agentsclimarketplace

Source grounded research

Skill Dokuganrryu/ai-agent-skills-and-subagents/skills/source-grounded-research

Portable AI agent skills and specialist subagents for prompt enhancement, workspace resume, source-grounded research, and release readiness.

Install
npx -y skills add Dokuganrryu/ai-agent-skills-and-subagents --skill source-grounded-research

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.
  • 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.

What its author says it does

Copied from the file, not written here

Use when the user asks for a current, evidence-backed research brief, tool/vendor evaluation, competitor/source scan, option ranking, or "is this real/good/clickbait" judgment that should combine local artifacts with current external sources and separate observation from inference.

SKILL.md

3.2 KB, as published. Nobody here has run it

Source-Grounded Research

Use this skill when the task is a bounded research or decision brief where stale facts, generic advice, or weak sourcing would make the answer worse.

Success Criteria

  • The research question or decision is explicit.
  • Important claims are tied to local evidence, current primary sources, or clearly labeled inference.
  • The output is compact enough to act on.
  • Any unverified or high-risk area is named instead of glossed over.

Workflow

  1. Define the decision

    • Restate the exact question, user goal, and stopping condition.
    • If the request is broad, narrow it to the smallest useful decision.

  2. Check local context first

    • Look for provided files, current repo docs, task ledgers, prior handoffs, configs, logs, screenshots, and memories.
    • Treat local artifacts as evidence about the user's project, not as proof of current external reality.

  3. Use current external sources for unstable claims

    • Browse for anything current, external, disputed, platform-specific, legal, financial, security-sensitive, pricing-related, or likely to have changed.
    • Prefer official docs, vendor pages, standards, public repos, release notes, and live pages.
    • Use third-party posts only as supporting context unless they contain directly verifiable data.

  4. Keep source classes separate

    • Local observation: files, logs, configs, runtime behavior, screenshots.
    • Current primary source: official docs, live vendor pages, public source code, authoritative records.
    • Secondary source: articles, forum posts, reviews, benchmarks, summaries.
    • Inference: your judgment from the above.

  5. Compare fit and risk

    • Explain what transfers to the user's exact project or device, and what may not.
    • Call out setup cost, maintenance burden, lock-in, safety, privacy, and data-quality risks when relevant.
    • Do not recommend risky live actions unless the user explicitly asks and the safe path is clear.

  6. Answer in the smallest useful shape

    • Lead with the recommendation or ranking.
    • Include a short evidence table only when it improves trust.
    • End with what is still unverified and the next validation step.

Optional Subagent

Spawn the evidence-scout custom agent when the research has many source surfaces or needs parallel discovery. Give it a narrow question, source boundaries, and output format. Do not let it make the final recommendation by itself.

Output Template

**Recommendation**
...

**Evidence**
| Claim | Source type | Evidence | Confidence |
| --- | --- | --- | --- |

**Fit For This Case**
...

**Unverified**
...

**Next Check**
...

Common Traps

  • Starting with a generic ecosystem tour before checking the user's files.
  • Treating memory or old notes as current external truth.
  • Hiding uncertainty inside confident prose.
  • Overfitting from a similar product, device, theme, repo, or market.
  • Creating a long bibliography when the user needs a decision.

Keep looking

Skills are one crate of 328,083. 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.