# ai-autopilot-pipeline

> Standardize and automate a Ticket → roles → file handoff → validation engineering pipeline (Command Center autopilot) for repos that use a single in-progress ticket queue (e.g. `docs/tickets/QUEUE.md`) and per-ticket run folders (e.g. `docs/ai/runs/TICKET_ID`). Use when you want multi-role AI execution with strict guardrails (Spec Keeper, Builder, Reviewer, Runner) and prompts/workers.json generated to run in parallel Windows terminals.

- Author: jinnlink
- Repository: jinnlink/skills-box
- Version: 20260123110756
- Stars: 0
- Forks: 0
- Last Updated: 2026-02-06
- Source: https://github.com/jinnlink/skills-box
- Web: https://mule.run/skillshub/@@jinnlink/skills-box~ai-autopilot-pipeline:20260123110756

---

---
name: ai-autopilot-pipeline
description: "Standardize and automate a Ticket → roles → file handoff → validation engineering pipeline (Command Center autopilot) for repos that use a single in-progress ticket queue (e.g. `docs/tickets/QUEUE.md`) and per-ticket run folders (e.g. `docs/ai/runs/TICKET_ID`). Use when you want multi-role AI execution with strict guardrails (Spec Keeper, Builder, Reviewer, Runner) and prompts/workers.json generated to run in parallel Windows terminals."
---

# AI Autopilot Pipeline

## What this enables

Turn a repo’s ticket workflow into a repeatable, script-driven “autopilot” run: generate role prompts + workers files, spawn role windows, and keep handoffs as files (not chat).

## Assumptions (repo conventions)

- Ticket queue exists and enforces exactly one “In Progress” ticket (default: `docs/tickets/QUEUE.md`).
- Ticket doc paths are referenced in backticks inside the queue (common pattern: ``(`docs/tickets/TICKET_*.md`)``).
- Run folder convention exists (default: `docs/ai/runs/<TICKET_ID>/` with `00/10/20/30/40` files).
- Contract priority is enforced by docs (typical: `SPEC.md` > `CLAUDE.md` > `IMPL.md` > other docs).

## Quick Start (Windows)

1. Create prompts + workers for the current In Progress ticket:
   - `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\ai-autopilot-pipeline\\scripts\\new_ticket_run.ps1" -RepoRoot "E:\\path\\to\\repo"`
2. Optional git preflight + branch:
   - `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\ai-autopilot-pipeline\\scripts\\git_preflight.ps1" -RepoRoot "E:\\path\\to\\repo" -RunDir "<runDir>"`
   - `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\ai-autopilot-pipeline\\scripts\\git_checkout_ticket_branch.ps1" -RepoRoot "E:\\path\\to\\repo" -TicketId "<TICKET_ID>"`
2. Run stage 10 (Spec Keeper):
   - `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\multi-window-runner\\scripts\\spawn_windows.ps1" -WorkersFile "<runDir>\\_mwr\\workers_10_spec_keeper.json"`
3. Run stage 20 (Builder) after stage 10 output exists.
4. Capture a diff for Reviewer (recommended):
   - `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\ai-autopilot-pipeline\\scripts\\collect_diff.ps1" -RepoRoot "E:\\path\\to\\repo" -RunDir "<runDir>"`
5. Run stage 30 (Reviewer); then stage 40 (Runner).
6. Generate a single run summary file:
   - `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\ai-autopilot-pipeline\\scripts\\synthesize_ticket_run.ps1" -RunDir "<runDir>"`

## Optional: AI Council (interactive)

Use Council only for a single concrete dispute/blocker; do not ask every model to read the whole repo.

- `powershell -NoProfile -ExecutionPolicy Bypass -File "$env:USERPROFILE\\.codex\\skills\\ai-autopilot-pipeline\\scripts\\offer_council.ps1" -RepoRoot "E:\\path\\to\\repo" -RunDir "<runDir>" -Council ask`

## Script Outputs

`new_ticket_run.ps1` creates/updates (under the repo):
- `docs/ai/runs/<TICKET_ID>/00_context.md`
- `docs/ai/runs/<TICKET_ID>/_mwr/*_prompt.txt`
- `docs/ai/runs/<TICKET_ID>/_mwr/workers_*.json`

The worker commands call your installed bridge (default: `codex-cli-bridge`) and write role outputs:
- `docs/ai/runs/<TICKET_ID>/10_spec_keeper.md`
- `docs/ai/runs/<TICKET_ID>/20_builder.md`
- `docs/ai/runs/<TICKET_ID>/25_diff.patch` (optional but recommended for review)
- `docs/ai/runs/<TICKET_ID>/30_reviewer.md`
- `docs/ai/runs/<TICKET_ID>/40_runner.md`
- `docs/ai/runs/<TICKET_ID>/50_conclusion.md`

## Roles (fixed outputs)

- Spec Keeper → `10_spec_keeper.md` (contract pass/fail + guardrails + acceptance reminders)
- Builder → code changes + `20_builder.md` (impact + file list + 3-step acceptance + rollback)
- Reviewer → `30_reviewer.md` (issues by severity + fix order)
- Runner → `40_runner.md` (commands + raw output; if blocked, record failure evidence)

## Notes

- This skill generates prompts/workers; it does not assume any specific language or framework.
- Council/debate is optional; for debate-style convergence use `ai-council`.
- Avoid “everyone reads the whole repo”: keep reads scoped to `docs/PROJECT_SNAPSHOT.md` (if present), the current ticket, and the run artifacts (`00/10/20/25/30/40`). Use `ai-council` only when there is a concrete dispute or blocker.

## References

- Prompt templates and output contracts: `references/prompt_templates.md`