agentleFS
jpoindexterjpoindexter/ai-agent-skills/source-grounded/SKILL.md

source-grounded

Skill1 starsChanged 5 months agoMITView on GitHub

Ground framework- and library-specific code in official documentation, not training data. Detect the installed version, fetch the relevant doc page, implement to the documented pattern, cite the source with a deep link, and flag anything you could not verify. Use when correctness depends on a specific framework or library version (routing, data fetching, forms, auth, state, build config).

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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. Very specific:

    Detect the stack + exact installed versions: Read the manifest first, then resolve ranges through the lockfile or installed runtime (npm …

    line 11
  2. Distinctive:

    Implement to the documented pattern: Use the signatures from the page.

    line 13
  3. Very specific:

    If network access is unavailable or prohibited, use bundled official documentation or the installed tool's help as local evidence and label …

    line 22

≈ “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

Source-Grounded

Training data goes stale, APIs deprecate, and "looks right" ships a pattern that broke two versions ago. Before writing framework-specific code from memory, verify it against the official docs for the installed version and cite the source so the user can check it. Version-agnostic logic — loops, data structures, renames, file moves — is exempt; this is for the version-dependent surface where the wrong pattern compiles and then bites.

Steps

  1. Detect the stack + exact installed versions. Read the manifest first, then resolve ranges through the lockfile or installed runtime (npm ls, python -m pip show, cargo metadata, and equivalents). State both the version and evidence source — React 19.1.0 (package-lock.json). Ask only after local manifests, lockfiles, and runtime metadata cannot establish the version.
  2. Fetch the exact doc page for the feature — not the homepage, not a web search. react.dev/reference/react/useActionState, not "react docs". Authority order: official docs → official blog/changelog → web standards (MDN, web.dev) → runtime/browser compat (caniuse, node.green). Stack Overflow, tutorials, and your own memory are never primary sources.
  3. Implement to the documented pattern. Use the signatures from the page. If the docs deprecate a pattern, don't emit it. If the docs don't cover it, say so and flag it unverified rather than filling the gap from memory.
  4. Surface conflicts, never resolve them silently. Docs recommend one thing, the existing codebase does another → present both (modern-per-docs vs match-existing) with the source link and let the user pick. Same when two official sources disagree (migration guide vs API reference) — name the discrepancy.
  5. Cite every framework-specific decision. Full URL, deep-linked to the anchor (/useActionState#usage — anchors survive doc restructuring better than top-level pages), in a code comment and/or the message. Quote the relevant passage when the decision is non-obvious.

Constraints

  • A pattern you could not find in official docs ships with an explicit UNVERIFIED — based on training data, may be outdated flag. Not silent, not a vague "this might be outdated" hedge — verify and cite, or flag clearly.
  • Citations are full URLs with anchors, never shortened links or a bare homepage.
  • Fetch the one relevant page, not the whole docs site — token frugality.
  • If network access is unavailable or prohibited, use bundled official documentation or the installed tool's help as local evidence and label it OFFLINE EVIDENCE; do not broaden network authority just to satisfy this skill.
  • Exempt: version-agnostic logic, renames, file moves, typo fixes, or when the user explicitly asks for speed over verification.
  • Confidence is not evidence. "I'm sure about this API" is the exact moment to check — training data is full of patterns that look correct and broke a major version ago.

Report format

STACK: <lib> <exact version> (from <manifest | lockfile | runtime>)
SOURCES:
  - <pattern> → <full doc URL#anchor>
OFFLINE EVIDENCE:
  - <pattern> → <bundled official doc | installed help | none>
UNVERIFIED:
  - <pattern> — no official doc found; training-data-based, verify before prod
CONFLICTS surfaced: <n> (resolved-by-user | pending)

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.