The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to build and execute shell commands (e.g., media video --prompt "TEXT") where the parameters are directly populated from user input. This pattern presents a significant risk of command injection if the input text contains shell metacharacters that are not properly sanitized or escaped before the tool is invoked.
[DYNAMIC_EXECUTION]: The skill implements a dynamic discovery pattern to locate its executable by reading from a JSON manifest file (~/.codex/codex-router/install-manifest.json). Executing code from paths resolved at runtime is a risk factor as it could be redirected to unauthorized binaries if the manifest file is modified by a malicious process.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external inputs including user-provided prompts and external image URLs, creating a surface for adversarial content to influence the agent's actions.
Ingestion points: The --prompt and --text flags accept raw user text, and the --image flag accepts external URLs or local file paths.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill uses the shell tool for command execution and has the ability to write files to the local filesystem using the --out flag.
Sanitization: There is no mention of input validation, escaping, or filtering to prevent malicious payloads in the external sources from affecting the host system.

codex-router-media