agentleFS
exm7777exm7777/co-founder/.agents/skills/research/SKILL.md

research

Skill0 starsChanged 3 months agoMITView on GitHub

Live, verified, cited answers for business decisions. Run when the user says "look this up", "get me the real numbers on...", "what's the market doing", "is this actually true", "research X", or when any recommendation needs its external receipt. Run it BEFORE advising on anything the training data would otherwise answer from memory, because memory is not research.

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use Problem-Solving Techniques, daily-arxiv 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.
  • 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 6 projects that use this

  • 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

  • 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 49 projects that use this

  • parcadeiSkill

    ast-grep-find

    AST-based code search and refactoring via ast-grep MCP

    Also in 4 projects that use this

  • SireJeffSkill

    context-optimize

    Context Optimization - Orchestrate RPI workflow TODO generation with interactive scoping

    Also in 4 projects that use this

  • +5 more from Lambenthan

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 170 edited versions we read keep these.

  • 5

    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:

    Before any step, read references/CONVENTIONS-core.md.

    line 30
  2. Very specific:

    Run the shared contract's Universal preflight.

    line 36
  3. Distinctive:

    Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or …

    line 53
  4. Distinctive:

    Use the best engines this environment has, in this order of preference: dedicated search and scrape tools if available (for example, …

    line 69
  5. Distinctive:

    Prefer primary sources over roundups: the study over the listicle, the pricing page over the review, the practitioner's own numbers over …

    line 78
  6. Very specific:

    STOP. Read references/verify-checklist.md before grading any claim: it holds the three traps that make false claims look corroborated, and …

    line 88
  7. Distinctive:

    Run the shared durable claim admission gate before filing.

    line 105
  8. Distinctive:

    Use the shared verified-absence record for every zero-result finding.

    line 111
  9. Very specific:

    Write the full dossier to wiki/research/<slug>.md with the universal wiki frontmatter, the base five plus sources (type: research, title, …

    line 117
  10. Very specific:

    Run scripts/graph-audit on every touched durable file, including the complete dossier.

    line 131

“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: receipts or it doesn't ship

Advice built on training data is pattern-matching dressed as knowledge. This skill goes and finds what is true right now, checks it, cites it, and files it where the business can use it again. It is how every other skill earns the right to recommend anything.

  the question
       |
  [ tier ]      quick: minutes, 2-3 sources   standard: multi-angle
       |        deep (opt-in): fan-out + independent fact-check
       |
  [ sweep ]     best engines available, multiple phrasings
       |
  [ verify ]    corroborate across INDEPENDENT sources
       |
  [ file ]      full dossier -> wiki/research/    chat gets the gist + link

Before any step, read references/CONVENTIONS-core.md. This resolves from the installed skill directory to the plugin's shared contract, independent of the founder's working folder. Its receipts, filing, and dedupe rules govern this workflow.

Step 0: Preflight

Run the shared contract's Universal preflight. A Fresh, Partial, or Wrong-folder result stops and runs the co-founder-setup skill before anything else (re-sync for Partial), without researching or filing. On Ready, read the charter's path map and the relevant vault context.

Done when: the founder system is Ready and the dossier destination is known.

Step 1: Sharpen the question

Restate what is actually being decided, in one line, and confirm it. "Research newsletters" is not a question; "what open rate should a 2k-subscriber list expect, so we can judge ours" is. A question doing double duty gets split.

Done when: one decision-shaped question is agreed.

Apply the shared parameter gate to any business-specific premise needed to shape or interpret the research. Search the vault, then ask the founder or label the premise as an unverified working hypothesis. Search-tier mechanics are operational; the business inputs are not yours to fill.

Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or citing a receipt for it.

Step 2: Pick the tier

  • Quick: a fact to check, minutes of work, 2-3 sources. Default for small stakes.
  • Standard: a decision input, multiple angles, 5-8 sources.
  • Deep: bet-the-business calls only, and the founder opts in (it costs real time and tokens): full fan-out plus the independent fact-check pass in step 4.

Uncertain resolves upward. State the tier and, for deep, get the yes first.

Done when: the tier is stated.

Step 3: Sweep

Use the best engines this environment has, in this order of preference: dedicated search and scrape tools if available (for example, Firecrawl or similar), social-pulse tools (recent posts and community threads), then the harness's native web search and page fetching. The engines vary by machine; the process never does.

Sweep rules:

  • Multiple phrasings per question: the practitioner's wording, the vendor's wording, the skeptic's wording. One phrasing finds one bubble.
  • Prefer primary sources over roundups: the study over the listicle, the pricing page over the review, the practitioner's own numbers over the guru's summary.
  • Capture every candidate claim WITH its source as you go. An uncited claim does not exist.
  • Record the exact query, engine, search date, and result count inspected for every negative finding. A finite search supports "none found in this search," never "none exists."

Done when: each angle of the question has candidate claims with named sources.

Step 4: Verify

STOP. Read references/verify-checklist.md before grading any claim: it holds the three traps that make false claims look corroborated, and skipping it is how slop gets certified.

Grade each claim: verified (independent sources agree), stated (one source claims it; carried as a quote, never as truth), contested (sources disagree; both sides carried). Deep tier adds the adversarial pass: each load-bearing claim gets a fresh check whose only job is to break it, using the checklist.

Research findings may be external facts without pretending to know what this business should do. A benchmark applied to this business, a "should" answer, or any other consequential recommendation uses the shared receipts contract in references/CONVENTIONS-receipts-contract.md: one internal receipt and one verified external source. One floor missing: deliver the factual findings, then write the recommendation in the canonical PARKED: internal receipt missing or PARKED: external receipt missing shape and name recall or research as the next action. Evidence gaps never move to "Needs your call".

Done when: every claim in the synthesis carries a grade and a source.

Run the shared durable claim admission gate before filing. Reopen every cited page and confirm the passage supports the written claim. Preserve the short supporting quote beside the citation. A source that disappeared or does not contain the evidence is removed; the claim is downgraded to stated with its honest source or omitted. Negative findings carry the saved search scope rather than upgrading verified search absence into a universal fact.

Use the shared verified-absence record for every zero-result finding. It may establish the bounded gap and support a PARKED recommendation; it never supplies the external leg for a SURVIVES/DIES call or proves that the missing fact does not exist.

Step 5: File and deliver

Write the full dossier to wiki/research/<slug>.md with the universal wiki frontmatter, the base five plus sources (type: research, title, status from the overall confidence, created, updated, sources: list), then the question, the synthesis, every claim with its grade and citation, Parked recommendations for missing receipt legs, and a "Needs your call" section only for judgment forks the evidence cannot settle. Include a Search record with the exact queries, engines, dates, and inspected-result counts. Update index.md, link hub/research.md, and append ## [date] research | <question> to log.md.

In chat: the answer in 3-6 lines, the strongest receipt, the link to the dossier. Never paste the dossier into chat; next month's question starts from the file, not from a re-run.

Mandatory filing gate

  1. Run scripts/graph-audit on every touched durable file, including the complete dossier.
  2. Fix every violation and rerun it. The dossier and research event remain incomplete until it exits zero.

Done when: the dossier is filed, indexed, logged, and the founder has the gist with its link.

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.