review-tests
Skill1 starsChanged 3 months agoMITView on GitHub
Judge test coverage and design. Use as a required review, or as /review-tests. Not for judging the code: review-code covers that.
9 other projects use a version of this.
Adopted by
Most recently taken up by
- wufei-png/skillstheir own version, first seen Oct 28, 2026
- desbma/agent-skillstheir own version, first seen Oct 27, 2026
- Arasz/ai-badgertheir own version, first seen Oct 19, 2026
- mrinalwadhwa/fluenttheir own version, first seen Oct 13, 2026
- YoniChechik/claude-code-configtheir own version, first seen Sep 18, 2026
- cjohnhanson/tisketword for word, first seen Aug 20, 2026
Other versions9 projects use a version of this.
Word for word · 1
- cjohnhanson/tisket★ 1100% same
Edited · 0
None found.
Wrote their own version · 8
- hashicorp/terraform-provider-aws★ 11ksame name
- mrinalwadhwa/fluent★ 125same name
- Arasz/ai-badger★ 3same name
- desbma/agent-skills★ 1same name
- wufei-png/skills★ 1same name
- hossainks/claude-skill★ 0same name
- simu5116/ClaudeCode★ 0same name
- YoniChechik/claude-code-config★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- review-codeSkill
- review-depsSkill
- review-docsSkill
- review-usabilitySkill
- signoff-driverSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 7 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 23
Read a state-machine suite as a design, not a score.
- Very specific: line 25
If testing-strategy is installed, load it.
- Distinctive: line 43
blocking: when its failure would be silent, and minor when a later layer catches it.
“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review: tests
You are one independent check on a change you did not write. Judge the
tests. Read signoff-driver for the sign-off line, the two severity
bands, and the not-applicable case.
Method
Mutation testing carries most of this skill. Remove a guard, run the suite, and name the test that fails. A guard no test kills is untested, whatever the coverage report says.
Two additions. Judge coverage against the change's goal, not its diff. Read a state-machine suite as a design, not a score.
If testing-strategy is installed, load it. Its mutation framework is
this method, and its coverage pass covers criterion 2. This skill differs
in making mutation mandatory for every new guard rather than periodic. Map
its bands as signoff-driver states. If testing-philosophy is
installed, load it for the QA-first stance.
Criteria
- Anchor on the goal. Find the goal the change serves: its issue, its commit message, or its pull-request body. Judge coverage against that goal. A diff-anchored review passes a change that is fully tested and half-built. If no goal is recorded anywhere, say so in the evidence and judge against the commit message. Minor when a goal exists and coverage misses part of it that no user reaches; blocking otherwise.
- Map behavior to tests. List every behavior the change adds or alters. Map each to a named test. A behavior with no test is blocking when its failure would be silent, and minor when a later layer catches it. Name the missing case either way.
- Kill each guard. A guard is a new
if, an early return, a bounds check, or a type check. Remove one, run the suite, and name the test that fails. A guard no test kills is blocking. This catches green tests over an unkilled guard: the suite passes whether the guard is there or not. Removing a guard may stop the code from running at all. Weaken it instead, so the code still runs and the behavior changes. - Grade a state-machine suite as a design. Some repositories test through a suite of named states and transitions rather than unit tests. Where one exists, ask four questions. Do states name real conditions, or numbered steps? Do transitions assert on behavior, or only on output shape? Does each path start from a clean state? Does any path depend on another running first? A suite that passes and answers these badly is blocking for the second and third questions and minor for the others.
- Find the vacuous assertion. An assertion failing for a reason
other than the one it names passes for the wrong reason. It is
blocking over a behavior this change adds, and minor
elsewhere. The common shape is a negated command that errors early:
! tool --flagpasses when--flagdoes not exist. Run each new assertion against a deliberately broken input and confirm it fails for its stated reason. A vacuous assertion over a new behavior is blocking. - Name the residual risk. Say what the suite cannot see. Rank what you name by how often the path runs and whether its failure is silent. This is minor by itself. It becomes the evidence for criterion 2 when a named risk has no test at all.
A change may add more guards than the suite's runtime allows. Say so, state the suite's measured runtime, and sample.
Sample by whether a failure would be silent, never by whether a user reaches the guard. The guards a user never reaches are the error paths, the bounds checks, and the permission checks, and those are exactly the ones whose failure passes unnoticed. Mutate every guard whose removal would produce a wrong result rather than a crash. Name the guards you left and why.
How to run the suite
Find the command the repository documents. Look in its contributing guide, its continuous-integration workflow, and its build manifest, in that order. If no command is discoverable, that is a blocking finding on its own: a suite nobody can run is not coverage.
Worked examples
A guard no test kills
A change adds a uid boundary to a directory walk, so the walk stops at a directory another user owns.
A weak review reads the diff, sees walk_up_stops_at_foreign_owner in the
test file, and passes.
This skill removes the uid check and runs the suite, which stays green: the fixture never creates a foreign-owned directory, so the boundary is untested.
signoff[review-tests] FAIL 4f1c2ab blocking: removing the uid
boundary leaves the suite green; the fixture owns every directoryThe same change, done right
The fixture creates a directory owned by another uid, and the boundary test fails when the check is removed.
signoff[review-tests] PASS 9d0e1f2 removed the uid boundary and
walk_up_stops_at_foreign_owner went red; all four new behaviors map
to named tests; minor: the fixture skips when run as root, and the
skip is silentDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

