pi-planner

name	pi-planner
description	Plan mode and plan proposal tools for consequential external actions. Use when: planning Odoo writes, email sends, calendar changes, deployments, cross-system workflows, or any irreversible operation that needs user approval. Triggers: plan mode, plan_propose, plan_approve, consequential action, external action, deploy plan, send plan, plan skill safety, safety registry.

Plan Mode

You have plan_mode and plan_propose tools for planning consequential actions.

Plan Mode — When to Enter and Exit

Call plan_mode(enable: true) when:

• The user says "let's plan", "prepare", "think through", or similar planning language
• You're about to research/gather context for external actions (Odoo writes, emails, deploys)
• The conversation shifts from implementation to planning consequential actions

Call plan_mode(enable: false) when:

• Planning is done (plan proposed, approved, or rejected) and you're resuming normal work
• The user asks you to do something that needs full tool access (edit, write, build, test)
• The user says "exit plan mode", "stop planning", or similar

In plan mode your tools are restricted to read-only: read, safe bash commands, and plan_* tools. File edits, writes, and destructive commands are blocked. This prevents accidental side effects while you research and plan.

When to Use plan_propose

Use plan_propose ONLY for consequential external actions:

• Writing to external systems (Odoo, email send, calendar, databases)
• Actions on behalf of other users
• Irreversible operations (deploy, send, delete)
• Multi-step workflows across systems

Do NOT plan for normal development work:

• File edits, code changes, refactoring
• Git operations (commit, push, branch)
• Build, test, lint commands
• Reading from any system
• Creating drafts or TODOs

Workflow

1. Enter plan mode: plan_mode(enable: true)
2. Gather context (read records, emails, data — all read-only tools work)
3. Propose: plan_propose(title, steps, context)
4. Wait for human approval (user uses /plan or /plans in TUI, or plan_approve tool)
5. Exit plan mode: plan_mode(enable: false)

Skill Safety Registry

When you read a skill that classifies operations with safety levels (READ, WRITE, DESTRUCTIVE, etc.), call plan_skill_safety to register those classifications. This enables intelligent plan mode filtering — READ operations (search, list, get) will be allowed in plan mode instead of blocked.

Extract command patterns from the skill docs and map them to READ or WRITE:

• READ: pure queries with no side effects (search, list, get, describe, show)
• WRITE: everything else — collapse DESTRUCTIVE/EXPENSIVE/SECURITY/FORBIDDEN to WRITE

Use glob patterns with * as wildcard. Patterns must start with the tool/CLI name.

plan_skill_safety({
  tool: "go-gmail",
  commands: {
    "npx go-gmail * search *": "READ",
    "npx go-gmail * get *": "READ",
    "npx go-gmail * thread *": "READ",
    "npx go-gmail * send *": "WRITE",
    "npx go-gmail * draft *": "WRITE"
  },
  default: "WRITE"
})

Call once per tool/CLI after reading its skill documentation. If in doubt, classify as WRITE.

Available Executor Tools

Check your context for the current tool inventory. Common tools:

• odoo-toolbox: read, write, create, delete, search
• go-easy: gmail (search, draft, send), calendar (list, create, delete)