GET /v1/agent/search?capability=web_search
SKILL.md v2 also accepted as compatibility input.
Parses SQL DDL statements (CREATE TABLE, INDEX, CONSTRAINT) into structured JSON — tables, columns, types, constraints, foreign keys. Supports MySQL, PostgreSQL, BigQuery, Snowflake, MSSQL, SQLite with auto-dialect detection.
Parse and validate TOML config files, extract key paths, and convert to JSON. Supports pyproject.toml, Cargo.toml, config.toml. Zero external dependencies (Python 3.11 stdlib tomllib).
Validates PHP code syntax using php -l (built-in lint). Returns structured JSON with error line numbers, messages, and types. Designed as a deterministic validation anchor for agent self-repair loops.
Open a SQLite .db file and run schema inspection, safe SELECT execution, EXPLAIN QUERY PLAN analysis, and index/stats queries. Zero external dependencies (Python stdlib sqlite3 only).
Convert JSON Schema / DDL / CSV headers to PySpark StructType code, and reverse-convert PySpark schemas to Avro or JSON Schema. No Spark runtime required.
Validates HTTP request objects (method, URL, headers, body) against RFC standards and OpenAPI specs, returning structured error lists. Prevents request-generation bugs before execution in agent API automation loops.
Send emails securely without exposing SMTP passwords, powered by MGC Blackbox.
Parse unified git diff text into structured per-file data: change type, line counts, hunk positions, language detection. Zero external dependencies.
Extract entities and relations from text and return node-edge graph data. Zero external dependencies.
Persistent key-value memory store with tags and FTS5 full-text search for agents across sessions. SQLite + trigram tokenizer. Zero external dependencies.
Deterministic JSON state machine for multi-step agent workflow tracking — checkpoint, restore, and failure recovery without LLM hallucination
Given a vulnerability type (SQL Injection, XSS, SSRF, etc.), returns OWASP/CWE-based fix patterns, safe code alternatives, and a validation checklist — no external dependencies.
Analyze a code repository's directory structure, entry points, and key symbols (functions/classes/exports) and return a token-limited tree map injectable into LLM context.
Extract text, tables, images, and metadata from DOCX files, and generate new DOCX from templates.
Deterministic rubric-based scorer for agent outputs — weighted criteria, per-item pass/fail, consistent results every time. No LLM needed.
Deterministic JSON Schema validator for LLM outputs. Returns full violation list with JSON Pointer paths ($.actions[0].priority), fix hints per keyword, and summary — self-correction loop anchor.
Stitch multiple images vertically, horizontally, or in a grid layout with resize, padding, and background color options. Pillow-only, no external APIs.
Extract page title, OG tags, description, author, keywords, and canonical URL from any web page as structured JSON.
Extract action items, assignees, deadlines, and decisions from meeting transcripts — USK v3 native, Korean/English, zero dependencies
Runtime context window management for AI agents — token budget estimation, message pruning, relevance reranking, and context splitting (zero dependencies)