| name | deepen-plan |
| description | Enhance an existing plan with parallel research and review agents to add depth, best practices, and implementation details. Use when a plan needs more detail, when audit finds plan insufficiency, or when the user says "deepen plan", "enhance plan", 深化计划, 增强计划, 补全计划. |
| argument-hint | [path to plan file] |
Deepen Plan
Announce at start: "I'm using the deepen-plan skill to enhance this plan with research."
Note: The current year is 2026. Use this when searching for recent documentation and best practices.
Overview
Core principle: Research first, original content preserved always.
This skill takes an existing plan and enriches each section by dispatching parallel research and review agents. The result is a deeply grounded, production-ready plan with concrete implementation details, best practices, and edge case coverage.
NEVER write implementation code. This skill only researches and enhances plan documents.
When to Use
- A plan exists but lacks depth, best practices, or implementation details
- Audit stage (Stage 5 in
run-ce-workflow) finds "plan insufficiency"
- Your human partner says "deepen plan", "enhance plan", "补全计划"
When NOT to use:
- No plan file exists yet — use
writing-arch-plans or writing-impl-plans first
- The plan is already execution-ready and the user wants to implement — use
executing-plans
The Process
Create TodoWrite todos from this workflow:
TodoWrite todos:
- id: "dp-1", content: "Phase 1: Locate and parse plan file", status: "in_progress"
- id: "dp-2", content: "Phase 2: Dispatch research and review agents", status: "pending"
- id: "dp-3", content: "Phase 3: Synthesize findings", status: "pending"
- id: "dp-4", content: "Phase 4: Enhance plan and write file", status: "pending"
Phase 1: Locate and Parse Plan
1.1 Resolve plan file path
If argument-hint was provided, use it. Otherwise:
- Check for recent plans: look in
docs/plans/ and docs/brainstorms/
- Ask: "Which plan would you like to deepen? Provide the path."
Do NOT proceed to Phase 2 until you have a valid, readable plan file.
If the file does not exist, STOP and ask the user for a correct path.
1.2 Parse plan structure
Read the plan file. Identify its major sections and create a section manifest:
Section 1: [Title] - [What to research for this section]
Section 2: [Title] - [What to research for this section]
...
Present the manifest to the user for confirmation before dispatching agents.
Phase 2: Dispatch Research and Review Agents
Dispatch ALL of the following agent categories in PARALLEL. Use the Task tool for each.
2.1 Per-section research agents
For each section in the manifest, dispatch an explore subagent:
Task explore: "Research best practices, patterns, and real-world examples for: [section topic].
Return concrete, actionable recommendations with references."
Use WebSearch for current best practices (2024-2026).
2.2 Skill-matched agents
Discover available skills from plugin and global paths. For each skill whose description matches the plan content, dispatch a generalPurpose subagent to apply that skill's lens to the relevant plan section.
2.3 Learnings and solutions
Check docs/solutions/ for documented past learnings. For each relevant learning, dispatch a subagent to extract applicable insights.
2.4 Review agents
Discover review agents from .cursor/agents/ (all agents are flat in this directory). Dispatch each as a Task with the appropriate subagent_type.
MUST skip workflow agents (bug-reproduction-validator, pr-comment-resolver, spec-flow-analyzer, lint) and design agents (design-iterator, figma-design-sync) — workflow agents are orchestrators not reviewers, and design agents operate on screenshots/visual comparisons not document text.
Phase 3: Synthesize Findings
Collect outputs from all dispatched agents, then:
- Deduplicate — Merge similar recommendations from multiple agents
- Prioritize — Rank by impact (high-value improvements first)
- Flag conflicts — Surface contradictory advice for human review
- Group — Organize findings by plan section
Phase 4: Enhance Plan and Write
4.1 Enhance each section
For each plan section, append research insights using the format from enhancement-template.md. MUST preserve all original content unchanged.
4.2 Add enhancement summary
Add an Enhancement Summary block at the top of the plan. See enhancement-template.md for the exact format.
4.3 Quality checks
Before writing the file, verify all items in the quality checklist from enhancement-template.md.
4.4 Write the enhanced plan
Write the enhanced plan to the original file. If the user prefers a separate file, add -deepened suffix to the filename.
Post-Enhancement Options
After writing the enhanced plan, present via AskQuestion:
| Option | Description |
|---|
| View diff (recommended) | Show what was added/changed |
| Run audit | Dispatch review agents on the enhanced plan |
| Execute plan | Load executing-plans to begin implementation |
| Deepen further | Run another round on specific sections |
| Revert | Restore original plan from git |
Stage 5 Integration (run-ce-workflow)
When dispatched from run-ce-workflow Stage 5 audit:
- The
agent-native-reviewer identifies "plan insufficiency" (vague code, missing paths, incomplete commands, pseudocode, cyclic dependencies, non-machine-evaluable verify criteria)
- This skill is the fix action for those findings — enhance the specific insufficient sections rather than making direct code edits
- After enhancement, return control to the audit loop
Reference Files