The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to use the --full-auto and --dangerously-bypass-approvals-and-sandbox flags when invoking the Codex CLI. The documentation within the skill explicitly states these flags are used to "skip all approval prompts" and "enable autonomous execution," which constitutes an intentional override of the platform's safety and oversight mechanisms.
[COMMAND_EXECUTION]: The workflow relies on the bash tool to perform tool discovery (which codex) and launch multiple parallel analysis tasks in the background. It passes complex natural language prompts directly into these shell-executed commands, increasing the risk of command-related issues when processing project metadata or content.
[INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: The skill is designed to crawl and analyze the entire project filesystem (e.g., App.tsx, package.json, pyproject.toml, API router files) to feed dimensions like navigation density and API health. 2. Boundary markers: The agent prompts for exploration tasks do not include delimiters or instructions to treat the analyzed code as untrusted data, allowing malicious instructions within the project files to potentially influence agent behavior. 3. Capability inventory: The skill has broad capabilities including arbitrary shell command execution via the bash tool and the ability to launch recursive sub-agent tasks. 4. Sanitization: No validation or escaping logic is present to sanitize code snippets or configuration values before they are interpolated into analysis prompts.

product-analysis