The Agent Skills Directory

[COMMAND_EXECUTION]: The pre-hook script executes a shell command echo "💻 Coder agent implementing: $TASK". This allows for arbitrary command execution if the $TASK environment variable contains shell metacharacters such as backticks or command substitution syntax.
[COMMAND_EXECUTION]: The post-hook script runs npm run lint --if-present. This executes scripts defined in the project's package.json file, which could be malicious if the agent is working on an untrusted repository.
[PROMPT_INJECTION]: The skill's use of the $TASK variable to influence hook logic creates an indirect prompt injection surface. Evidence: 1. Ingestion points: The $TASK environment variable in the pre hook. 2. Boundary markers: Absent. 3. Capability inventory: Host shell execution via echo, grep, and npm. 4. Sanitization: None identified for the $TASK input.

agent-coder