GET /v1/agent/search?capability=web_search
SKILL.md v2 also accepted as compatibility input.
Runtime context window management for AI agents — token budget estimation, message pruning, relevance reranking, and context splitting (zero dependencies)
Extract action items, decisions, owners, and deadlines from meeting notes — Korean/English bilingual, zero dependencies
Multi-mode text summarization toolkit — extractive, bullet, TL;DR, and summary comparison with zero external dependencies (Korean/English)
Track, check, and manage API rate limit state for agents calling external APIs — token bucket simulation, backoff strategy, and usage reporting
Agent conversation memory compactor: preserves decisions/identifiers/preferences using heuristics, compresses the rest by frequency/position scoring. No LLM calls, zero external dependencies. Korean decision expressions supported.
Prompt template engine with 5 actions: render, parse, validate, list_variables, audit. Zero external dependencies. Korean variable name support, conditional/loop blocks, HTML/SQL/Shell escaping, missing variable detection.
AI tool-call validator with 5 actions: validate_input, validate_output, dry_run, sanitize, audit_schema. Zero external dependencies. JSON Schema Draft 7 self-implemented. Detects path traversal, prompt injection, and type mismatches with auto-fix hints.
LLM API cost estimator with 5 actions: estimate, compare_models, project_budget, simulate_cache, audit_pricing_freshness. Zero external dependencies. Covers Claude/GPT-4o/Gemini pricing, Korean token heuristics, cache hit/miss simulation.
LLM prompt static analysis, auto-optimization, security/quality audit (15+ rules), and A/B comparison — zero external deps, Korean-first, agent-native fix_hint
텍스트의 LLM 토큰 수를 추정합니다. tiktoken 기반 GPT/Claude 정확 계산 및 단어 기반 빠른 추정을 지원합니다.