agentleFS
joris887joris887/exosuit/.claude/skills/research/SKILL.md

research

Skill9 starsChanged 3 months agoMITView on GitHub

Deep online research on any topic. Produces structured reports with citations and confidence scores. Three depth modes — quick scan, standard investigation, or deep dive.

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

Show its rules and text

We split this skill into the rules it gives an agent

28 rules. 0 are common, 25 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 170 edited versions we read keep these.

  • 25

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before starting research, assess whether the query is specific enough:

    line 26
  2. Distinctive:

    Ask 1–2 focused clarifying questions using AskUserQuestion

    line 29
  3. Distinctive:

    Ask about scope: "What specific aspect of [topic] are you interested in?"

    line 30
  4. Distinctive:

    Ask about context: "Is this for a specific project/technology, or a general investigation?"

    line 31
  5. Distinctive:

    Ask about depth: if not obvious from the query:

    line 35
  6. Distinctive:

    Use AskUserQuestion with options: "Quick scan (2-3 searches, fast answer)", "Standard (parallel research, 2-3 min)", "Deep dive …

    line 36
  7. Distinctive:

    If the query clearly implies depth (e.g., "comprehensive analysis of..." → deep), skip asking

    line 37
  8. Distinctive:

    Before doing new research, check if this topic has been investigated before:

    line 41
  9. Very specific:

    Search docs/research/ for matching topics (Grep on title/tags in frontmatter)

    line 43
  10. Very specific:

    Search docs/solutions/ for related findings

    line 44
  11. Very specific:

    Search docs/brainstorms/ for related design explorations

    line 45
  12. Distinctive:

    If prior research exists:

    line 47
  13. Distinctive:

    If it's recent and relevant: present it to the user, ask if they want fresh research or if prior findings suffice

    line 49
  14. Distinctive:

    If it's outdated or partially relevant: note it as context for the new research

    line 50
  15. Distinctive:

    Do NOT dispatch research agents until the user approves the research plan.

    line 74
  16. Distinctive:

    After plan approval, execute using the deep-research methodology:

    line 79
  17. Distinctive:

    After dispatch: "Researching [N] sub-questions in parallel..."

    line 88
  18. Distinctive:

    After collection: "Collected findings from [N] sub-questions.

    line 89
  19. Distinctive:

    Ask user: "These areas have low confidence: [gaps].

    line 102
  20. Distinctive:

    If yes: apply Phase 4 (Deepen) from deep-research methodology — targeted follow-up research

    line 103
  21. Distinctive:

    If no: proceed with current findings, noting confidence gaps

    line 104
  22. Distinctive:

    NEVER skip the clarification gate for vague queries — bad input produces bad research

    line 160
  23. Distinctive:

    NEVER present training-data knowledge as research findings — all claims must come from web sources

    line 161
  24. Distinctive:

    NEVER dispatch research without user approval of the plan

    line 162
  25. Distinctive:

    ALWAYS save research reports for future reference (they prevent redundant research)

    line 163
  26. Distinctive:

    ALWAYS show the executive summary inline — don't make users open the file to see results

    line 164
  27. Distinctive:

    ALWAYS check for prior research before starting new research

    line 165
  28. Distinctive:

    Keep the research plan visible and the user in control of depth and scope

    line 166

“Common” is measured: kept by at least 6 in 10 of the 170 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

research

Skill metrics: Emit a start event to the activity log:

echo "{\"type\":\"skill\",\"event\":\"start\",\"skill\":\"research\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" >> docs/sessions/.activity-log.jsonl

Researching: $ARGUMENTS

Phase 1: Clarification Gate

Before starting research, assess whether the query is specific enough:

If the query is vague (single word, no clear question, ambiguous scope):

  • Ask 1–2 focused clarifying questions using AskUserQuestion
  • Ask about scope: "What specific aspect of [topic] are you interested in?"
  • Ask about context: "Is this for a specific project/technology, or a general investigation?"

If the query is specific enough — proceed directly.

Ask about depth if not obvious from the query:

  • Use AskUserQuestion with options: "Quick scan (2-3 searches, fast answer)", "Standard (parallel research, 2-3 min)", "Deep dive (comprehensive, 5-10 min)"
  • If the query clearly implies depth (e.g., "comprehensive analysis of..." → deep), skip asking

Phase 2: Check Prior Research

Before doing new research, check if this topic has been investigated before:

  1. Search docs/research/ for matching topics (Grep on title/tags in frontmatter)
  2. Search docs/solutions/ for related findings
  3. Search docs/brainstorms/ for related design explorations

If prior research exists:

  • Load it and assess: is it still current? Does it answer the query?
  • If it's recent and relevant: present it to the user, ask if they want fresh research or if prior findings suffice
  • If it's outdated or partially relevant: note it as context for the new research

Phase 3: Plan Research

Compose the deep-research methodology (.claude/prompts/deep-research.md):

  1. Classify depth — Auto-classify using depth profile heuristics, or use the user's explicit choice
  2. Decompose query — Break into sub-questions per Phase 1 of deep-research methodology
  3. Present plan to user — Show the sub-questions and depth before executing

Present the research plan:

Research plan for: [query]
Depth: [quick/standard/deep]

Sub-questions:
1. [sub-question 1] — [web/code/hybrid]
2. [sub-question 2] — [web/code/hybrid]
3. [sub-question 3] — [web/code/hybrid]

Estimated: [N] searches, [time estimate]
<HARD-GATE> Do NOT dispatch research agents until the user approves the research plan. The user may want to adjust sub-questions, change depth, or add specific areas to investigate. </HARD-GATE>

Phase 4: Execute Research

After plan approval, execute using the deep-research methodology:

  1. Quick depth: Execute inline — search sequentially, no subagent dispatch
  2. Standard/Deep depth: Dispatch parallel subagents per sub-question using the Agent tool
    • Each subagent is a research-analyst (general-purpose agent with the research-analyst prompt)
    • Each receives: sub-question, search strategy, tool budget, output format instructions
    • Launch ALL subagents in a single message for parallel execution

Communicate progress to the user between phases:

  • After dispatch: "Researching [N] sub-questions in parallel..."
  • After collection: "Collected findings from [N] sub-questions. Synthesizing..."

Phase 5: Synthesize & Evaluate

Apply Phase 3 (Synthesize) from the deep-research methodology:

  1. Merge all subagent reflections into a coherent answer
  2. Resolve contradictions explicitly
  3. Assess overall confidence (weighted average of sub-question confidences)
  4. Compile knowledge gaps

If confidence is below threshold (70 for standard, 80 for deep):

  • Present initial findings with low-confidence areas highlighted
  • Ask user: "These areas have low confidence: [gaps]. Want me to dig deeper on any of them?"
  • If yes: apply Phase 4 (Deepen) from deep-research methodology — targeted follow-up research
  • If no: proceed with current findings, noting confidence gaps

Phase 6: Output & Persist

  1. Generate report using the research-report format from references/research-report-template.md
  2. Save to docs/research/<topic-slug>.md with full YAML frontmatter
  3. Present key findings to the user — don't just say "report saved", show the executive summary and top findings inline

Topic Slug Generation

  • Lowercase the topic
  • Replace spaces with hyphens
  • Remove special characters
  • Truncate to 50 characters
  • Example: "Modern auth token rotation" → modern-auth-token-rotation

Phase 7: Cross-Reference & Next Steps

After presenting findings:

  1. Check backlog impact — Search docs/reference/BACKLOG_INDEX.md for stories that might be affected by these findings
  2. Check solution overlap — Search docs/solutions/ for existing learnings that should be updated
  3. Check ADR relevance — Search docs/adr/ for accepted decisions about the researched topic. If the research contradicts an accepted ADR, flag it explicitly ("Research suggests X, but ADR-NNNN chose Y because Z — consider whether to supersede"). If the research validates an accepted ADR, note the confirmation.
  4. Suggest next steps based on findings:
    • "These findings suggest a brainstorm: /brainstorm [topic]"
    • "This could become stories: /ideate [idea based on findings]"
    • "A spike story would help validate: [specific uncertainty]"
    • "Findings are informational only — no action needed"

Skill metrics: Emit a completion event:

echo "{\"type\":\"skill\",\"event\":\"end\",\"skill\":\"research\",\"outcome\":\"success\",\"ts\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" >> docs/sessions/.activity-log.jsonl

Example

Input:  /research "Compare SQLAlchemy vs Tortoise ORM for async Python APIs"

Phase 1: Query is specific, implies comparison → auto-depth: Standard
Phase 2: No prior research found
Phase 3: Plan presented —
  Sub-questions:
  1. SQLAlchemy async capabilities and maturity
  2. Tortoise ORM async capabilities and maturity
  3. Performance benchmarks and community adoption
  4. Migration path and ecosystem compatibility

Phase 4: 4 subagents dispatched in parallel
Phase 5: Confidence 82/100 — all sub-questions well-covered
Phase 6: Report saved to docs/research/sqlalchemy-vs-tortoise-orm.md
Phase 7: No backlog impact. Suggested: /brainstorm if choosing ORM for new project

Rules

  • NEVER skip the clarification gate for vague queries — bad input produces bad research
  • NEVER present training-data knowledge as research findings — all claims must come from web sources
  • NEVER dispatch research without user approval of the plan
  • ALWAYS save research reports for future reference (they prevent redundant research)
  • ALWAYS show the executive summary inline — don't make users open the file to see results
  • ALWAYS check for prior research before starting new research
  • Keep the research plan visible and the user in control of depth and scope

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.