Skip to content

goal-mode-v2.md

Latest commit

 

History

History
74 lines (56 loc) · 2.07 KB

goal-mode-v2.md

File metadata and controls

74 lines (56 loc) · 2.07 KB
Raw

Goal Mode v2 — Autonomous Conductor Issue

Use this template when creating goal-driven autonomous work that should persist across multiple agents, turns, and evidence gates until the objective is achieved or a real blocker appears.

Required Header

/goal
GOAL_MODE: yes
GOAL_MODE_V2: yes
HEAVY_PATH: yes
SELF_AWARENESS_REQUIRED: yes

Required Sections

RAW_REQUIREMENT

<-- operator shorthand or full intent. 3-word shorthands are valid input; the conductor will expand them during DESIGNING. -->

NON_NEGOTIABLES

<-- what must NOT be touched: agents, runtimes, lanes, config, secrets. Be explicit. Default: do not touch Workbench Max, daemon/Desktop/core, live autopilot config, secrets, OAuth, Vercel production. -->

SUCCESS_METRIC

<-- what observable state proves the goal is achieved. Not a task list. -->

OPERATOR_CALL_CONDITIONS

<-- when to stop and ask the human. Default: permission/secret/payment/runtime mutation, blocked lane is only route, 2 prior failures, design trade-off needs taste judgment. -->

Conductor Workflow (for the assignee)

  1. Post GOAL_LOCK (per workbench-goal-mode-v2 skill)
  2. If Heavy: post SELF_AWARENESS_BOOTSTRAP
  3. If L2_PRESSURE: post RV_PRESSURE_CHECK
  4. Enter DESIGNING → produce DECISION_PACKET
  5. Enter DISPATCHING → create bounded child issues
  6. Enter OBSERVING → monitor, re-route only on new evidence
  7. Enter REVIEWING → evaluate against closeout gates
  8. Pass → LEARNING / ARCHIVING → NEXT_GOAL or DONE
  9. Block → BLOCKER_CLASSIFIED → operator call or cooldown retry

Required Closeout

GOAL_MODE_V2_CLOSEOUT
goal_id:
objective:
state_machine_path:
decision_packets_produced:
issues_dispatched:
evidence_harvested:
noise_cancelled:
operator_calls:
residual_risk:
archive_actions_taken:
verdict: PASS | FLAG | BLOCK

Related

  • Skill: skills/workbench-goal-mode-v2/SKILL.md
  • Autopilot: autopilots/goal-conductor.md
  • State machine: see SKILL.md for full diagram and transition rules
  • Dedupe/cooldown policy: see SKILL.md "Dedupe, Cooldown, and Noise Controls"