agent-skill-finder

Activates ultra-compressed response mode that reduces token usage by approximately 75% by eliminating filler words, articles, and pleasantries while preserving full technical accuracy and substance. Invoke when the user requests "caveman mode", "talk like caveman", "less tokens", "be brief", or uses the /caveman command.

Builds an interactive knowledge graph from any mix of code, docs, papers, images, and video/audio files. Extracts concepts, relationships, call graphs, and design rationale via AST (code) and Claude subagents (docs/media). Outputs interactive HTML, queryable JSON, and a plain-language audit report. Supports incremental updates, cross-repo merging, MCP server mode, and always-on assistant hooks for Claude Code, Cursor, Gemini CLI, Codex, and others.

Conducts a structured, relentless interview of the user's plan or system design, resolving each branch of the decision tree until full shared understanding is reached. Provides a recommended answer for each question asked. Invoke when the user wants to stress-test a plan, validate architecture decisions, or explicitly requests "grill me".

Guides the creation of new agent skills with correct structure, progressive disclosure, and bundled reference resources. Produces a SKILL.md with a routing-optimized description, optional reference files, and utility scripts where deterministic operations are needed. Invoke when the user wants to create, write, or build a new skill file.

Parses a git diff or patch file and produces a structured summary of changes: files modified, lines added and removed per file, a plain-English description of what changed, and a suggested conventional commit message following the feat/fix/chore/refactor/docs prefix convention.

Scans a git repository or file path for hardcoded secrets, API keys, tokens, and credentials using Gitleaks SAST rules. Returns findings with file path, line number, secret type (AWS key, GitHub token, Slack webhook, etc.), and a redacted preview. Returns clean=true when no secrets are detected.

Extracts structured action items from a meeting transcript or notes, identifying the task description, assigned owner name, due date or relative deadline, and priority level for each item. Returns a sorted action_items list and a one-paragraph meeting summary.

Validates an OpenAPI 3.x or Swagger 2.0 specification file (YAML or JSON) against the official schema, returning a list of validation errors with path, message, and severity. Checks for missing required fields, invalid $ref references, malformed response schemas, and security scheme misconfigurations. Returns valid=true when the spec has zero errors.

Retrieves the top-k most semantically relevant document chunks for a natural language query from a pre-built vector index (Chroma, Pinecone, or Weaviate). Accepts a query string and index connection config, returns ranked chunks with source file, similarity score, and chunk text for use in RAG pipelines.