From dominion-flow
Autonomously plans, executes, verifies, and advances through project build phases after PRD completion, with configurable phase range and retry limits.
npx claudepluginhub thierryn/fire-flow[--from-phase N] [--to-phase N] [--max-phase-attempts 3]# /fire-autonomous > Full autopilot after PRD is complete — plan, execute, verify, advance through all phases independently --- ## Purpose After discussions and PRD are complete (`/fire-1a-discuss`), this command takes over the entire build pipeline. Claude plans each phase, executes it, verifies the work (verifier + reviewer), fixes any issues, and advances to the next phase — all without human checkpoints. **Design principle (Boris Cherny):** "You don't trust; you instrument." All verification stays active. Verdicts auto-route to fix cycles instead of pausing for human input. The hum...
/buildingExecutes implementation plans through gated phases: Load plan, setup tracking, execute sections, verify tests, report status. Uses git worktrees for isolation.
/flowOrchestrates end-to-end feature development: creates, plans, implements, verifies, checkpoints, and merges. Supports autonomous batching, plan-only mode, fixes, and team ATDD.
/vt-flowOrchestrates full dev pipeline from feature (contracts→plan→implement→review→validate→PR) or bug (--fix) with interactive gates. Supports --to=STEP, --from=STEP, --deep, --auto flags.
/autonomousExecutes remaining milestone phases autonomously via discuss-plan-execute cycle per phase. Updates ROADMAP.md/STATE.md, creates phase artifacts, audits, completes, and cleans up.
/hatch3r-workflowGuides development lifecycle through Analyze, Plan, Implement, Review phases with agent orchestration; scale-adaptive Quick Mode for small tasks and optional Playwright browser verification.
/buildExecutes Shipyard phase plans using subagents or experimental teams, with review gates and state updates. Supports [phase-number], --plan N, --light flags.
Share bugs, ideas, or general feedback.
Full autopilot after PRD is complete — plan, execute, verify, advance through all phases independently
After discussions and PRD are complete (/fire-1a-discuss), this command takes over the entire build pipeline. Claude plans each phase, executes it, verifies the work (verifier + reviewer), fixes any issues, and advances to the next phase — all without human checkpoints.
Design principle (Boris Cherny): "You don't trust; you instrument." All verification stays active. Verdicts auto-route to fix cycles instead of pausing for human input. The human reviews the finished product, not intermediate steps.
Safety: Path verification, HAC, circuit breaker, fire-verifier, and fire-reviewer all remain active. Autonomous mode removes human pauses, not safety gates.
arguments:
--from-phase:
type: integer
default: "next incomplete phase from CONSCIENCE.md"
description: "Start from this phase number"
--to-phase:
type: integer
default: "last phase in VISION.md"
description: "Stop after completing this phase"
--max-phase-attempts:
type: integer
default: 3
description: "Max plan-execute-verify cycles per phase before escalating"
--max-iterations:
type: integer
default: 50
description: "Max iterations per execution loop (passed to fire-3-execute)"
--dry-run:
type: boolean
default: false
description: "Show what would be done without executing"
--token-efficient:
type: boolean
default: false
description: "Enable ALAS context slicing for ~60% token reduction. Passes --token-efficient to fire-3-execute. Default OFF for best quality."
Read .planning/VISION.md → extract all phases
Read .planning/CONSCIENCE.md → get current status, find next incomplete phase
IF --from-phase not specified:
from_phase = first phase with status != complete in CONSCIENCE.md
IF --to-phase not specified:
to_phase = last phase in VISION.md
Validate:
1. .planning/ directory exists
2. VISION.md has phases in range [from_phase, to_phase]
3. For each phase in range: MEMORY.md exists (PRD/discussion complete)
4. Path Verification Gate (MANDATORY — working directory matches project)
5. Cross-Phase DAG Validation (v13.0 — from fire-3-execute Step 3.55)
→ Check for circular dependencies between phases
→ Validate provides/requires contracts match across phase boundaries
→ STOP if cycles found or contracts broken
IF any phase lacks MEMORY.md:
Display:
"Phase {N} has no MEMORY.md — discussions not complete.
Run /fire-1a-discuss {N} first.
Autonomous mode requires all target phases to have context."
STOP
Same as all Dominion Flow commands:
1. Verify working directory matches expected project path
2. Check for cross-project contamination risk
3. Validate subagent path injection safety
This gate is NEVER disabled, not even in autonomous mode.
+--------------------------------------------------------------+
| AUTONOMOUS MODE ACTIVATED |
+--------------------------------------------------------------+
| |
| Project: {project name from VISION.md} |
| Phases: {from_phase} to {to_phase} ({count} phases) |
| Max attempts per phase: {max_phase_attempts} |
| |
| Active Safety: |
| Path verification — MANDATORY (cannot disable) |
| HAC enforcement — Active (confidence 5/5 rules) |
| Circuit breaker — Active (stall/spin/degrade detect) |
| fire-verifier — 70-point WARRIOR validation |
| fire-reviewer — 15-persona code review |
| |
| Token mode: {IF --token-efficient: EFFICIENT (ALAS) | FULL CONTEXT (best quality)} |
| |
| Human touchpoints: NONE until completion or blocker |
| |
| To interrupt: Ctrl+C or /fire-loop-stop |
| |
+--------------------------------------------------------------+
IF --dry-run:
DRY RUN — would execute:
Phase {N}: {name}
1. /fire-2-plan {N} --skip-checker
2. /fire-3-execute {N} --auto-continue --autonomous
3. Read verification + review results
4. Auto-fix if needed (up to {max_phase_attempts} attempts)
Phase {N+1}: {name}
...
No changes made. Remove --dry-run to execute.
STOP
autonomous_log = []
phases_completed = 0
start_time = now()
FOR phase_number in range(from_phase, to_phase + 1):
phase_name = VISION.md phase name
attempt = 0
phase_complete = false
Display:
"━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
" AUTONOMOUS: Phase {phase_number} — {phase_name}"
"━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
WHILE attempt < max_phase_attempts AND NOT phase_complete:
attempt += 1
Display: " Attempt {attempt}/{max_phase_attempts}..."
// ──────────────────────────────────────────────────────
// Step 3.1: AUTO-PLAN
// ──────────────────────────────────────────────────────
phase_dir = .planning/phases/{phase_number}-{phase_name}/
plan_files = glob("{phase_dir}/*.BLUEPRINT.md")
IF plan_files is empty:
Display: " Planning phase {phase_number}..."
Run /fire-2-plan {phase_number} --skip-checker
// skip-checker: plan issues surface during verification
// auto-replan on failure is more effective than blocking on warnings
ELSE:
Display: " Plans exist ({count} plans). Executing..."
// ──────────────────────────────────────────────────────
// Step 3.2: AUTO-EXECUTE
// ──────────────────────────────────────────────────────
Display: " Executing phase {phase_number}..."
Run /fire-3-execute {phase_number} --auto-continue --autonomous {IF --token-efficient: --token-efficient}
// --auto-continue: no breath-boundary interrupts
// --autonomous: merge gate auto-routes verdicts (v9.0, enhanced v12.0)
// --token-efficient: passed through if set (ALAS context slicing, ~60% token reduction)
// ──────────────────────────────────────────────────────
// Step 3.3: AUTO-VERIFY (read results)
// ──────────────────────────────────────────────────────
verification_file = {phase_dir}/{N}-VERIFICATION.md
review_file = {phase_dir}/{N}-REVIEW.md
Read verification_file → extract warrior_score, verifier_verdict
Read review_file → extract reviewer_verdict, finding_counts
// ──────────────────────────────────────────────────────
// Step 3.4: Evaluate Results
// ──────────────────────────────────────────────────────
IF verifier_verdict == "APPROVED" AND reviewer_verdict == "APPROVE":
phase_complete = true
phase_result = "CLEAN PASS"
Display: " Phase {N}: CLEAN PASS (verifier {score}/70, reviewer APPROVE)"
ELIF verifier_verdict in ["APPROVED", "CONDITIONAL"] AND reviewer_verdict in ["APPROVE", "APPROVE WITH FIXES"]:
phase_complete = true
phase_result = "PASSED WITH NOTES"
Display: " Phase {N}: PASSED WITH NOTES"
Display: " Verifier: {verifier_verdict} ({score}/70)"
Display: " Reviewer: {reviewer_verdict} ({finding_counts})"
Display: " Non-critical gaps logged to autonomous-log.md"
ELSE:
phase_result = "NEEDS FIX"
Display: " Phase {N}: NEEDS FIX (attempt {attempt})"
Display: " Verifier: {verifier_verdict}"
Display: " Reviewer: {reviewer_verdict}"
IF attempt < max_phase_attempts:
Display: " Re-planning for gaps..."
Run /fire-2-plan {phase_number} --gaps
// Loop back to Step 3.2 (next WHILE iteration)
ELSE:
Display: " Max attempts reached. Escalating."
GOTO Step 7 (Escalation)
// Log attempt
autonomous_log.append({
phase: phase_number,
attempt: attempt,
verifier: verifier_verdict,
reviewer: reviewer_verdict,
warrior_score: score,
result: phase_result
})
// ──────────────────────────────────────────────────────
// Step 3.5: Phase Transition
// ──────────────────────────────────────────────────────
IF phase_complete:
phases_completed += 1
Update CONSCIENCE.md:
Phase {phase_number}: complete
Autonomous: true
Attempts: {attempt}
Verifier: {score}/70
Reviewer: {reviewer_verdict}
// Commit checkpoint
git add -A && git commit -m "autonomous: Phase {phase_number} - {phase_name} complete"
Display: " Phase {phase_number} complete. Advancing..."
STOP AUTONOMOUS MODE immediately if ANY of:
1. Circuit breaker TRIPPED with state = BLOCKED
→ Agent cannot make progress. No amount of retry will help.
→ Display: "Circuit breaker tripped: {reason}"
2. Path verification gate fails
→ Working directory mismatch or cross-project contamination detected.
→ Display: "PATH VIOLATION: {details}"
3. HAC HARD BLOCK fires (confidence 5/5 rule match)
→ A known-bad action was attempted.
→ Display: "HAC BLOCK: Rule {N} — {rule statement}"
4. 3 consecutive phase attempts fail (max_phase_attempts exceeded)
→ Genuine blocker in the phase.
→ Display: "Phase {N} failed after {max_phase_attempts} attempts"
5. Context degradation detected (output volume decline >50%)
→ Context rot. Fresh session needed.
→ Display: "Context degradation detected. Sabbath Rest recommended."
On ANY blocker: GOTO Step 7 (Escalation)
Write .planning/autonomous-log.md:
# Autonomous Session Log
## Session: {timestamp}
| Key | Value |
|-----|-------|
| Phases | {from_phase} to {to_phase} |
| Completed | {phases_completed} of {total} |
| Duration | {elapsed_time} |
| Mode | Full autopilot |
### Phase Results
| Phase | Name | Attempts | Verifier | Reviewer | Result |
|-------|------|----------|----------|----------|--------|
| {N} | {name} | {attempts} | {score}/70 | {verdict} | {result} |
| {N+1} | {name} | {attempts} | {score}/70 | {verdict} | {result} |
### Gaps Noted
{List of non-critical gaps from PASSED WITH NOTES phases}
### Behavioral Directive Proposals
{Any HIGH/CRITICAL findings from reviewer that should become rules}
### DORA Metrics (v12.0 — AUTONOMOUS_ORCHESTRATION skill)
| Metric | Value | Notes |
|--------|-------|-------|
| Deployment Frequency | {phases_completed}/{elapsed_time} | Phases per hour |
| Change Lead Time | {avg time from plan to verified} | Plan→Execute→Verify avg |
| Change Failure Rate | {failed_attempts}/{total_attempts} | Lower = better quality plans |
| Recovery Time | {avg time from NEEDS FIX to next PASS} | Re-plan + re-execute avg |
### Supervised Autonomy Assessment (v12.0)
| Tier | Description | This Session |
|------|-------------|-------------|
| Human-in-the-loop | Every decision approved | Not used |
| Human-on-the-loop | Monitoring, intervenes on blockers | Active (via circuit breaker) |
| Human-out-of-loop | Full autonomy with instrumentation | {if no blockers: "Achieved"} |
+--------------------------------------------------------------+
| AUTONOMOUS MODE COMPLETE |
+--------------------------------------------------------------+
| |
| Phases completed: {phases_completed} of {total} |
| Total attempts: {sum of all attempts} |
| Duration: {elapsed_time} |
| |
| Results: |
| Phase {N}: {result} (verifier {score}/70, reviewer {verdict}) |
| Phase {N+1}: {result} (...) |
| ... |
| |
| Gaps noted: {count} |
| Log: .planning/autonomous-log.md |
| |
+--------------------------------------------------------------+
| NEXT |
+--------------------------------------------------------------+
| |
| Your work is ready for review. |
| |
| /fire-dashboard — See project status |
| /fire-4-verify {N} — Detailed verification |
| /fire-session-summary — Auto-generated (see above) |
| /fire-5-handoff — Full WARRIOR handoff (milestones) |
| |
+--------------------------------------------------------------+
+--------------------------------------------------------------+
| AUTONOMOUS MODE STOPPED — BLOCKER |
+--------------------------------------------------------------+
| |
| Completed: Phases {from_phase} to {last_completed} |
| Blocked at: Phase {blocked_phase}, attempt {attempt} |
| |
| Blocker: {description} |
| Type: {BLOCKED | PATH_VIOLATION | HAC_BLOCK | MAX_ATTEMPTS | |
| CONTEXT_DEGRADATION} |
| |
| What was tried: |
| - Attempt 1: {verifier_verdict} / {reviewer_verdict} |
| - Attempt 2: {verifier_verdict} / {reviewer_verdict} |
| - Attempt 3: {verifier_verdict} / {reviewer_verdict} |
| |
| Log: .planning/autonomous-log.md |
| |
+--------------------------------------------------------------+
| OPTIONS |
+--------------------------------------------------------------+
| |
| A) /fire-debug — Investigate the blocker |
| B) /fire-3-execute {N} — Manual execution with oversight |
| C) /fire-autonomous --from-phase {N} — Retry from here |
| D) /fire-dashboard — Review current state |
| |
+--------------------------------------------------------------+
Run /fire-session-summary
This generates a compact forward-looking summary with:
- Aggregate status (DONE/PARTIAL/BLOCKED for each work item)
- Readiness assessment (what's ready, blocked, needs-first)
- Outlook (trajectory, risk, momentum)
- Next steps (3 specific actions)
- Decisions made
Saved to ~/.claude/session-summaries/{project}_{date}.md
Auto-indexed into Qdrant on next session-end consolidation.
This is NOT a replacement for WARRIOR handoffs — it's the lightweight
forward-looking context that the memory system doesn't capture.
Update .claude/dominion-flow.local.md:
---
last_session: {timestamp}
command: "autonomous"
status: {complete | blocked}
phases_completed: {list}
blocked_at: {phase or null}
blocker_type: {type or null}
log: ".planning/autonomous-log.md"
---
# Autonomous Session State
## Last Run
- Date: {timestamp}
- Phases: {from} to {to}
- Completed: {count}/{total}
- Duration: {elapsed}
## Resume Point
IF blocked:
Resume with: /fire-autonomous --from-phase {blocked_phase}
ELSE:
All phases complete. Next milestone or new discussions.
IF phases_completed > 0: Create WARRIOR handoff with autonomous session context.
NEVER DISABLED in autonomous mode:
1. Path Verification Gate (Step 0.5)
— Wrong-repo editing is catastrophic and irreversible
— No confidence level justifies skipping this
2. HAC Hard Blocks (confidence 5/5 anti-patterns)
— Known-bad actions are blocked BEFORE execution
— These represent "we know this is wrong" conclusions
3. Circuit Breaker (BLOCKED state)
— Agent admits it cannot proceed
— No amount of retry will fix external blockers
4. Power-Verifier (70-point WARRIOR validation)
— Functional correctness checking stays active
— Verdicts auto-route but are never suppressed
5. Power-Reviewer (15-persona code review)
— Quality and simplicity checking stays active
— BLOCK verdicts force fix cycles, not human pauses
WHAT IS DISABLED in autonomous mode:
1. "READY FOR HUMAN" pause at merge gate
— Auto-proceeds to next phase instead of waiting
2. "Continue?" interrupts between breaths
— Handled by --auto-continue flag
3. Plan-checker blocking
— Handled by --skip-checker flag
— Plan issues surface in verification instead
4. Human review of individual phase results
— Results logged, human sees final product
# Full autopilot from next incomplete phase to end
/fire-autonomous
# Run phases 3 through 7
/fire-autonomous --from-phase 3 --to-phase 7
# Preview without executing
/fire-autonomous --dry-run
# Custom iteration limits
/fire-autonomous --max-phase-attempts 5 --max-iterations 100
/fire-1a-discuss — Gather context BEFORE autonomous mode (required)/fire-2-plan — Manual planning (autonomous mode auto-plans)/fire-3-execute — Manual execution (autonomous mode auto-executes)/fire-4-verify — Manual verification (autonomous mode auto-verifies)/fire-loop — Task-level loop (autonomous mode is phase-level)/fire-loop-stop — Emergency stop for autonomous mode/fire-dashboard — View results after autonomous runDominion Flow v12.0 — Full autopilot with instrumentation over restriction Boris Cherny principle: "You don't trust; you instrument."