The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes defensive instructions that command the agent to treat repository content, comments, and metadata as data rather than instructions, effectively mitigating potential indirect prompt injection attacks.
[DATA_EXFILTRATION]: While the skill accesses the local ~/.endorctl/config.yaml file, it incorporates explicit constraints to prevent the leakage of credentials or tokens, instructing the agent to only extract non-sensitive namespace information.
[COMMAND_EXECUTION]: The skill utilizes local build tools (e.g., Maven, npm) and CLI utilities (e.g., git, gh) for validation and remediation tasks. All such executions are governed by a safety contract requiring explicit user approval for any mutating actions.

sca-remediation