility**

The `task_create` prompt examples use `priority` as a top-level parameter, but the actual `task_create` tool (defined in `src/tools/task/task-create.ts`) has no `priority` arg. Priority must be passed inside the `metadata` field: `metadata: { priority: "high" }`. All 8 example calls will either silently drop the priority or fail validation when the AI follows this template.</violation>

<violation number="2" location="src/agents/prometheus/plan-generation.ts:51">
P1: Flawed few-shot example for task creation workflow encourages race condition/hallucinated IDs</violation>
</file>

<file name="src/agents/prometheus/markdown-renderer-v2.ts">

<violation number="1" location="src/agents/prometheus/markdown-renderer-v2.ts:20">
P2: Inefficient O(N²) tree reconstruction - `getNumberingPath` rebuilds the task tree for each task</violation>
</file>

<file name="src/features/claude-tasks/tree-numbering.ts">

<violation number="1" location="src/features/claude-tasks/tree-numbering.ts:86">
P2: Deleted root tasks are never numbered due to flawed lookup logic: `activeRoots` filters out deleted tasks, but the fallback numbering logic for deleted roots searches within `activeRoots`, causing a guaranteed lookup failure (-1) for any deleted root task. This also prevents numbering of their subtrees.</violation>

<violation number="2" location="src/features/claude-tasks/tree-numbering.ts:97">
P1: Incorrect sibling index calculation assigns the same duplicate index to all deleted siblings. The filter().length - 1 always returns the last position for every task, not each task's specific position.</violation>
</file>

---

Since this is your first cubic review, here's how it works:

• cubic automatically reviews your code and comments on bugs and improvements
• Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
• Add one-off context when rerunning by tagging @cubic-dev-ai with guidance or docs links (including llms.txt)
• Ask questions if you need clarification on any suggestion

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[rentiansheng]

• Pay close attention to src/agents/prometheus/plan-generation.ts and src/features/claude-tasks/tree-numbering.ts - prompt/tool mismatches and numbering logic errors.

5 issues found across 27 files

Confidence score: 3/5

• Several medium-high severity issues in src/agents/prometheus/plan-generation.ts could cause tool misuse: prompt examples reference a non-existent priority arg and promote a flawed task-creation workflow that may yield race conditions or hallucinated IDs.
• src/features/claude-tasks/tree-numbering.ts has concrete numbering bugs (duplicate sibling indexes and deleted roots never numbered), which can affect visible task ordering for users.
• Given multiple user-facing logic issues (severity ~6–7/10), there is some regression risk; this PR feels slightly risky to merge without fixes.
• Pay close attention to src/agents/prometheus/plan-generation.ts and src/features/claude-tasks/tree-numbering.ts - prompt/tool mismatches and numbering logic errors.

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/agents/prometheus/plan-generation.ts">

<violation number="1" location="src/agents/prometheus/plan-generation.ts:41">
P1: Rule violated: **Opencode Compatibility**

The `task_create` prompt examples use `priority` as a top-level parameter, but the actual `task_create` tool (defined in `src/tools/task/task-create.ts`) has no `priority` arg. Priority must be passed inside the `metadata` field: `metadata: { priority: "high" }`. All 8 example calls will either silently drop the priority or fail validation when the AI follows this template.</violation>

<violation number="2" location="src/agents/prometheus/plan-generation.ts:51">
P1: Flawed few-shot example for task creation workflow encourages race condition/hallucinated IDs</violation>
</file>

<file name="src/agents/prometheus/markdown-renderer-v2.ts">

<violation number="1" location="src/agents/prometheus/markdown-renderer-v2.ts:20">
P2: Inefficient O(N²) tree reconstruction - `getNumberingPath` rebuilds the task tree for each task</violation>
</file>

<file name="src/features/claude-tasks/tree-numbering.ts">

<violation number="1" location="src/features/claude-tasks/tree-numbering.ts:86">
P2: Deleted root tasks are never numbered due to flawed lookup logic: `activeRoots` filters out deleted tasks, but the fallback numbering logic for deleted roots searches within `activeRoots`, causing a guaranteed lookup failure (-1) for any deleted root task. This also prevents numbering of their subtrees.</violation>

<violation number="2" location="src/features/claude-tasks/tree-numbering.ts:97">
P1: Incorrect sibling index calculation assigns the same duplicate index to all deleted siblings. The filter().length - 1 always returns the last position for every task, not each task's specific position.</violation>
</file>

Since this is your first cubic review, here's how it works:

• cubic automatically reviews your code and comments on bugs and improvements
• Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
• Add one-off context when rerunning by tagging @cubic-dev-ai with guidance or docs links (including llms.txt)
• Ask questions if you need clarification on any suggestion

Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.

Pay close attention to src/agents/prometheus/plan-generation.ts and src/features/claude-tasks/tree-numbering.ts - prompt/tool mismatches and numbering logic errors.
opencode use compatibility mode, there is has diff.

[rentiansheng]

Going forward, I plan to add a todo list to the chat process before "Generate work plan to .sisyphus/plans/user-management.md", allowing discussion of the plan.

[sisyphus-dev-ai]

@rentiansheng Thanks for the PR! The rebase looks good and it is now mergeable.

However, the Cubic AI reviewer identified 5 issues that need attention before merging:

1. task_create examples have incorrect API usage - priority is shown as top-level param but should be inside metadata: { priority: "high" }. The actual tool definition requires this.

2. Task creation workflow has race condition risk - The few-shot example may encourage hallucinated IDs.

3. Performance issue - getNumberingPath rebuilds task tree O(N²) for each task.

4. Tree numbering bug - Deleted root tasks are never numbered (lookup in activeRoots fails).

5. Duplicate sibling indexes - Deleted siblings all get same index due to filter().length - 1 logic.

Also the CLA check is currently failing - please make sure to sign it.

Could you address these issues? Happy to re-review once fixed.

[rentiansheng]

I have read the CLA Document and I hereby sign the CLA

[rentiansheng]

recheck