LLM Documentation Index Generator
Turns project wiki content into llms.txt-compliant documentation for language-model discovery and context loading.
The skill requires resolving the git remote and branch, asking the user to clarify repository provenance, and avoiding overwrite of an existing root llms.txt not generated by deep-wiki. These provide some scoping and confirmation; deductions apply because sensitive-data handling, external data flows, backup/rollback, and link-safety checks are unspecified.
The instructions are broadly consistent and include prerequisites, format rules, and a validation checklist. However, there is no executable implementation, dependency guidance, abnormal-input handling, or designed failure feedback; key generation behavior is left to agent judgment, so the static score is capped below 10.
Activation scenarios are reasonably clear and cover llms.txt, llms-full.txt, and LLM-friendly documentation requests. Deductions apply because non-fit boundaries, repository-layout variants, Chinese-language support, and mainland-China network reachability are not adequately addressed.
The documentation is layered and includes metadata, format examples, prerequisites, output descriptions, and validation checks, with MIT licensing and version 1.0.0 stated. It lacks a changelog, explicit maintenance owner, update path, dependency/install notes, and troubleshooting guidance.
The target files, structure, link conventions, and validation requirements are concrete enough to make the core task plausible. No representative output or key-path reproduction evidence is provided, and complex or incomplete wikis still require substantial agent judgment, so the static score is capped below 7.
The skill links to the llms.txt specification and states concrete rules and checks, providing limited auditability. There are no skill-specific tests, committed generated artifacts, or third-party execution results, so the score remains below the static ceiling of 5.
- The skill requires git commands and repository-source clarification but does not define behavior when the commands, remote lookup, or branch detection fail.
- It may read the full wiki, yet gives no guidance for preventing secrets, internal links, or other sensitive data from being copied into llms-full.txt.
- The validation checklist is agent-executed without automated tests or reproducible samples; manually inspect links, completeness, and coverage before release.
- The workflow references the llmstxt.org specification but provides no local copy or fallback if that external reference is unreachable.
What it does & when to use it
This Microsoft Agent Skills skill generates llms.txt and llms-full.txt files for LLM-friendly project documentation. It is intended for developers and documentation maintainers who want project wiki content to be easier for language models and coding agents to consume. The skill resolves repository context before generating files, then derives links and full content from existing wiki pages. It is part of the deep-wiki plugin and is not a general Azure SDK development skill.
It first runs git remote get-url origin and git rev-parse --abbrev-ref HEAD to resolve the source repository URL and current branch, and asks the user to confirm whether the repository is local-only or has a source URL. It then reads existing wiki pages and produces ./llms.txt, wiki/llms.txt, and wiki/llms-full.txt. The index files contain relative links and page descriptions; the full file inlines page content while stripping YAML frontmatter and preserving Mermaid diagrams, citations, tables, and Sources comments. If no wiki exists, it can suggest /deep-wiki:generate or /deep-wiki:page, or create a minimal README-and-source-scan version without wiki page links. It must not overwrite an existing root llms.txt that was not generated by deep-wiki.
- A development team maintaining a project wiki needs a standard discovery file for language models.
- A documentation maintainer has generated wiki pages with deep-wiki and wants concise and full llms.txt outputs.
- A team deploying documentation with VitePress needs wiki-relative links for internal navigation.
- An engineering team needs Mermaid diagrams, citations, tables, and page content packaged into one context-expanded document for coding or documentation agents.
Pros & cons
- Produces both root-level and wiki-relative indexes for standard discovery and VitePress navigation.
- Preserves Mermaid diagrams, citations, tables, and source comments in the full output.
- Provides a validation checklist covering links, empty document blocks, duplicates, ordering, and size.
- MIT-licensed, with the README describing active maintenance and expanding tests.
- It works best when wiki pages already exist; without them, output is limited to a minimal README-and-source-scan version.
- Generation cannot proceed until repository context is resolved and the user confirms the repository type.
- The source provides no independent test coverage or concrete runtime-version requirement for this individual skill.
- It writes multiple documentation files and deliberately leaves a pre-existing non-deep-wiki root llms.txt untouched.
How to install
The repository README supports npx skills add microsoft/skills, followed by selecting skills in the installation wizard; skills are installed into the selected agent directory, such as .github/skills/ for GitHub Copilot. As a plugin, Copilot CLI users can run /plugin marketplace add microsoft/skills and /plugin install deep-wiki@skills. The source does not document a separate installation command for only the wiki-llms-txt folder.
How to use
After installing the skill in a project with available wiki pages, ask to “generate llms.txt and llms-full.txt” or to “make the project documentation LLM-friendly according to the llms.txt standard.” Before generation, confirm whether the repository is local-only or has a source repository URL; the skill then resolves the Git remote and current branch. If no wiki exists, run /deep-wiki:generate or /deep-wiki:page first.