What problem does it solve? AI-generated code for libraries, SDKs, and APIs is often written from stale training data, producing outdated signatures, deprecated patterns, and invented endpoints that look correct but break at runtime. This Skill enforces a workflow where every library-specific decision is grounded in a verifiable source — the installed library's code, official documentation, changelogs, or specs — and explicitly cited. ## Core Features & Use Cases - Version identification from the project: Reads declared and installed versions from requirements.txt, pyproject.toml, pip freeze, and runtime version attributes, with the installed version winning over the declared one. - Source authority hierarchy: Prioritizes the library's own source code, then official docs, changelogs, issues/PRs, and standards — explicitly rejecting Stack Overflow, blog posts, and training-data memory as authorities. - Version-pinned documentation retrieval: Uses Context7 (ctx7) queries with version-scoped identifiers to fetch docs matching the installed release, with a three-attempt cap and explicit unverified marking on failure. - Mandatory citation and honesty: Every non-trivial decision gets a full link or code reference; anything unverifiable is marked as unverified rather than silently guessed, and endpoints are never constructed by analogy. - Use Case: When writing a call against python-telegram-bot, MCP, or pymongo, the Skill reads the installed package source to confirm current class names and signatures, cites the exact file, and flags any conflict with existing project code for the user to decide. ## Quick Start Ask the assistant to implement an API call using the installed version of a specific library and require a verifiable source citation for every library-specific decision.
