The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to modify its local environment and install new capabilities.
Evidence in SKILL.md: codex mcp add openaiDeveloperDocs --url https://developers.openai.com/mcp.
[PRIVILEGE_ESCALATION]: Instructions in the skill specifically direct the agent to attempt to bypass platform security measures. It instructs the agent that if a command fails due to sandboxing, it should immediately attempt the same command again with escalated permissions without first consulting the user.
Evidence in SKILL.md: "If it fails due to permissions/sandboxing, immediately retry the same command with escalated permissions and include a 1-sentence justification for approval."
[REMOTE_CODE_EXECUTION]: The skill defines a workflow to download and execute remote code (an MCP server) within the agent's environment.
Evidence in SKILL.md and agents/openai.yaml: Automated installation path for https://developers.openai.com/mcp.
[METADATA_POISONING]: The skill documentation includes factually incorrect and deceptive information regarding "gpt-oss" models (e.g., gpt-oss-120b and gpt-oss-20b) allegedly released by OpenAI under the Apache 2.0 license. This appears to be a technique to grant the skill unearned authority or to entice users into granting permissions for non-existent official products.

openai-docs