Research verification
Skill kyt19970215/codex-skill-evolution-framework/skills/research-verification
Reusable external research and verification workflow for public tools, plugins, dependencies, APIs, OS behavior, browser/driver issues, version compatibility, security/anti-cheat claims, install/upgrade problems, exact error messages, stale claims, and research failure-learning rules. Use when the user asks to search, verify, compare current options, investigate a public project, remember or classify a public-source failure, or when local debugging hits an unclear environment/dependency/tooling issue.From its SKILL.md
npx -y skills add kyt19970215/codex-skill-evolution-framework --skill research-verificationAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 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
3.1 KB, 542 tokens by cl100k_base, as published. Nobody here has run it
Research Verification
Use this skill when a claim or technical behavior may have changed, depends on public tooling, or needs external precedent.
Research Workflow
- Collect minimal local evidence first:
- exact error or claim
- tool/package/project name
- version and OS/runtime
- command or reproduction step
- relevant local config/file names
- Form a narrow search query from public, non-sensitive details.
- Search 1-3 high-relevance sources first.
- Prefer sources in this order:
- official docs
- official changelog/release notes
- GitHub repo, issue, discussion, maintainer comment
- standards/specification
- Stack Overflow or high-signal forum
- blog/video/forum as hypothesis only
- Return to the local environment and verify the selected hypothesis.
- Report the source-backed conclusion, local verification result, and remaining uncertainty.
Search Budget
- Start narrow; do not browse broadly unless the user asks for deeper research.
- For recommendations that could cost significant time or money, verify current options.
- For legal, medical, financial, anti-cheat/security, or account-risk topics, browse by default and phrase cautiously.
Privacy And Safety
- Do not put private code, account names, tokens, full logs, local paths with sensitive names, or proprietary details into search queries.
- Search generic error fragments and public component names.
- Treat forum claims as leads, not facts.
- Never claim "no risk" for anti-cheat, account, security, or compliance topics.
Output Pattern
Use this shape when helpful:
- Conclusion
- Local evidence
- External evidence
- What was verified locally
- Risk / uncertainty
- Next step
Combine With Other Skills
- Combine with
coding-debug-ruleswhen researching a coding/tooling failure. - Combine with
project-rules-routerwhen a project-specific source priority exists. - Combine with the relevant project skill before broad research if the project has preferred sources or terminology.
Failure Learning
- When a public-source, version, API, dependency, browser/driver, install/upgrade, or stale-doc issue causes a failed attempt, capture the exact claim, source used, local version/context, corrected source, and verified outcome.
- Promote only reusable source-selection or verification guardrails, not long search transcripts.
- If the user asks to save the lesson, add it to
references/failure-learning.md.
References
references/source-quality.md: source strength and verification notes.references/failure-learning.md: durable research and version-verification failure shields.
What ships with it: 3 files
1.9 KB alongside SKILL.md
agents/
- openai.yaml255 B
references/
- failure-learning.md1004 B
- source-quality.md704 B