The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the terminal tool to execute shell commands for git operations and the Codex CLI. It explicitly recommends the use of the '--yolo' flag for faster performance, which by design bypasses the local sandbox and removes the requirement for user approval before file changes are committed.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the global installation of the '@openai/codex' CLI via the npm registry. While this tool originates from a well-known technology provider, it equips the agent with extensive capabilities for local code execution.
[REMOTE_CODE_EXECUTION]: The core functionality of the skill involves using Codex to execute tasks autonomously. This process involves the agent translating natural language prompts—which may be sourced from external repository content—into locally executed code and shell operations.
[DATA_EXFILTRATION]: By incorporating commands such as 'git push' and providing instructions for the GitHub CLI ('gh'), the skill grants the agent the technical capability to transmit local workspace data to external repositories.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted external data.
Ingestion points: External data enters the agent context through 'git clone' and 'gh pr checkout' operations during pull request reviews.
Boundary markers: The instructions do not define boundary markers or 'ignore' directives to prevent the agent from obeying instructions embedded within the external code or metadata.
Capability inventory: The agent maintains access to the 'terminal' tool for shell execution and 'codex' for automated file modification.
Sanitization: No sanitization or validation protocols are established for the content retrieved from remote repositories before it is processed by the autonomous agent.

codex