verify-end-to-end
Skill1 starsChanged 2 months agoMITView on GitHub
Prove a change works by running the real app, service, CLI, or job and observing outcomes across boundaries. For explicit end-to-end or runtime proof. Not for UI audits, tests, or diff review.
In this repository21 other files it gives its agents.
- change-production-data-safelySkill
- craft-contextual-uiSkill
- diagnose-root-causesSkill
- engineer-code-changeSkill
- engineer-reliable-data-pipelinesSkill
- engineer-secure-systemsSkill
- evaluate-agent-behaviorSkill
- handle-sensitive-data-safelySkill
- infer-intent-before-codingSkill
- migrate-contracts-safelySkill
- operate-production-safelySkill
- orchestrate-long-running-workSkill
- plan-decision-complete-changeSkill
- ponytailSkill
- publish-verified-github-changeSkill
- release-production-changesSkill
- respond-to-production-incidentsSkill
- review-code-with-evidenceSkill
- route-ultracode-subagentsSkill
- write-durable-documentationSkill
- write-tests-that-prove-behaviorSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Read repository instructions and any launch, seed, and test guidance.
- Distinctive: line 19
Run focused static, build, and automated checks first.
- Distinctive: line 24
Check preserved behavior and an adjacent unaffected path when regression risk warrants it.
- Distinctive: line 34
Never change an assertion, snapshot, seed, retry count, or timeout merely to obtain green.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Verify End to End
Close the gap between source-level checks and what a user, caller, operator, or downstream system actually experiences. A clean diff, build, type check, or unit suite is not end-to-end proof.
Establish the proof contract
- Read repository instructions and any launch, seed, and test guidance.
- Derive observable acceptance conditions from the requested outcome, specification, public contracts, and preserved invariants. Use the diff only to locate affected surfaces, never to weaken or invent conditions.
- Record the environment, starting state, primary changed path, highest-risk failure path, and behavior that must remain unchanged.
- Reuse the project's supported commands, fixtures, accounts, and local services.
Exercise the real surface
- Run focused static, build, and automated checks first.
- Launch the actual application, service, CLI, job, library consumer, or device flow in a safe environment.
- Exercise the primary path with representative input.
- Exercise one relevant failure, boundary, permission, or recovery path.
- Observe outputs and side effects: persistence, events, files, logs, network failures, cleanup, and subsequent reads. Restart or reload when durability matters.
- Check preserved behavior and an adjacent unaffected path when regression risk warrants it.
Trace cross-layer work through registration, configuration, transport, persistence, and consumption. Confirm new artifacts are actually referenced and that no stub or display-only behavior is mistaken for completion.
When the journey crosses a UI and craft-contextual-ui is not engaged, interact at representative viewport widths, inspect the rendered result, and check keyboard access plus loading, empty, and error states. For jobs, integrations, and migrations without their own owner, check repeat runs, partial progress, version skew, and rollback when reachable.
Loop on evidence
When a check fails, preserve the evidence and decide whether it is a product defect, a broken test, a fixture, a dependency, or the harness. A verification-only request does not authorize fixing any of them: report. If fixes are authorized, use diagnose-root-causes and engineer-code-change, then rerun the exact failed path and any earlier evidence the fix could invalidate.
Never change an assertion, snapshot, seed, retry count, or timeout merely to obtain green. Never retry without a new hypothesis or changed condition.
Stop and ask when
- proof requires seeding, resetting, migrating, deploying, or mutating a production, shared, credentialed, billed, or externally visible resource;
- the only available environment cannot represent the real surface;
- an observed failure contradicts an acceptance condition and fixing is not authorized.
Done when
Each material acceptance condition has direct observable evidence and no observed failure contradicts it. Report conditions and results, exact commands or interactions, environment, failures and rechecks, omitted checks with reasons, and residual risk. Never claim end-to-end success unless the runnable outcome was observed.
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.

