Find current documentation with ak:docs-seeker

Use ak:docs-seeker to locate current library or framework documentation from a focused topic or general library request. The Skill classifies the query, retrieves an llms.txt index through Context7 when available, follows the most relevant documentation pages, and reports source-qualified findings.

Choose ak:docs-seeker for external documentation

Use ak:docs-seeker when

• You need current API, configuration, installation, or feature documentation.
• A library version matters and you need sources that match it.
• An official documentation index can narrow a broad web search.
• Documentation is incomplete and you need a clearly labeled repository-code fallback.

Choose another workflow when

• You need facts from the current project. Use ak:scout.
• You already have source files and want one reviewable repository bundle. Use ak:repomix.
• You need a multi-source research argument rather than library documentation. Use ak:research or prepare a reusable brief with ak:research-prompt.
• You need to edit project documentation. Use ak:docs after gathering the source evidence.

Prepare the query and runtime

Before you start:

• Complete Onboarding, and confirm Engineer Kit is installed for the runtime and scope you use.
• Name the library, topic, desired version, target language, and question the documentation must answer.
• Prefer public inputs. Remove private repository names, tokens, customer data, and confidential code unless their use is necessary and approved.
• Decide whether network access, repository cloning, and local bundle creation are permitted if the documentation index is unavailable.

Run the Skill

/ak:docs-seeker "Next.js 16 cache invalidation APIs. Use official version-matched sources, cite every behavior claim, and do not clone repositories."

/ak:docs-seeker "Next.js 16 cache invalidation APIs. Use official version-matched sources, cite every behavior claim, and do not clone repositories."

$ak:docs-seeker "Next.js 16 cache invalidation APIs. Use official version-matched sources, cite every behavior claim, and do not clone repositories."

The Skill accepts a library name and optional topic as free-form input. It has no published mode flag. Topic-specific requests take the shortest path; general requests can return a larger index that needs prioritization.

Expected execution shape

The bundled detector and index analyzer run locally and print JSON. The fetcher makes HTTPS requests to context7.com, optionally with CONTEXT7_API_KEY, and prints the fetched content inside JSON. They do not write a cache or llms.txt file unless the surrounding workflow explicitly redirects or saves output.

Follow the evidence path

1. The Skill classifies the query. It extracts a library and short topic or treats the request as general.
2. It retrieves an index. Topic-specific Context7 lookup falls back to a general library lookup when the topic endpoint is unavailable.
3. It prioritizes pages. Getting-started and installation pages are grouped separately from API guides, advanced material, migration notes, and other supplementary sources.
4. It opens primary documents. Findings should come from the official, version-matched pages linked by the index, not from the index title alone.
5. It falls back deliberately. When official docs are missing, the Skill can inspect the official repository and label claims inferred from README, tests, examples, or implementation code.
6. It synthesizes with citations. Each behavioral or version claim points to the page that supports it; conflicts and documentation gaps remain visible.

Keep source access safe

• Environment lookup can read .env files at Skill, shared-Skills, and runtime levels; process environment values take precedence. Never print or cite token values.
• CONTEXT7_API_KEY is optional for the bundled fetcher. GITHUB_TOKEN and GEMINI_API_KEY appear in the example environment but are not required by the core detect, fetch, and analyze scripts.
• Cloning a repository, checking out a tag, installing Repomix globally, and writing repomix-output.xml are external or persistent effects. Request approval and use a bounded destination before taking them.
• Public web access may still incur provider quotas. External models, paid APIs, or broad agent distribution require separate cost approval.
• Do not present community tutorials as equal to official versioned docs. When sources conflict, show the conflict and explain the selected authority.

Verify the result

A complete result should name the library and version, distinguish direct documentation from code inference, link each primary source, summarize the answer to the requested topic, record fallbacks, and identify unresolved gaps. If a repository was cloned or packed, it should also report the destination, generated bundle, cleanup status, and disk impact.

The bundled tests cover query normalization and classification, Context7 URL construction, URL parsing and priority grouping, and distribution suggestions. They do not prove that a live provider is available or that a retrieved page is correct for the requested release.

Troubleshoot or continue

Know the current limits

• Query classification is pattern-based and can misclassify ambiguous wording.
• Context7 coverage, page freshness, rate limits, and linked URLs are external state; verify important claims against the official page.
• Repository analysis is code-based evidence, not a substitute for a documented public contract.
• The source's timing and agent-count suggestions are projections, not guarantees or required parallelism.
• Stable and beta package identical ak:docs-seeker scripts and workflows.