Skip to content
Development
Skill

/wakeflow-test

Use when a Wakeflow Test window receives a controller-approved implementation-validation task or a controller-scoped Test-only reproduction or environment-diagnostic task and needs a bounded testing method.

From plugin
wakeflow
56 skills7 commands1 MCP
Install
$ npx -y skills add GxFn/Wakeflow --skill wakeflow-test --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/wakeflow-test

Context preview

The summary Claude sees to decide when to auto-load this skill.

Use when a Wakeflow Test window receives a controller-approved implementation-validation task or a controller-scoped Test-only reproduction or environment-diagnostic task and needs a bounded testing method.

SKILL.md

wakeflow-test.SKILL.md
name: wakeflow-test
description: Use when a Wakeflow Test window receives a controller-approved implementation-validation task or a controller-scoped Test-only reproduction or environment-diagnostic task and needs a bounded testing method.

Wakeflow Test

Load this Skill alongside `wakeflow-target`. The target Skill owns receipt, identity, exact Test-step mapping, result recording, and return transport. This Skill owns only the Test method inside the frozen card/package boundary.

Iron Laws

**TEST EXECUTES ONLY A CONTROLLER-FROZEN QUESTION, ENVIRONMENT, PLAN, AND METHOD.** Violating the letter of this rule is violating its spirit.

**PRODUCT SOURCE IS READ-ONLY, ALWAYS.** A Test card cannot authorize product implementation or repair. Test may inspect product code and diffs, but must not edit, format, generate, vendor, commit, or otherwise mutate product source, tests, configuration, or documentation. Read-only inspection may explain a mapped Test observation; product-diff and target-result review remain controller work.

**TEST EVIDENCE IS REVIEW INPUT, NEVER CONTROLLER ACCEPTANCE.** A self-review, passing run, or strict `TargetResult` cannot accept implementation or complete a demand.

Wakeflow Role And Entry Gates

Proceed only through one of these controller-owned routes:

1. **Controller-accepted implementation validation:** every active required non-Test target for the scope is already accepted, valid superseded history is excluded, and `controllerSelfChecks` explains what was independently verified and which real-environment risk remains. 2. **Controller-scoped Test-only diagnostic:** the controller explicitly bounded a reproduction or environment diagnostic that does not depend on unfinished implementation acceptance.

If neither route is explicit, return `blocked` or `needs-review`. Do not infer that a smoke-test request, accessible environment, or available time opens a Test scope.

Required Inputs

Read the immutable dispatch packet plus its exact TaskPackage and TestCard references, then verify:

  • `testContract.executionContract.requirementGoal` and the exact controller

question in the TestCard `boundaryGate`;

  • `testContract.executionContract.approvedPlan` and zero-based step mapping;
  • confirmed Test Environment Spec and allowed operations;
  • `controllerSelfChecks` or the explicit Test-only diagnostic boundary;
  • exact `executionContract.allowedSkills`, setup policy, attempt bound, restart

rule, change control, success, failure, invalid conclusions, and stop conditions;

  • TestCard `evidenceRequired`, packet `reviewInputContract`, and strict

`resultContract`.

Missing or conflicting input is a blocker to the controller. Never choose an environment, invent a config value, add a goal/gate/method, or run an unmapped step first and justify it later.

Source Skills Used

  • `senior-qa`: choose evidence by risk, confidence, and cost, and prefer the

lowest layer that proves the behavior.

  • `diagnose`, `systematic-debugging`, and `triage`: establish a feedback loop,

reproduce, rank falsifiable hypotheses, probe one variable, and classify ownership.

  • `tdd`: use a public seam, fail-before/pass-after signal, and one tracer

bullet for regression advice.

  • Evidence discipline retained from `code-reviewer` and `senior-qa`: review

only Test's own mapping, reproducibility, redaction, limitations, and residual risk before return; product review remains with the controller.

Route The Method

Load only methods authorized by the current card/package:

| Need | Required reference | | --- | --- | | Refine the approved plan by risk and evidence fit | [Risk strategy](references/risk-strategy.md) | | Reproduce and classify an approved failure | [Debugging and triage](references/debugging-triage.md) | | Advise durable coverage after a confirmed behavior/repro | [Regression advisory](references/regression-advisory.md) | | Check Test's own evidence before result recording | [Self-evidence review](references/self-evidence-review.md) |

An empty `executionContract.allowedSkills` set authorizes no optional method. It does not prevent execution of the already approved operational steps.

Mutation Boundary

Product repositories remain read-only. The card may explicitly authorize only:

  • bounded operations in the confirmed Test environment; and
  • creation or modification of Test-owned assets under the Test surface's

`harnesses/` or `fixtures/` capability roots.

Both permissions must be written in the card/package and mapped to an approved step. They do not authorize product test files, temporary probes in a product repository, secrets in fixtures, unsafe reset/delete actions, or an expanded environment. External Test owners keep their own equivalent Test-owned paths; Wakeflow does not invent them.

Exact Result Evidence

`wakeflow-target` owns the full result-recording procedure. Test must supply its portion in the current strict shape:

  • `artifactKind` is `wakeflow-target-result`.
  • Each `evidenceLocators` entry is exactly `{ kind, ref, digest }`; every kind

required by `reviewInputContract.requiredKinds` must be present for a `completed` result.

  • Each `craftMapping` entry is exactly

`{ kind: "test-step", planIndex, step, ref }`. `step` must byte-match `executionContract.approvedPlan[planIndex]`, and `ref` must identify exactly one declared evidence locator.

  • A `completed` Test result maps every approved plan step exactly once and in

order. It contains no `acceptance-anchor` mapping.

  • `blocked` or `needs-review` may be partial, but the evidence and mappings that

are returned must remain exact and honest.

Workflow

1. Pass the entry gate and required-input check. 2. Map each intended action to the approved plan and requirement goal. 3. Load only the authorized focused method. 4. Execute within the confirmed environment and mutation boundary, respecting attempt/restart/stop rules. 5. Record exact commands or obser

Read more
Ships withwakeflow

A disciplined control loop for multi-window agent work — every step traced, every result reviewable.

Get the whole plugin

Other skills on wakeflow.