agentleFS
unclejobs-aiunclejobs-ai/second-claude-code/skills/research/SKILL.md

research

Skill41 starsChanged 7 days agoMITView on GitHub

Use when finding sources through iterative web exploration and synthesis

  • Reads credentials

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use daily-arxiv, exp-eval 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.
  • LambenthanSkill

    daily-arxiv

    每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新

    Also in 5 projects that use this

  • LambenthanSkill

    exp-eval

    实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges

    Also in 5 projects that use this

  • truongtv22Skill

    Problem-Solving Techniques

    Apply systematic problem-solving techniques for complexity spirals (simplification cascades), innovation blocks (collision-zone thinking), recurring patterns (meta-pattern recognition), assumption constraints (inversion exercise), scale uncertainty (scale game), and dispatch when stuck. Techniques derived from Microsoft Amplifier project patterns adapted for immediate application.

    Also in 5 projects that use this

  • liufeng1976Skill

    setup-ts-deep-modules

    Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.

    Also in 43 projects that use this

  • skyllwtSkill

    exp-pilot-eval

    Pilot result evaluation — read pilot results, apply success criteria, update idea page (pilot_result, failure_reason if failed), generate PILOT_VERDICT_REPORT. Called by /ideate Phase 5 after /exp-pilot-run.

    Also in 4 projects that use this

  • skyllwtSkill

    exp-pilot-run

    Pilot experiment execution — read Pilot Spec YAML, write pilot code, run experiment(Confirm with the user before operation and require the applicant to conduct manual inspection), return results. Called by /ideate Phase 5. Does NOT modify wiki pages or judge pass/fail.

    Also in 4 projects that use this

  • +5 more from Lambenthan
  • +1 more from skyllwt

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 2 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 169 edited versions we read keep these.

  • 2

    Distinctive

    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
  1. Very specific:

    When mmbridge is detected (see references/mmbridge-integration.md), dispatch mmbridge research in parallel with the internal researcher.

    line 38
  2. Distinctive:

    Verify actual search counts before output.

    line 65
  3. Very specific:

    When --engine legacy is set or $JINA_API_KEY is unavailable, replace "Jina Search" with "WebSearch" and "Jina Reader" with "WebFetch" in …

    line 83
  4. Very specific:

    news: Add after:30d and prefer established editorial sources over blogs.

    line 91
  5. Very specific:

    When --engine legacy or $JINA_API_KEY is unavailable:

    line 102
  6. Very specific:

    When --interactive is set, skip Jina/WebFetch and use Playwright for every URL.

    line 107
  7. Distinctive:

    Write the full brief using the Write tool and tell the user the saved path.

    line 117

“Common” is measured: kept by at least 6 in 10 of the 169 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Resolve runtime paths before operations.

Iron Law

No guessing. If there is no data, collect it first.

Red Flags

  • "This one article covers everything" → STOP. Single-source = blind spots. Depth minimums (3/5/10) exist for a reason.
  • "I already know the answer, just need a citation" → STOP. Confirmation bias produces cherry-picked sources.
  • "Source is from 2022 but topic hasn't changed" → STOP. Staleness must be verified, not assumed.
  • "Skip the analyst step, just list links" → STOP. Raw link dumps without gap analysis are rejected.
  • "Page was blocked so I'll summarize from the URL title" → STOP. Use fallback chain (Jina → Playwright → Gaps), never hallucinate from metadata.

Research

Autonomous multi-round web research that produces structured Research Briefs.

When to Use

  • User asks to research a topic, trend, or question
  • Another skill (e.g., /scc:write, /scc:analyze) needs source material
  • User wants competitive intelligence, market data, or literature review

Web Engine

The researcher uses Jina Search (s.jina.ai) as the primary tool. Fallback chain: Jina Search → WebSearch + WebFetch → the unblock chain (/scc:unblock, zero-key 9-phase) → Playwright. See references/jina-guide.md and docs/skills/unblock.md.

MMBridge Enhancement

When mmbridge is detected (see references/mmbridge-integration.md), dispatch mmbridge research in parallel with the internal researcher.

Dispatch

At Step 1, also run via Bash:

mmbridge research "<topic>" --type <type> --json > /tmp/mmbridge-research-${RUN_ID}.json
  • --type code-aware: when topic relates to the current codebase
  • --type open: for general topics unrelated to code
  • --depth shallow: skip mmbridge
  • --depth medium|deep: mmbridge enabled

Merge

At Step 3, parse the mmbridge JSON output file as supplemental source material. Count each distinct cited URL from mmbridge as one source and deduplicate overlaps during gap analysis.

Step-by-Step

  1. Auto-load references/research-methodology.md and references/jina-guide.md.
  2. Dispatch researcher (sonnet): run depth-appropriate Jina Search calls with varied query phrasings. If $JINA_API_KEY is unavailable, fall back to WebSearch + WebFetch. If mmbridge is available at medium or deep depth, dispatch it in parallel.
  3. Validate sources: reject login walls, blocked pages, empty content. For blocked URLs, run the unblock chain (node "<plugin-root>/skills/unblock/engine/cli.mjs" "<URL>" --json); read the trace before any retry.
  4. Dispatch analyst (sonnet): merge internal and mmbridge findings, identify gaps, and apply conflict-resolution rules from references/research-methodology.md.
  5. Optional second round: only when depth allows and the analyst found critical gaps.
  6. Dispatch writer (sonnet): synthesize the final brief with conflict annotations.
  7. Verify actual search counts before output. If counts exceed the depth cap, discard excess results and re-synthesize.

Options

FlagValuesDefaultEffect
--depthshallow|medium|deepmediumHARD limit on search rounds — see Depth Behavior
--sourcesweb|academic|newswebConstrains search domain
--langko|en|autoautoOutput language. When called from another skill (write, analyze), inherits the caller's --lang value.
--interactiveflagoffForce Playwright for all URL fetches (useful for SPAs, dashboards)
--enginejina|legacyjinajina: use Jina Search API. legacy: use WebSearch + WebFetch

Depth Behavior

  • shallow: EXACTLY 3 Jina Search calls. No deep reads. No gap analysis round.
  • medium: EXACTLY 5 Jina Search calls plus up to 2 Jina Reader deep reads.
  • deep: 10+ Jina Search calls, unlimited Jina Reader deep reads, and at most 3 gap-fill rounds before documenting remaining gaps.

When --engine legacy is set or $JINA_API_KEY is unavailable, replace "Jina Search" with "WebSearch" and "Jina Reader" with "WebFetch" in the above counts.

Coverage requirements and conflict resolution rules are in references/research-methodology.md.

Source Domain (--sources)

  • web (default): General web search. No domain restrictions.
  • academic: Bias at least half the calls toward Google Scholar or arXiv.
  • news: Add after:30d and prefer established editorial sources over blogs.

Fallback Chain

Jina Search (s.jina.ai) ─── primary: search + content in one call
  └─ blocked/empty → Jina Reader (r.jina.ai) with X-Engine: browser
       └─ still fails → Playwright (browser_navigate + browser_snapshot)
            └─ unavailable → note in Gaps & Limitations

When --engine legacy or $JINA_API_KEY is unavailable:

WebSearch → WebFetch → Playwright → Gaps & Limitations

When --interactive is set, skip Jina/WebFetch and use Playwright for every URL.

Cost controls: Max 3 Playwright navigations per round. Exceeding the cap means skip the rest and note the gap.

Graceful degradation: Both Jina and Playwright are optional. If $JINA_API_KEY is missing, fall back silently. If Playwright MCP is unavailable, proceed without it and note unreachable URLs.

Auto-Save

  • Path: .captures/research-{slug}-{YYYY-MM-DD}.md
  • {slug} = topic lowercased, spaces to hyphens, max 40 chars
  • Write the full brief using the Write tool and tell the user the saved path.

Gotchas

Failure ModeMitigation
Stops after 1 searchresearcher MUST meet depth minimums: shallow=3, medium=5, deep=10. Fewer = restart.
Lists links without analysisanalyst step is required. Raw link dumps are rejected.
Hallucinated sourcesEvery URL must come from an actual Jina Search/WebSearch result. writer cannot invent URLs.
Duplicate queriesresearcher must vary phrasing with synonyms and different angles per query.
Jina API key missingFall back to WebSearch + WebFetch silently. Do not error.

Subagents

researcher: { model: sonnet, tools: [Bash, WebSearch, WebFetch, browser_navigate, browser_snapshot], constraint: "Jina Search when $JINA_API_KEY set, else WebSearch+WebFetch; meet depth minimums, vary phrasing, validate content, flag staleness; Playwright only when Jina+WebFetch both fail; max 3 navigations" }
analyst: { model: sonnet, tools: [], constraint: "produce gap list, flag data conflicts, verify coverage requirements" }
writer: { model: sonnet, tools: [], constraint: "every claim needs a source, no invented URLs, include conflict annotations" }

Output & Integration

See references/research-methodology.md for output format template, extended gotchas table, and integration notes.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.