Add HITL planner agent skill and update API skills with spike findings

[nickmisasi]

Summary

• New /launch-planner-agent skill documenting the HITL plan loop pattern, based on a live spike test against the Cursor Cloud Agent API
• Updated /debug-cursor-api with conversation retrieval and planner agent curl examples
• Updated /modify-agent-prompt with the planner prompt pipeline section

Context

A spike test was run launching a read-only planner agent against mattermost/mattermost for MM-66620. Key validated findings:

• Conversation API returns clean JSON with plan in the last assistant_message
• Strong "DO NOT MODIFY CODE" prompting works — agent stayed read-only
• autoBranch defaults to true and creates orphan branches — must explicitly set false
• Planner took ~11 min on the monorepo, produced a ~3.5KB plan

These findings are codified in the new skill so future implementation work has validated patterns to follow.

Test plan

• Skills are markdown files — no build/test impact
• Verified skill descriptions render correctly in Claude Code skill listing

🤖 Generated with Claude Code

Summary by CodeRabbit

• Documentation
  • Added API reference documentation for retrieving agent conversations and launching a read-only planner agent.
  • Documented the planner agent workflow, including API interaction patterns, plan extraction, and iteration behavior.
  • Added guidance on configuring planner settings and integrating with HITL planning workflows.

[coderabbitai]

Caution

Review failed

The pull request is closed.

📝 Walkthrough

Walkthrough

This PR adds comprehensive documentation for a new read-only Planner Agent feature within the HITL workflow. It includes API reference examples for agent conversation retrieval and planner agent launch, detailed feature documentation covering workflow patterns and plan extraction, and prompt configuration guidance for the planner system.

Changes

Cohort / File(s)	Summary
Planner Agent Documentation .claude/commands/debug-cursor-api.md, .claude/commands/launch-planner-agent.md, .claude/commands/modify-agent-prompt.md	Adds documentation for the new read-only Planner Agent feature, including API reference examples (Get Agent Conversation, Launch Planner Agent), end-to-end workflow patterns, plan extraction and iteration mechanisms, integration with pollers, and planner system prompt configuration for HITL planning.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 Behold! A planner agent so bright,
Read-only wisdom, planning right,
With HITL loops and prompts galore,
Documentation opens up the door,
No code is changed, just thoughts divine! ✨

✨ Finishing touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch hitl-spike-skills

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}