test
Skill0 starsChanged 8 days agoMITView on GitHub
Create meaningful regression coverage, audit weak or redundant tests, and optimize test lanes using measured cost and change impact. Use for test creation, suite audits, failing-test investigation, or CI test optimization; browser journey QA stays with qa/browse.
185 other projects use a version of this.
Adopted by
Most recently taken up by
- br3eze-code/br3eze-codetheir own version, first seen Nov 16, 2026
- vericontext/vibeframetheir own version, first seen Nov 15, 2026
- TJaenichen/claude-code-harnesstheir own version, first seen Nov 15, 2026
- Renozoic-Foundry/forge-publictheir own version, first seen Nov 15, 2026
- nazmirket/skillstheir own version, first seen Nov 15, 2026
- RohiRIK/skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use agent-context-isolation, agentica-infrastructure and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
parcadeiSkillagent-context-isolation
Agent Context Isolation
Also in 3 projects that use this
parcadeiSkillagentica-infrastructure
Reference guide for Agentica multi-agent infrastructure APIs
Also in 3 projects that use this
- +10 more from parcadei
Other versions185 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 185
- facebook/react★ 253ksame name
- react/react★ 252ksame name
- apache/skywalking★ 25ksame name
- livewire/livewire★ 24ksame name
- NVIDIA/TensorRT★ 13ksame name
- FalkorDB/FalkorDB★ 11ksame name
- FastLED/FastLED★ 7.5ksame name
- fleetdm/fleet★ 7.2ksame name
- parcadei/Continuous-Claude★ 3.9ksame name
- parcadei/Continuous-Claude-v2★ 3.9ksame name
175 more not listed
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 repository27 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- agent-first-docsSkill
- analytics-reviewSkill
- autopilotSkill
- browseSkill
- create-prSkill
- design-uiSkill
- foremanSkill
- harness-auditSkill
- image-genSkill
- implementSkill
- investigateSkill
- marketing-videoSkill
- plan-draftSkill
- qaSkill
- reconSkill
- release-prdSkill
- releaseSkill
- review-prSkill
- semantic-decisionSkill
- setup-projectSkill
- teamSkill
- telemetry-reviewSkill
- token-auditSkill
- triageSkill
- ui-auditSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 9 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 158 edited versions we read keep these.
9
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Follow preamble for bootstrap, telemetry and standalone/ticket handling, and the Auto-Decision Framework for decisions.
- Distinctive: line 20
Read the changed behavior, its callers and existing tests.
- Very specific: line 27
Use semantic-decision kind testcase when choosing affected existing cases.
- Distinctive: line 41
Run relevant existing tests first.
- Distinctive: line 43
Import and exercise production code.
- Distinctive: line 50
Never mutate a user's dirty source to manufacture evidence.
- Distinctive: line 54
Run the affected lane plus required checks.
- Very specific: line 63
Never weaken assertions, bless snapshots, catch errors into passing defaults, comment tests out, add .skip/.todo, broaden excludes, use || …
- Distinctive: line 76
Look for copied implementations, self-derived oracles, empty fixtures, conditional assertions that never execute, swallowed failures, mocks …
- Distinctive: line 93
Read measurement and lanes.
- Distinctive: line 106
Do not create branches, commit or push.
“Common” is measured: kept by at least 6 in 10 of the 158 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
test
Prove behavior at the cheapest layer that can actually observe the failure. A smaller suite is useful only if it retains its ability to catch regressions.
Modes: create [scope], audit [scope], optimize [scope]. With no mode,
infer it from the request; ordinary implementation uses create scoped to the
change, not a whole-project audit.
Follow preamble for bootstrap, telemetry and
standalone/ticket handling, and the
Auto-Decision Framework for decisions.
Shared refs are filesystem paths beside this skill's directory, so read them
by path, not as skill://.
Before adding a test
Read the changed behavior, its callers and existing tests. Answer briefly:
- What observable behavior changes, and what requirement defines the answer?
- What plausible input, state, failure or timing would break it?
- Which existing tests already distinguish correct from broken behavior?
- What is the lowest credible layer that observes that failure?
- Is another test necessary, or does existing evidence suffice?
Use semantic-decision kind testcase when
choosing affected existing cases; supply the behavior/caller map and evidence.
Preserve mandatory repository checks and regression reproducers regardless of
that choice. No ticket, provider or orchestration setup is needed to reason
about tests; use the calling LLM fallback if the decision CLI is unavailable.
Pure rules often need unit tests; SQL predicates/locks need real disposable DB
coverage; HTTP authorization needs the actual middleware/handler boundary;
React effects need a mounted hook/component; browser APIs or user journeys may
need browse/qa. A mock of the mechanism under test cannot prove it works.
create
- Scout the package scripts, runner configuration, fixtures and existing coverage before choosing a new framework or dependency.
- Run relevant existing tests first. Add only missing behavior, boundaries and realistic failures. A docs-only or already-covered change may need no new test.
- Import and exercise production code. Keep expected results independent of the calculation being checked. Do not transcribe the implementation into a test-only model, mock the subject itself, or assert merely that a mock returns what the fixture supplied. A model is useful only when compared to production.
- Tie each new assertion to a named regression. For a bug fix, demonstrate failure before the fix when feasible; a bounded mutation in a disposable copy is another option. Verify the intended assertion failed, not setup/imports. Never mutate a user's dirty source to manufacture evidence.
- Stub external transport, not the contract being tested. Restore global state, clocks and resources. If tests pass alone but fail together, investigate runner isolation and fixture ownership before touching expectations.
- Run the affected lane plus required checks. Report what those tests cannot prove (e.g. DOM emulation is not browser playback or layout).
When a test fails
Reproduce with the exact command and capture the failure. Classify it from evidence: product regression, invalid fixture/expectation, isolation leak, missing infrastructure, or an existing defect. Check alone and with neighbors when order matters. Repair the responsible layer and rerun it.
Never weaken assertions, bless snapshots, catch errors into passing defaults,
comment tests out, add .skip/.todo, broaden excludes, use || true, or raise
retries/timeouts merely to obtain green. An intentional behavior change can
require a new expectation, but cite the requirement and retain the invariant
it replaces. Skipping/quarantining a test requires explicit user authorization;
record the reason and restoration condition. Existing skips remain visible in
the report and never count as passed coverage. Missing required infra is
NOT RUN/BLOCKED, not a passing lane. Existing failures are investigated and
reported; calling one "unrelated" does not discharge it or authorize disabling it.
audit
Inventory collected files/cases, scripts and CI/gates, then inspect candidates in production context. A regex hit or high assertion count is only a lead. Look for copied implementations, self-derived oracles, empty fixtures, conditional assertions that never execute, swallowed failures, mocks that ignore predicates, source-string checks claimed as runtime coverage, duplicate cases, stale skips and tests absent from all documented runners.
For each finding record: file/test, claimed behavior, concrete defect it fails to catch, evidence, and keep/rewrite/merge/delete with rationale. Architectural source guards and fixed protocol constants can be legitimate contracts; don't delete them just because they read source or assert literals.
Audit is read-only unless the request also authorizes cleanup/fixes. When changing the suite, map every removed behavior to a retained/replacement test, or explain why it was never meaningful. Replace weak coverage of a real risk before removing it. Fewer tests and a green run alone prove neither coverage nor correctness. Do not delete a failing test as a substitute for investigating.
optimize
Read measurement and lanes. Measure first; do not assume the biggest file, most assertions or most cases dominate cost. Preserve full-suite stable-release checks. For routine pushes/PRs, select affected lanes from the behavior/caller map plus mandatory checks; do not default to the whole suite when a validated impact map suffices. Never call a faster run with missing coverage an optimization.
Output
Return a concise behavior → failure → existing/new evidence → lane map. Include exact commands and results, collection/pass/fail/skip/not-run counts when available, prerequisites, and unresolved risks. Audit findings and optimization measurements must separate observed facts from estimates. Publish the report through the caller's handoff if a ticket exists; otherwise return it directly. Do not create branches, commit or push. CI lane changes stay within the requested optimization scope; deployment and release authority stay with the caller.
STATUS: DONE | DONE_WITH_CONCERNS | NEEDS_CONTEXT | BLOCKED
VERDICT: TESTED | AUDITED | OPTIMIZED | FAIL
SUMMARY: <behavior covered, meaningful removals, checks and remaining gaps>
NEXT: <remaining action or none>TESTED means the selected checks ran and passed, not that unrun lanes passed.
An audit can finish with findings; unresolved failures prevent a test/release
PASS. Claim OPTIMIZED only with comparable measurements and retained coverage.
Does 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.

