Pipeline templates¶

Templates live in data/pipelines/*.yml (project-level overrides are supported). Pick one per kind of work.

Template	Stages	Use it for
`aidlc-feature`	init → specify → clarify → plan → constitution → tasks → testplan → parallelize → implement → orchestrate → review → verify → deliver	A full feature with the implementation harness and delivery loop
`aidlc-mvp`	init → specify → clarify → plan → tasks → testplan → parallelize → implement → verify	Lean end-to-end delivery without review/deliver loops
`aidlc-classic`	init → specify → clarify → plan → tasks → testplan → parallelize → analyze	Design-only: stop before implementation
`aidlc-enterprise`	init → specify → clarify → checklist → plan → constitution → tasks → testplan → parallelize → implement → orchestrate → verify	Governance-heavy delivery
`aidlc-express`	init → specify → clarify → implement → verify	Small change, fast
`aidlc-bugfix`	init → specify → clarify → implement → verify	Fix a specific bug; verify loops back on failure
`aidlc-refactor`	init → specify → clarify → plan → implement → verify	Behaviour-preserving refactor
`aidlc-infra`	init → specify → clarify → plan → implement → verify	Infrastructure changes
`aidlc-security-patch`	init → specify → clarify → checklist → implement → verify	Security fix with a checklist
`aidlc-poc`	init → specify → plan → tasks → implement → verify	Proof of concept, no clarify
`aidlc-verify-loop`	init → specify → plan → tasks → testplan → implement → verify	Verify-fix loop until green
`aidlc-workshop`	init → specify → clarify → plan → tasks → implement → verify	Teaching / demo flow
`test-minimal`	init → specify	Smoke test

Schema¶

name: aidlc-feature
version: 1
description: >
  Full feature delivery.
steps:
  - id: code-generation          # unique step id (branch targets)
    stage: implement             # one of the stages listed in Pipelines & stages
    phase: construction          # initialization | ideation | inception | construction
    role: developer              # persona from data/personas
    model: anthropic/claude-sonnet-4-5   # optional per-step model
    thinking: medium             # off | minimal | low | medium | high | xhigh | max
    review: true                 # second-agent review harness
    humanGate: true              # pause for approval (autonomous mode auto-approves)
    maxIterations: 4             # cap when this step is a loop target
    onComplete:
      branch:
        - when: "verification_status == 'fail'"   # variables: verification_status,
          goto: code-generation                    # code_review_status, delivery_status, iteration
        - when: "true"
          goto: end
  retry:                          # optional run-level retry policy
    max: 2
    backoffMs: 30000

Branch expressions support <var> == 'literal', <var> != 'literal', true and false. goto names a step id or end; loops are capped by the target's maxIterations (default applies otherwise).

End-to-end test¶

bun run e2e [template …] creates a fixture project per template, waits for onboarding, enables autonomous + fast mode, starts a run and drives it to completion, then writes e2e-report.md. Environment: E2E_CONCURRENCY, E2E_TIMEOUT_MIN, E2E_MODEL, E2E_KEEP=1 (keep projects), E2E_REPORT.

The report has each template's tokens (input, cache reads, output) and cost per stage and the size of every artifact, also written as JSON (E2E_JSON, default e2e-report.json). To check that a prompt change saves tokens without losing stages or artifacts, run the same templates before and after and pass the first JSON as E2E_BASELINE:

E2E_JSON=before.json bun run e2e aidlc-mvp       # on main
E2E_BASELINE=before.json bun run e2e aidlc-mvp   # on the branch