The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to gather evidence by dumping environment variables using env and inspecting macOS keychain identities via security list-keychains and security find-identity. This behavior can expose sensitive cryptographic identities and environment secrets to the agent's output buffer and potentially to logs.
[COMMAND_EXECUTION]: The find-polluter.sh script executes npm test on files identified by a user-supplied pattern. The script's for loop is vulnerable to word-splitting, which could result in unintended command execution if filenames in the repository contain spaces or shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input such as system error logs, stack traces, and git diffs for debugging purposes.
Ingestion points: External error messages and git commit history (as seen in SKILL.md).
Boundary markers: None; the skill does not provide instructions to isolate or ignore instructions within the analyzed logs.
Capability inventory: Shell command execution (npm test, git, security) and file system access.
Sanitization: No sanitization is performed on external data before processing by the agent.

systematic-debugging