agentsclimarketplace

Flow test

Skill maklybae/maklyskills/skills/flow-test

Run the repository's real verification — scoped while iterating, full before declaring anything done — read the output, diagnose failures at the root rather than at the assertion, and record the evidence in the task ledger. Use this skill when the user says "прогони тесты", "проверь что всё зелёное", "run the tests", "does it build", when code has just been written or changed and its state is unknown, and always before claiming that work is complete, fixed, or passing. Prefer it over asserting from memory that something passes: a claim without fresh output in this session is a guess.From its SKILL.md

Install
npx -y skills add maklybae/maklyskills --skill flow-test

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.6 KB, as published. Nobody here has run it

Flow test

Two rhythms

Narrow while iterating. Run the package or module you just touched, as often as you like. Fast feedback is what keeps a mistake one edit old instead of ten.

Broad before declaring done. Once, at the end, run what the profile's ## Commands calls the full verification — plus lint and build if they are separate. Scoped runs miss exactly the failures that matter most: the ones your change caused somewhere you were not looking.

Where the language has a race detector or equivalent sanitiser, use it — in the scoped runs too, not only the final one. A race caught at the stage that introduced it is a one-edit fix; caught in the closing sweep, it surfaces after other code has been built on top of it. Races are invisible without the detector, and in some toolchains a single race takes down the whole test binary and reports as unrelated neighbouring failures.

Evidence

No claim without output produced in this session. "Tests should pass now" is not a status; it is a hypothesis with the experiment skipped. Say what you ran and what came back — counts, not adjectives.

This is not ceremony. The failure mode it prevents is specific and common: code is changed, the change looks obviously right, the claim is made, and the suite was never run because the last run was four edits ago.

Diagnosing a failure

Default assumption: the code is wrong, not the test. A test is a recorded expectation, usually written when the behaviour was clearer than it is now.

Changing an assertion to match new behaviour is legitimate — the requirement may genuinely have changed — but only when you can say why the old expectation was wrong. Write that sentence in the ledger. If you cannot write it, you are editing the test to make the failure go away, which converts a caught bug into a shipped one.

An intermittent failure is a finding, not noise. Re-running until green hides a real defect that happens to be rare; concurrency bugs almost always present this way first.

"Pre-existing" is a claim that needs evidence. When something fails in code you did not touch, check whether your change reaches it before saying so — the same one line proves or disproves it. If it really is pre-existing, note it in the ledger and leave it alone; fixing it silently inside this change makes the diff harder to review.

Record it

Append to ## Verification in the ledger — ~/.claude/projects/<project-root-as-dashes>/flow/<task-slug>/ledger.md, schema in ../flow/references/ledger.md relative to this skill's directory — the date, the exact command, and the result. Later stages and review subagents read those lines to know what has actually been proven, and a stale one is worse than none.

Not in this stage

Adding coverage belongs to flow-implement, deleting worthless tests belongs to flow-cleanup. This stage runs what exists and repairs what it breaks — if the run reveals a gap, note it as a finding rather than expanding scope here.

Keep looking

Skills are one crate of 608,475. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.