The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions to bypass security constraints, specifically telling the agent to 'Run Context7 CLI requests outside Codex's default sandbox' and to 'rerun it outside the sandbox' in case of failures. This is a direct attempt to circumvent safety boundaries.\n- [EXTERNAL_DOWNLOADS]: The skill mandates the installation of an unverified third-party CLI package (ctx7) from the public NPM registry. This package does not originate from a trusted vendor, and its behavior is not verified.\n- [COMMAND_EXECUTION]: The skill requires the agent to execute shell commands for global package installation and to run the ctx7 tool with dynamically generated queries, increasing the risk of command injection and unauthorized system modification.\n- [PROMPT_INJECTION]: The skill processes documentation content from external sources, presenting an indirect prompt injection surface where malicious instructions could be embedded in the retrieved data.\n
Ingestion points: Documentation content retrieved from the ctx7 docs command output.\n
Boundary markers: None; the instructions do not specify any delimiters to separate documentation from agent instructions.\n
Capability inventory: Global package installation (npm install -g) and CLI tool execution (ctx7).\n
Sanitization: None; the skill does not include steps to sanitize or validate the external content before it is processed by the agent.

find-docs