How to use the Test Evidence Review agent skill

What does the Test Evidence Review skill make the agent do?

The skill goes beyond checking that test files exist. Its own description says it evaluates assertion coverage, meaning it looks at whether the tests check the behaviour the story requires. A test file that exists and passes can still leave critical behaviour uncovered, and this skill is built to catch that gap.

For each story it reviews, the agent gives one of four verdicts: ADEQUATE, INCOMPLETE, MISSING, or NOT ASSESSED. It flags blocking gaps, meaning gaps that would stop the story from being closed. It writes a summary in the chat and can optionally save a report file.

• Reads the story's acceptance criteria and Test Evidence section
• Locates the test files or evidence documents for the story
• Counts assertions per test function and looks for edge-case coverage
• Gives each story a verdict and lists blocking gaps
• Optionally saves a report to production/qa/evidence-review-[date].md

How do you install the skill?

Download the ZIP file from this page. You can install it in one of two ways, depending on where you use Claude.

For Claude Code, unzip the download into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that project. The skill folder should contain SKILL.md.

In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal use across projects: unzip into ~/.claude/skills/
• Single project only: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What does the skill need before it works well?

The skill is designed for a Claude Code environment that has the studio's .claude folder, including its rules, docs, hooks, and the story and sprint file layout. It depends on those files and paths. If they are missing, the skill will report gaps or NOT ASSESSED rather than a clean result.

The skill's allowed tools reference two helpers: the hooks/yaml-helper.sh config helper and .claude/scripts/review-receipts.sh. The bundled config and receipt helper scripts are not included in the reviewed files, so check that they exist in your project before you rely on the skill.

• A Claude Code environment with the studio's .claude folder
• Story files under production/epics/ and sprint plans under production/sprints/
• The hooks/yaml-helper.sh config helper
• The .claude/scripts/review-receipts.sh helper

How do you trigger a review?

Run the skill with an argument that sets its scope. The argument can be a story path, the word sprint, or a system name. The skill reads the scope you give it and reviews the stories in that scope.

If you run it with no argument, the skill asks which scope you want: a single story, the current sprint, or a system. You can also ask the agent in plain words, for example that it check whether the tests for a particular story cover its acceptance criteria. Naming the story, sprint, or system makes the scope clear.

• Single story: pass the story file path, for example a file under production/epics/
• Current sprint: pass sprint, which reads the most recently modified file in production/sprints/
• One system: pass the system name, which reviews production/epics/[system-name]/story-*.md
• No argument: the skill asks you to choose the scope

How are the verdicts and blocking items decided?

Whether a gap counts as blocking depends on the story type and the studio's testing settings. The skill reads the qa.level and testing.strict settings and applies the same rules that /story-done uses to close stories, so a BLOCKING item here is one that would stop /story-done. Logic, Integration, Visual/Feel, and UI gaps are blocking by default. Config/Data gaps are advisory by default.

A test function with zero assertions is flagged as blocking, because it passes without proving anything. A test function with one or two assertions is noted as potentially thin. At qa.level: minimal, Logic, Integration, and Config/Data stories with no test evidence are marked WAIVED rather than MISSING, and they raise no blocking item. Visual/Feel and UI evidence is never waived.

• ADEQUATE: the tests appear to cover the acceptance criteria
• INCOMPLETE: some behaviour or edge cases are not covered
• MISSING: no test evidence was found for a story that needs it
• NOT ASSESSED: the skill could not check the story, for example because the scope was empty or its files were absent
• WAIVED (qa.level: minimal): no test evidence required for this story type at this level

What about safety, scripts, and network access?

The skill runs scripts: its allowed tools include Bash calls to the config helper and the receipt helper. It does not use the network. The shown SKILL.md only reads and reviews files, and it can write the optional report file.

The bundled helper scripts are not included in the files reviewed for this page, so read them yourself before installing if you need to confirm what they do. The skill does not run your tests. It judges the test and evidence documents that already exist.

• Runs local scripts through allowed Bash commands
• Does not use the network, according to this listing
• Does not run your tests
• Can write a report file when you ask for one