The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a specialized persona and research workflow without any attempts to bypass safety filters or override system guardrails.\n- [DATA_EXFILTRATION]: No network access, API calls, or instructions to access sensitive credential files were found. The scope is confined to the analysis of project source code.\n- [REMOTE_CODE_EXECUTION]: The skill does not involve downloading external scripts, installing third-party packages, or executing remote code.\n- [COMMAND_EXECUTION]: There are no instructions for the agent to execute shell commands, manage system services, or perform other dangerous OS-level operations.\n- [SAFE]: The core invariants and evidence standards in the skill promote accuracy and transparency by forcing the agent to verify claims against the implementation, which is a positive safety behavior for analysis tasks.

wiki-researcher