The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to collect local repository evidence, including files, git diffs, and branch metadata, and transmit them to an external service (ChatGPT). Although it targets a well-known service and includes instructions to exclude secrets, the core function involves the automated transfer of local code to a remote environment.
[COMMAND_EXECUTION]: The skill utilizes a codex tool to execute shell commands locally for the purposes of evidence collection and applying fixes. It supports an execution mode (--ask-for-approval never) that allows for unattended command execution, though this is protected by an internal 'Local Codex Gate' requiring user selection.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests findings from an external AI and uses them to drive local repository changes.
Ingestion points: External ChatGPT responses are captured into review.md and processed by the agent to determine fixes.
Boundary markers: The skill does not define structural delimiters (like XML tags or specific markers) to isolate external input from the agent's internal instructions.
Capability inventory: The agent has the capability to write files and execute arbitrary shell commands via the codex exec command.
Sanitization: The skill relies on 'local verification' by the agent to validate findings, but lacks programmatic sanitization of the externally sourced content.

chatgpt-review-bridge