The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines clear boundaries for sub-agents, using specific prompts that explicitly prohibit delegation or unauthorized actions. No patterns attempting to bypass safety filters or override system instructions were found.- [DATA_EXFILTRATION]: While the skill accesses repository content and code paths for analysis, it contains no network operations, external data transfer commands, or access to sensitive credential files.- [COMMAND_EXECUTION]: The workflow involves standard software development activities such as running local tests and repository checks. There is no evidence of arbitrary command execution, privilege escalation, or persistence mechanisms.- [SAFE]: The use of 'fresh-context specialists' is a significant security feature. By providing child agents with raw data while withholding the main agent's implementation plan, the skill reduces the risk of successful indirect prompt injection and prevents confirmation bias.

preflight