| name | create-plan |
| description | Create detailed implementation plans through interactive research and iteration. Use when starting a new feature or ticket that needs a comprehensive technical plan. Saves plans to the repo-appropriate plans location and emits Personal Knowledge events when applicable. |
| allowed-tools | Read, Write, Bash, Agent, mcp__plugin_claude-mem_mcp-search__smart_search, mcp__plugin_claude-mem_mcp-search__smart_outline, mcp__plugin_claude-mem_mcp-search__smart_unfold |
Create Plan
Create detailed implementation plans through an interactive, iterative process.
Exploration Tools Priority
For code file exploration, prefer smart-explore tools (4-18x token savings):
| Tool | Use instead of |
|---|
smart_search(query, path) — discover files + symbols in one call | Glob + Grep discovery loop |
smart_outline(file_path) — structural skeleton (~1-2k tokens) | Read on files > 100 lines |
smart_unfold(file_path, symbol_name) — one symbol's full source | Read on large files for one function |
Rule: Use Read only for files < 100 lines or non-code files (markdown, JSON, config, tickets).
Initial Response
If no parameters: ask for task/ticket description and wait.
If file path provided: read it FULLY and begin immediately.
Tip: /create-plan thoughts/shared/tickets/eng_1234.md
Process
1. Read Context Files Fully
Read all mentioned files fully (no limit/offset). For code files > 100 lines, prefer smart_outline + smart_unfold. Do NOT spawn sub-tasks before reading.
1.5. Query claude-mem for prior context
Before spawning sub-agents, pull past work on this feature/ticket.
- Search:
mcp__plugin_claude-mem_mcp-search__search(query="<feature/ticket topic>", project="<repo>", limit=15, orderBy="relevance")
- Fetch the 3-6 most relevant IDs:
mcp__plugin_claude-mem_mcp-search__get_observations(ids=[...])
Embed results into sub-agent prompts and include a brief summary in "Present Understanding" so the user sees what past work informs this plan. Skip silently if unavailable.
2. Research in Parallel
Spawn parallel agents:
- codebase-locator: Find related files
- codebase-analyzer: Understand current implementation
- thoughts-locator: Find existing thoughts on this feature
- linear-ticket-reader: Get full ticket details (if Linear ticket mentioned)
Wait for ALL to complete. Then read all identified relevant files.
3. Present Understanding + Questions
Based on [ticket] and codebase research, I understand we need to [summary].
Found:
- [Current implementation at file:line]
- [Pattern to follow]
Questions (only what code can't answer):
- [Question 1]
4. Develop Plan Structure
Propose phasing first, get buy-in before writing details.
5. Write Plan
Path:
- Personal Knowledge repo:
plans/YYYY-MM-DD-description.md
- HumanLayer thoughts repos:
thoughts/shared/plans/YYYY-MM-DD[-ENG-XXXX]-description.md
- Other repos: use the nearest established
plans/, docs/plans/, or thoughts/shared/plans/ convention.
# [Feature] Implementation Plan
## Overview
## Current State Analysis
## Desired End State
### Key Discoveries
## What We're NOT Doing
## Implementation Approach
## Phase 1: [Name]
### Overview
### Changes Required
#### 1. [Component]
**File**: `path/to/file`
**Changes**: [summary]
### Success Criteria
#### Automated Verification:
- [ ] `make test`
#### Manual Verification:
- [ ] Feature works as expected
**Note**: Pause after automated verification for human manual testing confirmation.
---
## Testing Strategy
## References
6. Emit Personal Knowledge Event When Applicable
If the plan is created inside /Users/fredrick/Desktop/02_Personal/personal-knowledge or is about Fredrick's Personal Knowledge system, append a metadata-only event to:
/Users/fredrick/Desktop/02_Personal/personal-knowledge/queues/events/knowledge-events.jsonl
Event rules:
- Use
source_class: "plan" and signal_type: "artifact_completed".
- Use a repo-relative
source_ref, such as plans/YYYY-MM-DD-description.md.
- Keep
summary short and non-sensitive; do not copy plan body into the event.
- Set
privacy from the plan context when obvious; otherwise default to sensitive.
- Set
suggested_route to wiki_candidate or wiki_source_queue.
- Set
status: "queued".
- Add a candidate file only when the plan contains durable architecture, operating rules, or decisions that should later be reviewed for wiki ingestion.
- Validate the JSONL after writing:
jq -c . /Users/fredrick/Desktop/02_Personal/personal-knowledge/queues/events/knowledge-events.jsonl >/dev/null
7. Sync and Review
For Personal Knowledge repo plans, sync with git:
git status --short
git add plans/YYYY-MM-DD-description.md queues/events/knowledge-events.jsonl
git commit -m "docs(plans): add <description> plan"
git push origin main
For HumanLayer thoughts repos:
humanlayer thoughts sync
Present plan location and ask for feedback. Iterate until satisfied.
Guidelines
- Be skeptical: question vague requirements
- Be interactive: get buy-in at each step
- Be thorough: specific file paths and line numbers
- No open questions in final plan — resolve before writing
- Separate automated vs manual success criteria