agentleFS
iangogenticiangogentic/venture-lab/.claude/skills/compose-report/SKILL.md

compose-report

Skill0 starsChanged 4 months agoMITView on GitHub

Render a run's findings as one evidence-first report.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write the report of this run for one reader: a founder deciding where to spend their next month.

    line 8
  2. Distinctive:

    Keep the register visible in the prose: what a source said ("developers report…"), what was inferred from it ("which suggests…"), and what …

    line 106
  3. Very specific:

    Every claim points at the artifact that carries it, citing ids inline in parentheses exactly as they appear in the inputs — (ev_…), (rb_…), …

    line 113
  4. Very specific:

    Do not invent totals, percentages, or market numbers that no artifact carries.

    line 126

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Compose the report

Write the report of this run for one reader: a founder deciding where to spend their next month. They were not in the pipeline. They will read this document alone, act on it, and be right or wrong with real time and money. Truth over optimism — a report that flatters the findings costs them a month.

You are narrating, not concluding. Every claim below already exists in an artifact; your job is the through-line, not new judgement. The verdicts were taken by the decision stage and are not yours to soften or upgrade.

The question this run set out to answer

${question}

The research briefs

${briefs}

The pain clusters (may be empty)

${clusters}

The opportunities

${opportunities}

The decisions

${decisions}

The market analyses (may be empty)

${market}

The competition analyses (may be empty)

${competition}

The contradiction analyses (may be empty)

${contradictions}

The interview plans (may be empty)

${interviews}

The harvested leads (may be empty)

${leads}

What to return

  • executive_summary — the findings without the argument, for the reader who stops here. State what was decided and what it rests on, in a short paragraph.
  • highlights — the few lines worth reading if nothing else is. Findings, not headings.
  • body — the full report, in Markdown.

The body: one evidence-first section per opportunity

Open with the question and how much evidence the run actually gathered — say plainly when it is thin. Then, for each opportunity, in this order:

  1. The thesis — what the opportunity claims, and the pain cluster and briefs it grew from.
  2. Market and competition, in brief — only what the analyses established. An unsized market is missing information, not a small market; report the basis honestly rather than dressing the gap.
  3. What contradicts it — the counter-evidence, given its own weight, never softened and never netted away. A blocking finding leads this subsection; an empty contradiction analysis is reported together with what was searched for, so the reader can judge whether the silence means anything.
  4. The decision and why — the verdict, its stated reasons, and its confidence. Include what argued the other way.
  5. The next validation step — the decision's cheapest next action, and the interview plan or leads that would serve it when the run has them.

Separate observation, inference, and hypothesis

Keep the register visible in the prose: what a source said ("developers report…"), what was inferred from it ("which suggests…"), and what is still only proposed ("the untested assumption is…"). A report that flattens these into one confident voice is wrong even when every sentence is individually true.

Cite ids for every claim

Every claim points at the artifact that carries it, citing ids inline in parentheses exactly as they appear in the inputs — (ev_…), (rb_…), (dec_…), or several at once (ev_…, cx_…). A claim you cannot cite does not go in the report.

Ids nested inside the artifacts above are citable too, and are often the better citation: the evidence a brief quotes, the cluster an opportunity answers, the opportunity a decision decided. Copy such an id exactly as it is written — never adapt, shorten or reconstruct one, and never write an id no input contains. The application checks every citation against the run's artifacts, and a single id that names nothing fails the whole reply.

What not to compute

Do not invent totals, percentages, or market numbers that no artifact carries. The application computes the title, the covered period, and the verdict tallies from the artifacts themselves; anything you return in verdict_counts is discarded.

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.