The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted session transcripts which may contain malicious instructions. It implements significant mitigations by explicitly instructing sub-agents to ignore directives within the transcript and treat the data as untrusted.
[COMMAND_EXECUTION]: While the skill orchestrates agents that may use tools (shell, MCP, etc.), it mandates explicit user approval for every proposed external write or skill modification.
[REMOTE_CODE_EXECUTION]: The skill mentions external models and MCP tools, but does not provide mechanisms for unverified code execution. Access to external data (tickets, traces) is scoped to context referenced in the transcript.

reflect