Navin

Plan

Navin Plan and Mission: /blueprint, /forge, /cruise and a native ledger

August 5, 2026 · 2 min read · Navin team

Native Magentic-style orchestration: Task Ledger, Progress Ledger, mandatory evidence, pause/resume and Build handoff - no external framework.

Plan without executing, execute without losing the thread, resume a mission after a pause: that is the /blueprint/forge/cruise / /mission trio. Navin does not invent a todo in chat: it keeps a native Task Ledger and Progress Ledger on the board.

Inspired by Magentic-One / Deep Agents loops, implemented without LangGraph or an external package - on Navin's board and AgentLoop.

The four levels

IntentCommandBehavior
Plan only/blueprintGoal analysis + Task Ledger + board tasks. No code.
Plan then buildBuild → /forgeExecute the plan, Progress Ledger on every step.
Autopilot/cruisePlan, execute, test, detect stalls, local replan until done or paused.
Long mission/missionDurable multi-turn goal + ledger + checkpoints + resume + final report.

Composer UI modes stay plan / agent. Tools stay Navin's (board, spawn, verify, test_run, files, shell, MCP…).

The protocol on every step

On /forge, /cruise, /mission:

  1. Read the ledger (ledger_get) + board next.
  2. Pick one ready step (or spawn independent ready steps).
  3. Act with tools.
  4. Verify per step validation (test / lint / verify / manual / none).
  5. Write ledger_progress with evidence.
  6. Decide: next | retry | ledger_replan | ledger_pause.
  7. Never mark done without evidence when validation requires it.

If missing_info blocks progress → pause and ask the human. Do not invent facts.

What is persisted

.navin/board/mission.json next to the board:

  • Task Ledger: goal, status, version, constraints, facts, missing_info, acceptance_criteria, steps[], history[]
  • Progress Ledger: current_step_id, evidence, stall_count, loop_detected, replan_needed, budgets, next_actor

Board tasks carry acceptance, validation, retry_count, agent, evidence. A done with validation in test|lint|verify without evidence is rejected.

Pause, resume, human edit

  • ledger_pause / ledger_resume
  • Human edits via ledger_update bump version and append history (actor=human)
  • Chat plan panel: goal, vN, stall badge, last change reason, Build after /blueprint

Sub-agents

spawn briefs copy the step's acceptance + validation. The parent merges evidence before done / ledger_progress.

Concrete examples

/blueprint Add GitHub OAuth without breaking Google Auth
# review plan panel → Build

/forge Implement the approved plan

/cruise Ship dark mode end to end with tests

/mission Migrate auth to Supabase across the monorepo

Plan/Mission + board autonomy

Once the plan is approved, board autonomy chains ready work: isolated branch, PR on done, issues sync. The ledger decides what; autonomy decides how to ship in git.

Also see task discipline: expert board.

FAQ

Does /blueprint touch code?

No. Design only. Code starts with Build / /forge.

/cruise vs /mission?

/cruise: delivery autopilot. /mission: long goal, checkpoints, multi-session resume, final report.

Who replans?

Local replan first (ledger_replan). Pause on budget / human / missing info.

Download Navin · Features · Plan Mode docs

Conclusion

Plan/Mission is the difference between an agent that "codes a bit" and an orchestrator that plans, proves, replans and resumes - like a serious team.

Try Navin on your machine

Local agent, cross-platform. Code, debug, scrape, leads, security and review - without leaving Navin.

Related reading