vibe-coding-rescue
Skill0 starsChanged 2 months agoMITView on GitHub
Use when an AI-built, Vibe Coding, or inherited software project cannot start, build, test, validate, run locally, pass CI, or match its README; also use when environment, dependency, API contract, database migration, test fixture, deployment, or fake-completion failures must be diagnosed before repair.
- Reads credentials
- Commits and pushes
In this repository42 other files it gives its agents.
- ai-pr-redteamSkill
- ai-servicesSkill
- build-work-memorySkill
- ceo-calendar-inviteSkill
- ceo-document-reviewSkill
- ceo-email-classifierSkill
- ceo-feedback-iterationSkill
- ceo-feedback-processingSkill
- ceo-mail-reviewSkill
- ceo-meeting-workSkill
- ceo-message-triageSkill
- ceo-minutes-syncSkill
- ceo-personnel-communicationSkill
- ceo-wechatSkill
- ceo-work-trackingSkill
- codex-model-capacity-retrySkill
- daily-frontier-tech-discoverySkill
- dengbao-code-auditSkill
- dingtalk-browser-exportSkill
- dingtalk-knowledge-organizeSkill
- dingtalk-minutes-access-requestSkill
- dingtalk-oa-approvalSkill
- dingtang-okr-reviewSkill
- fundflow-investor-meetingSkill
- fxiaoke-crm-cliSkill
- internal-app-standardsSkill
- ocrSkill
- pm-data-adaptation-assistantSkill
- project-delivery-document-collectionSkill
- qa-generated-test-caseSkill
- senior-technical-product-evaluationSkill
- spec-intakeSkill
- sso-oidc-onboardingSkill
- stardust-interviewSkill
- production-devops-sreSkill
- stardust-ttsSkill
- startask-io-contract-auditSkill
- task-managementSkill
- transcribeSkill
- veyra-timesheetSkill
- vibe-codingSkill
- video-transcribeSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
17 rules. About 0 are common, 12 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈12
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
Use vibe-coding for formal Spec-to-code delivery from an approved engineering-ready Spec.
- Very specific: line 13
Use internal-app-standards for broad stack selection, architecture design, or internal-system production review.
- Very specific: line 14
Use dengbao-code-audit for MLPS/等保 security audit.
- Distinctive: line 15
Use this skill first when the immediate problem is rescue: startup failure, red tests, broken README, CI failure, environment drift, …
- Distinctive: line 20
Read the README, manifests, lockfiles, env templates, and the user-reported failing command or area.
- Very specific: line 21
Run the read-only context collector when possible.
- Distinctive: line 27
Read the CI/deploy files, migrations, tests, generated contracts, and code paths indicated by the collector and failure report.
- Distinctive: line 29
If the command would mutate production, require credentials, or call an external system, stop and state the safe substitute or missing …
- Distinctive: line 33
Read references/failure-taxonomy.md when classifying install, startup, build, test, API, database, CI/deploy, README drift, or AI …
- Distinctive: line 34
Read references/rescue-workflow.md before changing files or deciding the repair sequence.
- Distinctive: line 35
Read references/output-contract.md before reporting a fix, partial fix, or blocker.
- Distinctive: line 39
Do not edit before baseline reproduction unless the baseline command is unsafe.
- Very specific: line 43
Never copy secrets, private URLs, tokens, cookies, database contents, .env values, or local browser state into reports, commits, fixtures, …
- Distinctive: line 44
Do not claim "fixed", "green", "newcomer-ready", "CI-ready", or "deployable" without fresh command evidence.
- Distinctive: line 45
Do not claim "newcomer-ready" unless the README path was verified from a clean checkout/container or an equivalent isolated state with only …
- Distinctive: line 46
When Git push is requested, prefer an isolated branch unless the user explicitly requested direct push and repository policy permits it.
- Distinctive: line 47
If remote access, credentials, private packages, databases, external services, or production permissions are missing, report a blocker with …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Vibe Coding Rescue
Recover an AI-built or inherited project to a known-good engineering baseline. The job is not to make a confident guess; it is to reproduce the real failure, identify the first broken boundary, make the smallest justified repair, and prove the original failure no longer reproduces.
Use This Instead Of
- Use
vibe-codingfor formal Spec-to-code delivery from an approved engineering-ready Spec. - Use
internal-app-standardsfor broad stack selection, architecture design, or internal-system production review. - Use
dengbao-code-auditfor MLPS/等保 security audit. - Use this skill first when the immediate problem is rescue: startup failure, red tests, broken README, CI failure, environment drift, contract mismatch, migration failure, or AI-generated fake completion.
Required Start
- Protect the worktree: run
git status --short --branchand identify user changes before editing. - Read the README, manifests, lockfiles, env templates, and the user-reported failing command or area.
Run the read-only context collector when possible. Write its JSON outside the target repo, such as
/tmpor a separate work directory, unless the user explicitly asks for an artifact in the repo:python3 <skill-dir>/scripts/collect_rescue_context.py <project-root> --output <workdir>/rescue-context.json- Read the CI/deploy files, migrations, tests, generated contracts, and code paths indicated by the collector and failure report.
- Reproduce the reported failure with the exact command or the closest safe substitute. Record command, exit code, and key log lines before changing code.
- If the command would mutate production, require credentials, or call an external system, stop and state the safe substitute or missing authorization.
Reference Routing
- Read references/failure-taxonomy.md when classifying install, startup, build, test, API, database, CI/deploy, README drift, or AI fake-completion failures.
- Read references/rescue-workflow.md before changing files or deciding the repair sequence.
- Read references/output-contract.md before reporting a fix, partial fix, or blocker.
Hard Rules
- Do not edit before baseline reproduction unless the baseline command is unsafe; then document why and use a dry-run, local-only, or synthetic substitute.
- Fix one proven failure chain at a time. Do not batch dependency upgrades, framework swaps, format churn, or broad refactors into a rescue.
- Separate environment/config, dependency, contract, code, test, data, documentation, and deployment causes. Do not label a test or script broken until upstream configuration and contract evidence has been checked.
- For behavior changes, use a failing test first when feasible. For config/docs/runbook fixes, prove them with the command they are supposed to repair.
- Never copy secrets, private URLs, tokens, cookies, database contents,
.envvalues, or local browser state into reports, commits, fixtures, or logs. - Do not claim "fixed", "green", "newcomer-ready", "CI-ready", or "deployable" without fresh command evidence.
- Do not claim "newcomer-ready" unless the README path was verified from a clean checkout/container or an equivalent isolated state with only documented setup inputs.
- When Git push is requested, prefer an isolated branch unless the user explicitly requested direct push and repository policy permits it. Never mix user changes into the rescue commit.
- If remote access, credentials, private packages, databases, external services, or production permissions are missing, report a blocker with the exact missing artifact and the last safe local evidence.
Expected Output
Lead with the rescue verdict: fixed, partially fixed, or blocked.
Then provide:
- Scope and original failure command.
- Failure chain: symptom -> failing boundary -> root cause -> repair.
- Files changed and why.
- Verification commands actually run, with exit status and remaining failures.
- README/newcomer runbook status.
- Remaining risks, skipped unsafe checks, and required owner action.
- Git commit, branch, PR, or push status when delivery was requested.
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.

