mrzhangguoguo Ported from oh-my-codex
pipeline. OMX runtime conventions ($macroinvocation,omxCLI,.omx/state directory) are replaced with WorkBuddy idioms (Skill tool, Agent tool, task list,.workbuddy/memory).
Pipeline Skill
pipeline sequences stages through a uniform flow: each stage is an ordered skill invocation (or Agent delegation), with progress tracked in the task list and durable artifacts in .omw/.
Default Autopilot Pipeline
The default Autopilot pipeline sequences:
deep-interview -> plan (consensus) -> ultragoal (+ team if needed) -> code-review -> ultraqa
team is conditional: invoke it only inside an active Ultragoal story when independent lanes or broad verification justify parallel execution. Explicit legacy Ralph pipelines remain available as custom stages, but Ralph is not the advertised default loop.
Configuration
Pipeline parameters are configurable per run:
| Parameter | Default | Description |
|---|---|---|
maxRalphIterations |
10 | Quality-gate retry ceiling; legacy option name retained for compatibility |
workerCount |
2 | Number of parallel Agent workers for a team stage |
agentType |
executor |
Role framing for team workers (expressed as the Agent prompt role) |
Stage Interface
Every stage implements the same shape:
name: stage identifier.run(ctx): perform the stage, accumulating artifacts from prior stages.canSkip?(ctx): return true only when its required durable artifacts already exist and are approved in order.
Stages receive a StageContext (accumulated artifacts from prior stages) and return a StageResult (status, artifacts, duration).
Built-in Stages
- deep-interview: Requirements clarification and ambiguity gate. Invoke the
deep-interviewskill. - plan (consensus): Consensus planning (Planner + Architect + Critic). Invoke the
planskill with consensus. Skips only when bothprd-*.mdandtest-spec-*.mdplanning artifacts already exist and durable consensus evidence records Architect approval followed by Critic approval. Plan/test-spec files alone are not consensus evidence. If either review is missing, blocked, out of order, or non-approving, stay inplanand fail with an explicit blocker/max-iteration outcome rather than progressing to execution. Carry anydeep-interview-*.mdspec paths forward for traceability. - ultragoal: Durable goal-mode execution; record ledgers under
.omw/ultragoal/. Launchteamonly from inside an Ultragoal story when parallel lanes are warranted. - code-review: Merge-readiness review gate. Invoke the
code-reviewskill. - ultraqa: Adversarial QA gate after a clean review; docs-only/trivially non-runtime changes may record an explicit skip reason.
- team-exec and ralph-verify: Legacy/custom pipeline adapters retained for explicit non-default pipelines.
State Management
Track pipeline state in the task list (one task per stage) rather than a hidden state file. The built-in WorkBuddy status display shows progress automatically. Resume is supported from the last incomplete stage.
- On start: create tasks for each stage; mark the first as
in_progress. - On stage transitions:
TaskUpdatethe current stage tocompletedand the next toin_progress. - On completion: mark all stages
completed; append a short summary note to.workbuddy/memory/YYYY-MM-DD.md.
If a durable record is needed across sessions, write the pipeline state to .omw/pipeline/<slug>.md (active stage, per-stage status, artifact paths).
Relationship to Other Skills
- deep-interview / plan / ultragoal / code-review / ultraqa: the default pipeline delegates to these skills in order.
- team: optional parallel execution engine inside an Ultragoal story when lanes are needed.
- ralph: available only for explicit legacy/custom pipelines.