The Agent Skills Directory

[COMMAND_EXECUTION]: The maintenance script scripts/review-openai-models.mjs uses execFileSync to run git diff --name-only. This is a standard administrative operation to track file changes within the repository and does not pose a risk.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to api.openai.com to fetch updated model information. These requests are restricted to official OpenAI endpoints. Furthermore, the skill's web-search tool uses a strict domain allowlist (developers.openai.com, platform.openai.com) to prevent ingestion of untrusted external content.
[DATA_EXPOSURE]: The skill uses OPENAI_API_KEY to authenticate with the OpenAI API. This is standard practice for model-integrated automation. There are no signs of credential exfiltration or access to sensitive local environment files.
[DYNAMIC_EXECUTION]: While the script processes JSON data from a remote API, it uses a strict JSON schema for validation and escapes markdown characters before writing them to disk, effectively mitigating potential injection or formatting attacks.
[INDIRECT_PROMPT_INJECTION]: The skill has a small attack surface via its documentation update script which reads external documentation. However, it implements robust boundary markers and validation logic, including source verification and structured output parsing, which are effective mitigations. The severity is considered low/safe.

allocate-agent-models