plan-run

name	plan-run
description	Execute a YAML plan file, running each task in a fresh session with checks and auto-commit.
argument-hint	<plan-file> [--dry-run] [--task <id>] [--model <name>] [--no-commit]
user-invocable	true

Plan Run

Execute a YAML plan file generated by /blueprint. Each task runs in a separate session with automated checks, retries on failure, and git commits on success.

This spawns child sessions — the plan executor runs independently and reports results back here.

Process

1. Identify the plan file: Use $ARGUMENTS to find the plan file path. If no path is given, look for .yaml files in .claude/plans/ and list them for the user to pick.

2. Validate first: Run a dry-run to verify the plan is well-formed before executing:

   claude-hooks-plan --dry-run <plan-file>
   

3. Show the plan summary to the user: goal, number of steps/tasks, which checks will run. Ask for confirmation before proceeding.

4. Execute the plan:

   claude-hooks-plan <plan-file> 2>&1
   

   Common options to pass through from $ARGUMENTS:

   • --dry-run — validate only, don't execute
   • --task <id> — resume from a specific task (skip earlier ones)
   • --model <name> — model for child sessions (default: sonnet)
   • --no-commit — skip auto-commits after each task
   • --max-retries <n> — max check retries per task (default: 7)

5. Report results: Show which tasks succeeded or failed, and if any failed, which check blocked them.

Rules

• Always dry-run first to catch plan errors before spawning sessions
• Ask the user for confirmation before executing (this will spawn multiple sessions)
• Pass through any flags from $ARGUMENTS to the underlying command
• If a task fails, report which check failed and suggest using --task <id> to resume
• Do not modify the plan file — if changes are needed, tell the user to edit it or re-run /blueprint