agentleFS
Sign inStart free
theonly1metheonly1me/shadowclone/skills/research-primary-sources/SKILL.md

research-primary-sources

Skill25 starsChanged 32 days agoMITView on GitHub

Use when a decision depends on how a library, platform, protocol, or tool behaves, and that behavior can change by version. Also use it when the user says "check the docs", "is this still true?", or "what does the spec say?". Finds the first-party source for the installed version and reads it. Separates what the source guarantees from what you infer. Not for debugging a failure in your own code (use `diagnose-before-editing`).

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If no source answers the question, did you say so and keep the choice easy to reverse?

    line 21
  2. Distinctive:

    Write the decision as a question.

    line 25
  3. Very specific:

    Write the answer, the link or file:line for each claim, and the limit that it puts on the change.

    line 29
  4. Distinctive:

    Hidden cost: If memory is wrong, an install can write a file that hides a file of the team, and nobody notices.

    line 35
  5. Distinctive:

    Keep credentials, private links, and user data out of research notes.

    line 43
  6. Distinctive:

    If the evidence is old, say so, and name the version that it covers.

    line 44

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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 Primary Sources

Use when

A decision depends on outside facts about a library, a platform, a protocol, a standard, or a tool. Those facts change with versions, and memory or a search snippet can be out of date.

Gates

  1. Is the question specific enough to have one answer?
  2. Did you record the installed version, the runtime, and the platform that the answer must match?
  3. Did you read the first-party source itself, such as the specification, the official documents, the source code, or the release notes?
  4. Does each claim say whether the source guarantees it or you infer it?
  5. If no source answers the question, did you say so and keep the choice easy to reverse?

Process

  1. Write the decision as a question. Record the version, the runtime, the platform, and the limits that apply.
  2. Find the source that defines the behavior. Prefer the specification, the official documents, the source code, a release note, or a first-party issue.
  3. Open the exact page or file. Check that its version and date match yours.
  4. Compare the source with the installed code. Explain each difference that comes from a version or an environment.
  5. Write the answer, the link or file:line for each claim, and the limit that it puts on the change.

Example

Situation: A change must know which instruction file a coding agent reads when two files exist in one folder. Easy route: Answer from memory: "it reads both". Hidden cost: If memory is wrong, an install can write a file that hides a file of the team, and nobody notices. Best route: Open the official documents for the installed version, find the section about file order, and quote it. Evidence: The documents said that the agent reads the override file first, and reads at most one file in each folder. The install now stops before it writes a file that hides another one.

Guardrails

  • A search result is a lead. Read the source that it points to.
  • Quote only the lines that support the decision.
  • Keep credentials, private links, and user data out of research notes.
  • If the evidence is old, say so, and name the version that it covers.

Completion

  • The answer, or the exact unknown.
  • Each claim with its source: a link, or a file:line.
  • The version that each source covers.
  • The limit that the answer puts on the change.

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.

Check your setup, free