The Agent Skills Directory

[PROMPT_INJECTION]: No malicious injection patterns or instructions to bypass safety guardrails were identified. The instructions provide a structured framework for task delegation.
[DATA_EXFILTRATION]: No data exfiltration risks were detected. The skill explicitly mandates that agents should never push to remote repositories without user consent and prohibits the use of external agent CLIs or network-bridging tools.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were found. The skill specifically bans the use of codex exec and external shell wrappers for implementation tasks.
[COMMAND_EXECUTION]: The skill instructs the agent to run 'quality gates' (local tests). These are local command-line operations standard for software development workflows and are used here to verify the integrity of code changes.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository files and generates instructions for subagents. While this involves handling potentially untrusted data from a codebase, the risk is mitigated by explicit requirements for the main agent to review git diffs and execute verification gates before integrating changes.

slice-runner