maceff-autonomous-sprint

Launch and sustain an autonomous sprint — time-bounded self-directed work with Markov-guided mode transitions.

When to Invoke

• •User authorizes an autonomous sprint ("ride the tsunami", "work for N hours", "GO!")
• •User says AUTO_MODE with a time allotment
• •Post-compaction recovery when a sprint timer is still active

Step 1: Activate AUTO_MODE

Invoke the auto-mode skill to handle authorization, permissions, and mode switch:

Skill(skill: "maceff-auto-mode")

Follow its protocol for authorization and activation. AUTO_MODE now auto-starts the Transcript Monitor for idle detection.

Step 2: Session Restart Gate (MANDATORY)

AUTO_MODE permissions only take effect after a CC session restart. You MUST negotiate this with the user before proceeding.

Print this message and STOP:

Permissions require session restart. Please restart (Ctrl-D then claude -c) and say GO to resume the sprint.

DO NOT proceed to Step 3 until the user has restarted and returned. Creating tasks is OK, but do not start work.

Step 3: Read Sprint Policy

macf_tools policy navigate autonomous_sprint
macf_tools policy read autonomous_sprint

Extract requirements by answering:

• •What does the policy specify about gate point behavior?
• •What interaction protocol does the policy define for recommendations?
• •What timer discipline does the policy specify?
• •What CL thresholds does the policy define?
• •What anti-patterns does the policy name?

Step 4: Set Up Sprint Infrastructure

1. •

   Scope the sprint task with timer (MANDATORY for timed sprints):

   bash

   macf_tools task scope set <task_id> --timer <minutes>
   

   The timer is the wind-down mechanism. Without it, wind-down becomes arbitrary.

2. •

   Verify TM is running (should be auto-started from Step 1):

   bash

   macf_tools transcript-monitor status
   

3. •

   Document sprint start in task notes:

   bash

   macf_tools task note <id> "Sprint launched at <time>. Timer: <N> min. Goal: <description>."
   

Step 5: Begin Work

Work within the scoped task. The two-gate stop mechanism governs the sprint lifecycle:

Scope gate + Timer gate: Both must clear before stopping.

• •Timer active, scope active: Scope gate blocks with Markov recommender. Follow the recommendation to continue productive work.
• •Timer expired, scope active: Complete the task with a report to clear scope gate.
• •Both cleared: Stop is allowed. AUTO_MODE persists until user returns.

Task note discipline (REQUIRED):

• •Prefix all notes with work mode: DISCOVER: analyzed v2.1.109 strings
• •Document ideas as task notes, not formal idea CAs: DISCOVER: idea — REPL registerTool for dynamic tools
• •After sprint, curate ideas from notes with user guidance

NO new task creation in AUTO_MODE unless the user directs it. Activity goes in task notes on the scoped task.

Scope completion before timer is ENCOURAGED: When the scope gate fires with timer still active, the Markov recommender suggests the next work mode. This is the designed transition mechanism — follow it to continue related productive work.

Anti-Patterns

• •No timer set: Inventing arbitrary wind-down timing instead of using --timer
• •Emergency de-escalation as wind-down: Using macf_tools mode set MANUAL_MODE to stop — that's for emergencies only
• •Scope stretching: Padding work to avoid hitting the timer gate — complete honestly, let Markov guide the next activity
• •Idea CA creation during sprint: Creating formal idea JSON files — use task notes instead, curate with user later

This skill chains: maceff-auto-mode (authorization + TM start) → session restart → autonomous_sprint policy (sustainability) + mode_system policy (recommender). All needed for a well-governed sprint.