agentleFS
holon-runholon-run/holon/skills/research-artifact/SKILL.md

research-artifact

Skill197 starsChanged yesterdayApache-2.0View on GitHub

Create a compact, auditable research package that keeps the question, source ledger, claim matrix, decision brief, and follow-up queue linked by stable IDs.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 4 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill to turn a completed or paused research pass into a package that another person can inspect and resume.

    line 10
  2. Distinctive:

    Create a directory or document with these sections, in this order:

    line 16
  3. Very specific:

    Run citation-verifier before marking a claim supported.

    line 27
  4. Very specific:

    Use stable IDs such as Q-001, SRC-001, CLM-001, and FU-001.

    line 35
  5. Distinctive:

    Start from the smallest package that answers the decision.

    line 56
  6. Very specific:

    Run citation-verifier and carry its statuses and conflicts into the decision brief.

    line 60
  7. Very specific:

    Put the next useful action in follow-up-queue, including whether operator input or external access is required.

    line 64
  8. Distinctive:

    Default output is a local report in the current response or an explicitly requested workspace location.

    line 69
  9. Very specific:

    Keep facts, interpretations, recommendations, and open questions in separate sections.

    line 74
  10. Very specific:

    If sources conflict, retain both records and document the dimensions of the conflict.

    line 76

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Research Artifact Skill

Purpose

Use this skill to turn a completed or paused research pass into a package that another person can inspect and resume. The package is a record of evidence and uncertainty, not an implicit knowledge base and not permission to publish.

Package layout

Create a directory or document with these sections, in this order:

  1. question-card — question ID, decision, audience, scope, definitions, time window, acceptance test, assumptions, and owner.
  2. search-plan — sub-questions, source types, queries or collection methods, date bounds, coverage notes, and deliberate exclusions.
  3. source-ledger — one row per source with source_id, title, author or publisher, locator, source type, publication/update date, access date, relevant scope, extracted evidence, and limitations.
  4. claim-matrix — one row per material claim with claim_id, wording, source IDs, exact locator, status, confidence, counter-evidence, and reason. Run citation-verifier before marking a claim supported.
  5. decision-brief — answer first, then evidence, competing explanations, caveats, confidence, and what would change the conclusion.
  6. follow-up-queue — unresolved questions, priority, blocking reason, next action, and the condition for closing each item.
  7. provenance — artifact version, creator, verification time, source access notes, and any operator approvals.

Use stable IDs such as Q-001, SRC-001, CLM-001, and FU-001. Keep IDs stable when the package is revised; append a revision note instead of silently rewriting history.

Markdown and JSONL contract

Markdown is the human-readable default. When machine processing is useful, add JSONL records with one record per line and a record_type field:

{"record_type":"source","source_id":"SRC-001","locator":"...","accessed":"YYYY-MM-DD"}
{"record_type":"claim","claim_id":"CLM-001","source_ids":["SRC-001"],"status":"unverified"}

JSONL is an index of the Markdown package, not a second contradictory source of truth. Every claim record must link to source IDs, and every supported claim must have a locator. Do not emit fabricated placeholder evidence as if it were verified.

Production procedure

  1. Start from the smallest package that answers the decision; do not create a broad archive by default.
  2. Record sources before prose conclusions and preserve inaccessible/negative results as limitations.
  3. Run citation-verifier and carry its statuses and conflicts into the decision brief.
  4. Mark the package complete, paused, or blocked; never use complete when material claims remain unverified without an explicit caveat.
  5. Put the next useful action in follow-up-queue, including whether operator input or external access is required.

Permissions and safety

  • Default output is a local report in the current response or an explicitly requested workspace location. Do not write project files, upload sources, publish results, or create a persistent archive without exact authorization.
  • Preserve original source locators and access dates. Do not copy long copyrighted passages; summarize and point to the locator.
  • Keep facts, interpretations, recommendations, and open questions in separate sections. A polished brief does not upgrade unverified evidence.
  • If sources conflict, retain both records and document the dimensions of the conflict. If a source cannot be inspected, keep the claim blocked or unverified.

Completion checklist

  • Question and acceptance test are explicit.
  • Material sources have stable IDs and traceable locators.
  • Material claims link to source IDs and verifier statuses.
  • Counter-evidence and limitations are preserved.
  • Freshness and access dates are recorded where relevant.
  • The brief states confidence, unresolved questions, and next actions.
  • No external write or publication occurred without authorization.

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.