Plan Skill

Automatically detect when users want to create development plans and handle them using the plan-forge MCP server.

Trigger Patterns

This skill should be invoked when the user's request matches patterns like:

• •"plan <task> in the background"
• •"create a development plan for <task>"
• •"start planning <task>"
• •"plan out <task>"
• •"make a plan for <task>"
• •"I need a plan for <task>"
• •"generate a plan for <task>"

Behavior

When triggered:

1. •Extract the task from the user's request
2. •Spawn background agent using the Task tool with plan-forge agent and run_in_background: true
3. •Acknowledge immediately - don't wait for planning to complete
4. •Provide status check instructions

Response Template

I've started a background planning session for: "<task>"

The plan-forge agent is now:
1. Analyzing the codebase
2. Generating a comprehensive development plan
3. Reviewing and refining until quality threshold is met

This typically takes 1-5 iterations (30 seconds to a few minutes).

**Check progress:** `/plan status`
**View results:** `/plan get plan`

I'll continue to be available while planning runs in the background.

Example Interactions

Example 1: Explicit background request

User: Plan adding OAuth authentication in the background

Claude: I've started a background planning session for: "adding OAuth authentication"
        [spawns plan-forge agent with run_in_background: true]
        Check progress with `/plan status`

Example 2: Implicit planning request

User: I need a development plan for implementing user roles

Claude: I've started a background planning session for: "implementing user roles"
        [spawns plan-forge agent with run_in_background: true]
        Check progress with `/plan status`

Example 3: Status check follow-up

User: How's that plan coming along?

Claude: [calls plan_status via MCP]
        Session: implementing-user-roles
        Status: in_progress
        Iteration: 2 of 5
        Still working on it...

Integration with /plan Command

This skill works alongside the /plan slash command:

• •Skill: Auto-detects planning intent in natural conversation
• •Command: Explicit invocation with /plan <task>

Both use the same underlying plan-forge MCP tools and background agent.

Handling needs_input Response

When you check on a background agent (via TaskOutput) and it returns with status="needs_input":

1. •

   Store the agentId for later resume

2. •

   Present questions clearly to user:

   "Planning paused - the reviewer needs your input:

   {reason from agent response}

   Please provide your answers and I'll continue planning."

3. •

   When user responds, resume the same agent:

   code

   Task tool:
   - prompt: "User provided these answers: {user_response}"
   - subagent_type: plan-forge
   - resume: {stored agentId}
   - run_in_background: true
   

This uses Claude Code's resumable subagents feature - the agent continues with full context and remembers what questions it asked.

Notes

• •Always use background execution for new plans
• •User can continue conversation while planning runs
• •Plans are saved to dev/active/<slug>/ when complete
• •If plan needs human input, agent stops and returns with needs_input status
• •Use the resume parameter to continue the same agent with user's answers