Autopilot

Autopilot

Run the full pipeline front-to-back: gather context, panel discussion, plan waves, execute, audit.

Human approval is requested only at two gates: after the panel produces its recommendations, and after the pre-ship audit. Everything else runs automatically.

---
config:
  layout: elk
---
graph LR
    A["/autopilot [work]"] --> B{".context/codebase/?"}
    B -->|No| C["/gather-context"]
    B -->|Yes| D["Use existing map"]
    C --> D

    subgraph PANEL["Panel Discussion"]
        E["3 stakeholder agents"] --> W1["Wave 1: positions"]
        W1 --> W2["Wave 2: debate"]
        W2 --> CONSENSUS["Consensus"]
    end

    D --> E
    CONSENSUS --> F["User approves"]
    F --> G["/plan-waves"]
    G --> H["/execute"]
    H --> AUDIT["/audit"]
    AUDIT -->|PASS| DONE(["Shipped"])
    AUDIT -->|FAIL| FIX["Fix"] --> AUDIT

Process

Step 1: Gather Context (auto)

Check if .context/codebase/ exists and is current.

• If missing: Run /gather-context automatically — spawn 4 mapper agents, write 7 docs, verify. No human approval needed.
• If exists: Use as-is. Print a one-liner: Codebase map current — skipping gather-context.

Step 2: Scope the Work

If work description provided as argument, use it. Otherwise ask:

What should autopilot build?

This is the only open-ended question. Everything after this is automated.

Step 3: Panel Discussion

Run /discuss --auto [work description].

This spawns three AI stakeholders who debate the implementation approach across 2+ waves, then produce consensus recommendations in .context/CONTEXT-<SLUG>.md and a full transcript in .context/DISCUSSION_LOG-<SLUG>.md.

See /discuss Auto Mode for the full panel discussion process.

Human gate: After the panel completes, present the consensus summary and wait for approval before continuing to planning.

Step 4: Plan Waves (auto)

Run /plan-waves with the slug from Step 3. The planner receives the panel's consensus, conventions, and concerns. No human approval needed — the panel already debated the structure.

Print the plan summary when done:

Plan generated: .context/PLAN-<SLUG>.md
[N] waves, [M] tasks

Step 5: Execute (auto)

Run /execute with the slug. All waves dispatched automatically. Cross-review runs per wave. Fix cycles run automatically (max 2 per wave). Commit per wave.

Print wave-by-wave progress:

Wave 1: [theme] — [N] tasks dispatched... complete (0 fix cycles)
Wave 2: [theme] — [N] tasks dispatched... complete (1 fix cycle)
...

Step 6: Pre-Ship Audit

Run the pre-ship audit from /execute Step 5:

• Convention compliance against CONVENTIONS.md
• Concern avoidance against CONCERNS.md
• Decision compliance against CONTEXT-.md
• Test coverage against TESTING.md

If FAIL: fix automatically (up to 2 attempts), then re-audit. If still failing after 2 attempts, pause and present failures to user.

If PASS: report completion.

Autopilot complete: [work description]

Pipeline: gather-context -> panel (3 stakeholders, [N] waves) -> plan -> execute -> audit
Waves executed: [N/N]
Tasks completed: [M/M]
Pre-ship audit: PASS

Files changed:
[consolidated list]

Artifacts:
- .context/CONTEXT-<SLUG>.md (decisions)
- .context/DISCUSSION_LOG-<SLUG>.md (panel transcript)
- .context/PLAN-<SLUG>.md (wave plan)