com.dmzagent/skill-builder-mcp

Fetch the complete record for one skill: metadata plus the full markdown body (the actual prompt/instructions). Resolves by id or name. Read-only — unlike an install, this does NOT count as a download. Args: - skill (string): skill id ("@handle/slug" or bare "slug") or its display name. - response_format ('markdown' | 'json'): default 'markdown'. Returns: name, id, type, category, tags, author, version, downloads, dependency list (for meta skills), the web URL, and the full markdown body. JSON format returns the raw skill record. Example: skill_info { "skill": "@kmd_ai/bakefail-pre-flight-auditor" } If you intend to use the skill in this project, follow up with skill_install.

Get the registry's taxonomy: the full set of categories, tags, authors, and types with counts, plus the total number of skills. Use it to discover valid filter values for skill_search. Read-only. Args: - response_format ('markdown' | 'json'): default 'markdown'. Returns: total skill count and faceted lists of categories, types, authors, and tags, each with a count.

Search and browse the skill-builder registry of reusable AI skill prompts. Use this first to discover what skills exist before installing. Supports free-text search plus filters and sorting. Read-only; never writes files. Args: - query (string, optional): free-text search across names, descriptions, tags. - category (string, optional): exact category filter (e.g. "Developer Tools"). - author (string, optional): publisher handle (with or without leading @). - tags (string[], optional): require these tags. - type ('basic' | 'meta', optional): 'meta' = skills that orchestrate dependencies. - sort ('relevant' | 'recent' | 'popular' | 'downloads'): default 'relevant' when a query is given, otherwise 'recent'. - page (number): 1-based page (default 1). - page_size (number): results per page, 1-100 (default 20). - response_format ('markdown' | 'json'): default 'markdown'. Returns: a list of matching skills (id, name, description, category, type, dependency count, downloads). JSON format additionally returns total/page/pageSize. Example: skill_search { "query": "data pipeline", "type": "meta" } To install one of the results, call skill_install with its id.

Autocomplete/typeahead over the registry: returns matching skills, tags, authors, and categories for a (possibly partial) query. Useful for resolving a vague request to a concrete skill id before calling skill_info or skill_install. Read-only. Args: - query (string): partial term to complete (min 1 char). - limit (number): max suggestions, 1-25 (default 10). Returns: a ranked list of suggestions, each tagged with its kind (skill | tag | author | category). Example: skill_suggest { "query": "lit" }

Install a skill from the registry into the current project, writing the files the target coding tool expects. If the skill is a META skill, its full dependency tree is resolved (deduped, cycle-safe) and every required skill is installed alongside it (pass with_dependencies=false to install only the meta skill itself). Where files are written, by target: - claude: appends the skill markdown to CLAUDE.md (under a "## <name>" heading with a hidden skill-id marker; re-installs are skipped if already present). - codex: same, but to AGENTS.md. - cursor: writes .cursor/rules/<slug>.mdc with frontmatter. - file: writes <slug>.md plus a <slug>.json manifest into output_dir. Args: - skill (string): id ("@handle/slug" or "slug") of the skill to install. - target ('auto' | 'claude' | 'codex' | 'cursor' | 'file'): default 'auto', which sniffs the project (.cursor/, .claude/, CLAUDE.md, AGENTS.md, env) and falls back to 'file'. - project_dir (string, optional): project root to install into. Defaults to SKILL_PROJECT_DIR or the server's working directory. - output_dir (string, optional): for target 'file', dir relative to project_dir (default "."). - agents_file (string, optional): override the CLAUDE.md / AGENTS.md path (relative to project_dir) for targets 'claude'/'codex'. - with_dependencies (boolean): install a meta skill's dependencies too (default true). - dry_run (boolean): plan and report what would be written without writing (default false). - response_format ('markdown' | 'json'): default 'markdown'. Returns: the resolved target and project dir, the ordered install plan, and the files written/planned for each skill, plus any dependencies that could not be resolved. This tool WRITES FILES (non-read-only). It never deletes anything. Example: skill_install { "skill": "@kmd_ai/paperdistillery-orchestrated-literature-to-briefing-pipeline" }