agentleFS
iangogenticiangogentic/venture-lab/.claude/skills/research-brief/SKILL.md

research-brief

Skill0 starsChanged 4 months agoMITView on GitHub

Synthesise collected evidence into cited research briefs.

20 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use meeting-notes and weekly-review.

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.
  • claude-office-skillsSkill

    meeting-notes

    >

    Also in 3 projects that use this

  • QuixiAISkill

    weekly-review

    Review the past week across goals, tasks, calendar, journal, and commitments, then propose the next week

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 19 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not inflate it, and do not let the number of evidence records stand in for it.

    line 80
  2. Distinctive:

    "Many users report": when it was two, or one source quoted twice.

    line 86

“Common” is measured: kept by at least 6 in 10 of the 19 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 brief

Say what this evidence adds up to — and, just as importantly, how much it is worth.

The question

${question}

The evidence

${evidence}

The job

A brief is what every later stage reads instead of the raw evidence. Someone who never sees these excerpts must be able to act on your brief and know how far to trust it. That second half is not metadata; it is half the output.

Cover one coherent slice of the question per brief. If the evidence genuinely splits into separate stories — different segments, different problems, different timeframes — write a brief for each. Do not fuse unrelated material because it arrived together, and do not split one story into fragments to look thorough.

Four different things, kept apart

Signals are what the evidence shows: one pattern per signal, stated so it could be checked on its own. "Teams re-key the same order into two systems every morning" is a signal. "There are issues around data entry" is not. Tie each signal to the specific evidence ids carrying that statement, not to the whole pile you read — which subset supports which claim is exactly what a reader needs.

Tag every signal:

  • observation — the supplied evidence says this directly. You could quote it.
  • inference — you reasoned it from the evidence. The reasoning must be legible in the statement itself.
  • hypothesis — you are proposing it. Nothing supplied establishes it yet.

Most signals in a first pass are inference. Labelling one as observation because it feels solid is the failure this field exists to catch.

Quotes are verbatim excerpts, kept attributable. They are the one place a downstream reader can audit the pipeline, so a quote that cannot be traced back to its evidence is worse than no quote. Never paraphrase inside a quote, never tidy the grammar, never merge two speakers.

Contradictions are places where sources disagree — a topic plus at least two positions, each stated in its own terms so whoever holds it would recognise it. Do not resolve them by majority, by recency, or by which reads better. Do not average them into a claim neither source made. A single position with a caveat is not a contradiction. Note also which kind of disagreement it is: about what is happening, about what causes it, or about whether it matters — those are three different findings.

Unknowns are what this evidence does not settle. Be concrete about what would settle each one — the observation, source or measurement that would close the gap. These drive the next collection round, so "more research needed" is useless; name the research. An empty unknowns list is a very strong claim about your own coverage, and is almost never true.

Grade your own evidence honestly

  • Quality is what the evidence is: nothing behind it, a single unverified report, several independent sources agreeing, quantitative data, or something confirmed against a primary source. Grade the base as a whole, at the level the weakest load-bearing part supports — not at the level of your best single source.
  • Density is how much of it there is. Sparse means a handful of sources, or many voices tracing back to one origin. Moderate means enough to see a pattern but not to size it. Dense means many genuinely independent sources. On a normal run, sparse is the truthful answer, and reporting it costs nothing — reporting dense over eight forum posts costs the decision at the end of the pipeline.
  • Source count is distinct sources. One source restated in three excerpts counts once. Do not inflate it, and do not let the number of evidence records stand in for it.

Failure modes

  • Laundering: a claim in the summary that appears in no signal and no evidence.
  • Smoothing: presenting a contested point as settled because most excerpts agreed.
  • "Many users report" when it was two, or one source quoted twice.
  • Recapitulation: replaying the excerpts in order instead of saying what they mean together. If your summary would be unchanged with half the evidence removed, you summarised the pile rather than synthesised it.
  • Confident prose over an anecdote. Match the register to the grounding.
  • Tagging inference as observation, or grading density up because the brief felt thin.
  • Importing a fact you happen to know to fill a gap the evidence left.

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.