What does this skill do, and when should you use it?
This skill generates high-quality AGENTS.md files for repository folders, covering build commands, testing instructions, project structure, code style, and operational boundaries. It checks each folder independently and generates a file only when AGENTS.md is missing. It derives commands, paths, and conventions from existing manifests, CI configuration, and source files instead of inventing them. It is best suited to teams adding structured coding-agent context to existing repositories, including monorepos and generated wiki sites.
Reads target folders, configuration files, CI workflows, and selected source files; checks whether AGENTS.md and CLAUDE.md already exist; detects the language, framework, build tool, test runner, structure, naming, imports, and other conventions; generates AGENTS.md for applicable roots, wiki folders, source and test directories, packages, applications, services, and folders with build manifests; generates the fixed CLAUDE.md companion only when it is absent; and reports existing files as skipped.
- A team maintaining a large repository needs agent guidance at the root and in relevant nested projects.
- Developers inheriting a repository need executable build and test instructions extracted from existing configuration.
- A monorepo team needs folder-specific AGENTS.md files for packages, apps, or services without duplicating root guidance.
- A deep-wiki-generated VitePress site needs a dedicated wiki/AGENTS.md file.
- A team wants new context files created safely while protecting existing AGENTS.md files from overwrites.
How do you install this skill?
- Generation writes AGENTS.md and potentially CLAUDE.md, but explicit user confirmation, backup, and rollback procedures are not specified.
- Command and template generation depends on detected project files; complex, nonstandard, or multilingual repositories may receive incomplete guidance.
- Chinese output handling and dedicated automated quality evidence for generated files are not documented.
- Shell / CLI
- Local filesystem
Install the collection with npx skills add microsoft/skills, then select the needed skill in the wizard. The collection installs into the selected coding agent directory, such as .github/skills/ for GitHub Copilot, and can be symlinked for multiple agents. The source does not document a separate install command for wiki-agents-md; the skill is located at .github/plugins/deep-wiki/skills/wiki-agents-md/SKILL.md.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/microsoft/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/.github/plugins/deep-wiki/skills/wiki-agents-md" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
With the skill installed in a repository, use a request such as: Generate AGENTS.md files for applicable folders that do not already have one, and create CLAUDE.md only where it is missing. Inspect real commands, structure, conventions, and boundaries; never overwrite existing files. The source does not document a specific agent trigger syntax.
What are this skill's strengths and limitations?
- The non-overwrite guard is explicit and applies independently to every folder.
- The workflow requires evidence from real manifests, CI files, and source code, reducing invented commands and paths.
- It handles repository roots, nested folders, wiki sites, source and test directories, and monorepo packages.
- It can create a protected CLAUDE.md companion that redirects agents to AGENTS.md.
- The skill has no dedicated test suite or acceptance scenarios; the README documents testing infrastructure for the broader collection.
- Output quality depends on the target repository exposing usable configuration, CI, and source-code evidence.
- It is designed to generate missing files, not update existing AGENTS.md content.
- The source does not specify agent-specific trigger commands or permission behavior.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| AGENTS.md Context Builder this page ✓ Microsoft · Official | 54 · Use with care | ★ 2.7k | 3mo ago | MIT |
| Harness Creator: Harness Engineering for AI Coding Agents | 56 · Use with care | ★ 20k | 24d ago | MIT |
| AionUi Architecture Conventions | 57 · Use with care | ★ 34k | 11d ago | Apache-2.0 |
| Claude Reflect Self-Learning Memory | 57 · Use with care | ★ 1.8k | 22d ago | MIT |
| faf-cli — Persistent Project Context (FAF Expert Skill) | 50 · Use with care | ★ 41 | 10d ago | MIT |
AGENTS.md provides project context such as commands, conventions, and boundaries, while README.md is intended for human readers; the two are complementary. .github/agents/*.agent.md defines agent personas rather than repository context.
How did FollowSkills review this skill?
The skill enforces a missing-file guard, lists directories to skip, and warns against secrets and dangerous operations, providing some permission and data-safety boundaries. It does not require user confirmation before writes and does not specify backups, rollback, write scope, data flows, or dependency security, so points are deducted.
The workflow is internally consistent and includes existence checks, configuration inspection, real-path validation, and handling for missing project features. However, this static review finds no executable tests for the skill's key paths, generated-output validation, or detailed recovery feedback; under static calibration the score remains at or below 10.
The audience, AGENTS.md generation scenarios, root/nested-folder rules, and missing-only trigger are reasonably clear. Non-fit cases, false-trigger boundaries, and handling of complex projects are not fully specified. The core function is local-repository based rather than entirely dependent on overseas services, but Chinese-language support is not addressed, so points are deducted.
Metadata provides a name, description, MIT license, author, and version. The documentation has clear structure, templates, examples, quality principles, anti-patterns, and maintenance-oriented guidance. It lacks a changelog, explicit maintenance owner, update path, and installation/dependency notes, so points are deducted.
The skill gives a directly actionable process for scanning folders, extracting commands, detecting structure, defining boundaries, and composing templates, covering the core task. Still, much of the example material is generic, no representative output is statically verified, and comparative benefit over manual authoring is not demonstrated; the score is conservatively limited.
The skill specifies auditable paths, commands, and a no-invention validation rule, while the repository provides generic CI and test evidence. The supplied tests do not specifically cover this skill's key paths, and there is no independent reproduction result or multi-source corroboration, so only a low score is justified.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →