---
name: "source-command-pipeline-pipetest"
description: "Deterministic QA gate for explicitly scoped canonical testing tasks."
---

# source-command-pipeline-pipetest

Use this skill when the user asks to run the migrated source command `pipetest`.

## Command Template

# 🧪 Pipetest (deterministic testing-lane gate)

Targets: **$ARGUMENTS** (task ids whose canonical status is `testing`, or
`--all --tasks <scope>` for every testing task in one explicit scope).

You are running the ADR-0055 deterministic QA sign-off. The verdict is the
SUITE'S EXIT CODE — never your opinion of the code. Use task plan/checklist for the steps.

## 1. Run the suite

Run the project's full gate — in this repo `npm run ci`; in an installed
project, the runner `/qa-signoff` would use (check `qa.*` config / package
scripts). Capture the exit code and a one-line summary (suites passed, file
count, timestamp).

## 2. Green → complete what is proven

For each targeted canonical task in `testing`:

1. Verify the task's `acceptance[]` against the test output, then run immediately:
   `node contextkit/tools/scripts/pipeline.mjs auto-transition <id> done --tasks <scope> --evidence "<runner> exit 0 — <summary> @<ISO date>"`
   The verb re-validates testing status and requires the bound automated-test
   evidence. No additional human test is required. If it refuses, relay its
   reason verbatim; never fall back to a
   bare `move` to force the result.
2. If an acceptance criterion is unproved: report it and leave the task in
   `testing`.

## 3. Red → report; bounce only what is attributable

A red suite NEVER mass-bounces the lane. Identify which failing test belongs to
which card (its own new tests / its touched files). Only for an attributable
task: `node contextkit/tools/scripts/pipeline.mjs qa-reject <id> "<the failing
output tail>" --tasks <scope>`. Rejection returns the task to `backlog` and
clears stale current-cycle evidence. Everything else: report the failure and
stop — fixing is a separate decision.

The same `qa-reject` command handles later human feedback on a `done` task. If
its Workflow is already complete, the aggregate and full package reopen first;
the task then begins a fresh backlog → working → testing → done cycle.

## 4. Report

One table: card → verdict (approved / left-in-testing + why / rejected) +
the suite line. Note any card the human still has to look at.

## Hard rules (ADR-0055 / ADR-0043)

- A green automated receipt uses `auto-transition ... done`; `qa-approve`
  remains the explicit QA equivalent. Never use free-form `move` to fabricate
  test evidence.
- Evidence is mandatory and goes on the card + the event log (actor
  `automated-test` for the automatic green path, or `qa` for an explicit QA
  transition).
- A scoped controller may invoke `/pipetest` after implementation; it must not
  leave a proven task parked in `testing` waiting for a second approval.

## Token economy (ADR-0103)

Run the test suite through the compact runner so only a bounded summary enters
context, not the full log:

```
node contextkit/tools/scripts/economy/run-compact.mjs <your test command>
```

The full log is written to `runs/<id>/` (gitignored); the exit code is the only
pass/fail source. Skip only if `economy.compaction.enabled` is false in config.

If the host exposes quota/usage data during the run, write a metadata-only quota
snapshot so `/token-report` has real data:

```
node contextkit/tools/scripts/economics/quota-snapshot.mjs --write --source pipetest --capture-method manual <quota flags>
```

If quota data is not visible, report `quota-snapshot skipped: no host quota data`.