The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides detailed instructions on how to implement security rails for command execution. It specifically includes code examples for creating command denylists using regex to block dangerous operations like rm -rf /, sudo, chmod, chown, and dd. It also highlights advanced security techniques, such as using word boundaries (\b) in patterns to prevent bypasses via command chaining.
[EXTERNAL_DOWNLOADS]: The skill references established tools and libraries, including the official @anthropic-ai/claude-code SDK and the codex CLI. These resources are from well-known and trusted organizations and are essential for the integration tasks the skill describes.
[PROMPT_INJECTION]: The skill focuses on architectural patterns for session management and does not contain any instructions that attempt to bypass safety guidelines, extract system prompts, or override agent constraints. It encourages the use of developer instructions to define agent roles safely.
[DATA_EXFILTRATION]: The described network communication patterns (JSON-RPC over stdio, SSE streams for chat updates, and standard HTTP endpoints) are integral to the functionality of the sidecar backend being built. There is no evidence of unauthorized data transfer or exfiltration to untrusted domains.

multi-turn-runtime-adapters