The Agent Skills Directory

[COMMAND_EXECUTION]: The shell script scripts/load.sh executes platform-specific binaries (ch-load-*) from the bin/ directory. These pre-compiled binaries are opaque and their source code is not provided for security auditing, which is a high-risk practice as they can perform unauthorized actions with the user's system permissions.
[COMMAND_EXECUTION]: The skill uses the exec "$LOAD_BIN" "$@" pattern to run the selected binary with all arguments passed from the agent. This creates a direct execution path for parameters that may be derived from external user input.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection. Ingestion points: A conversation UUID is extracted from the user's request as defined in SKILL.md. Boundary markers: The extracted UUID is enclosed in double quotes in the bash command template. Capability inventory: The skill executes local binaries and retrieves full message transcripts from a history database. Sanitization: No input validation or sanitization is performed on the arguments within the shell script; the skill relies entirely on the agent's ability to correctly extract a safe UUID string.

load-conversation