The Agent Skills Directory

[SAFE]: The skill is a legitimate development utility for architecting and scaffolding LangGraph CLI applications.
[SAFE]: Documentation and recipes (e.g., AWS assistant) include robust security guidance, such as stripping managed flags to prevent account escaping and using environment isolation for shell-based tools.
[SAFE]: External dependencies are restricted to well-known, established libraries from the LangChain and React ecosystems.
[SAFE]: The skill implements a mandatory review gate (/codex:review) to ensure generated code is audited before completion.
[SAFE]: Data handling for telemetry (LangSmith) is gated by a mandatory privacy disclosure and user confirmation step.
[COMMAND_EXECUTION]: The skill scaffolds agents capable of executing bash commands and AWS CLI commands. It provides comprehensive safety mechanisms, including command classification (read vs. write), mandatory human-in-the-loop approval for mutations, and strict environment variable stripping to prevent privilege escalation or credential leakage.
[PROMPT_INJECTION]: The skill scaffolds agents that ingest content from a local AGENTS.md file (see references/chat-aws-repo.md). Ingestion points: AGENTS.md file content is loaded and appended to the system prompt; Boundary markers: The content is wrapped in a markdown fenced block; Capability inventory: The scaffolded agents can execute shell commands via a bash tool and AWS commands via an aws_cli tool; Sanitization: Content is loaded and trimmed but not explicitly sanitized for prompt instructions.

langgraph-adk