agentleFS
LiriRaidLiriRaid/waymark/skills/library-docs/SKILL.md

library-docs

Skill2 starsChanged 6 days agoMITView on GitHub

Before code that uses a library API not verified this session, and for setup or migration: \"cómo se usa\", \"migrar a la versión\", \"sigue sin funcionar\".

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before writing or changing code that calls a library API not verified in this session.

    line 20
  2. Very specific:

    Check the session's tool list (or ToolSearch) for the exact tool names before calling.

    line 30
  3. Distinctive:

    When the route falls back, say so in one line (for example: "framework MCP not connected.

    line 40
  4. Very specific:

    Before querying, read facts/<library>.md (format in facts/README.md).

    line 71
  5. Very specific:

    If the fact is really a convention of the stack (how the user's projects should use the library), propose it for …

    line 73
  6. Distinctive:

    Read only the mode the task needs (one file).

    line 77

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Library Docs

Precondition. Tool of waymark, shared by every dept-* department. If no department skill is loaded in this conversation, load the owner of the task first and use its brief (what, why, where, how) as the input of this skill. Skip for L0 edits and for questions that do not depend on a library's behavior.

Approach

Training data ages; libraries do not wait. Before code leans on an API, this skill pins down which version the project really runs and asks a source that speaks for that version.

  • Version first, question second. The installed version decides the answer. A correct answer for the wrong major is a wrong answer.
  • Most authoritative source wins. A framework's own MCP beats a general docs index; a docs index beats web search; web search beats memory. On the web, use the official site of the project (e.g. angular.dev, tailwindcss.com, docs.nestjs.com, supabase.com/docs, prisma.io/docs, postgresql.org/docs, react.dev) and its versioned pages; blogs, tutorials and Q&A sites only to locate the official page, never as the source.
  • One concept per question. Narrow queries return focused, citable snippets; broad ones return noise.
  • Budget the lookups. At most 3 docs calls per question. If three calls did not settle it, say what is still uncertain instead of guessing.
  • Nothing private leaves the machine. Queries carry API names and symptoms, never keys, tokens, connection strings, customer data or proprietary code.
  • Remember only what surprised you. Verified, non-obvious facts are saved per library and version so the next session does not pay for the same lookup.

When to run

  • Before writing or changing code that calls a library API not verified in this session.
  • Setup, configuration, CLI flags, migration between versions, deprecations, breaking changes.
  • An error message that points at library behavior (wrong signature, removed option, changed default).
  • Escalation — mandatory, even if the API "looks known":
    • L2+/L3 tasks that rely on framework or platform features (rendering, routing, forms, styling system, ORM, auth, realtime, database features), before the first edit;
    • after a failed attempt, before trying again: re-read the official docs for the exact API and version instead of retrying variations;
    • the user asks for the same thing again or says it still does not work ("sigue sin funcionar", "otra vez", "no quedó"): stop, consult official docs, re-diagnose, and record the cause in project memory → Gotchas so it is not repeated.
  • Skip it for general programming questions, business logic and code review that does not hinge on an API.

Provider routing

Check the session's tool list (or ToolSearch) for the exact tool names before calling; servers rename tools between releases. A server listed as failed to connect counts as unavailable.

Use the MCP servers this user actually has. They are listed in ../waymark/skill-registry.md → MCP sections (sync indexes every configured server) and in the session tool list. Never assume a server exists because another user or a stack profile mentions it.

LibraryFirst providerFallback
The framework or library has its own docs MCP connected (vendor or framework server)that MCP — check its exact tool names in the sessiongeneral docs index
Any librarya general docs-index MCP the user has connected (e.g. context7)official site
No docs MCP connectedthe official site, versioned pages (web fetch or search scoped to it)memory, labelled as unverified

When the route falls back, say so in one line (for example: "framework MCP not connected; used the docs index + the official site").

Guardrails

  • MUST detect the installed version before the first query.
  • MUST quote or paraphrase the docs that justify the code, with the version and the source.
  • MUST NOT send secrets, environment values, personal data or large private code excerpts in a query.
  • MUST NOT exceed 3 docs calls per question without telling the user why.
  • MUST NOT present memory as documentation; when unverified, say so.

Inputs

InputSource
Briefthe department brief: what code depends on which library API
Stack conventions../waymark/stacks/<stack>.md → Tools (docs.library row) and Official docs
Project context<project>/.waymark/memory.md → Identity (stack, versions), Gotchas
Known factsfacts/<library>.md in this skill (verified, version-scoped)

Output contract

Return to the department, in this shape:

Docs · <library> <installed version> · provider: <MCP name | web (official site) | memory (unverified)>
Answer: <one or two sentences>
Source: <tool + library id or doc page>
Applied to: <file:line or "answer only">
Fact saved: <facts/<library>.md | none (not novel) | proposed for stacks/<stack>.md>
Open risk: <anything still uncertain, or none>

Pattern library (grows with use)

This skill grows facts instead of UI patterns:

  • Before querying, read facts/<library>.md (format in facts/README.md).
  • After a lookup confirmed something non-obvious (a renamed option, a changed default, a version-specific signature, a deprecated path that still compiles), run the novelty check (waymark references/learning.md: grep the fact in this file, the stack profile, project memory and facts/). Novel and verified → append it to facts/<library>.md with its version range and source.
  • If the fact is really a convention of the stack (how the user's projects should use the library), propose it for ../waymark/stacks/<stack>.md → Conventions instead of saving it here.
  • A newer version that changes a fact → mark the old line as superseded with its version range; never silently rewrite it.

Modes

Read only the mode the task needs (one file); never load all modes.

  • lookup — an API, option, signature or behavior must be confirmed before coding. → modes/lookup.md
  • migrate — upgrading a major or minor, removing a deprecation, adopting a new API across the code base. → modes/migrate.md
  • configure — setting up or changing a library's configuration, provider wiring, CLI flags or build integration. → modes/configure.md

Stack adapters

Stack-specific notes → references/stack-adapters.md (read only the project's stack row).

Learned notes

Grows with use (waymark references/learning.md). Dated, non-obvious notes about using this tool. When there are more than ~10, fold them into the body above and clear this list.

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.