The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the vinceamstoutz/symfony-security-auditor package and various AI platform bridge modules (e.g., symfony/ai-anthropic-platform) from the public Composer registry.
[DATA_EXFILTRATION]: The auditing process involves reading sensitive project files, including source code (.php), templates (.twig), and configuration (.yaml, .xml), and sending their content to external AI platforms for processing. While these platforms (Anthropic, OpenAI, Google) are recognized providers, this behavior represents a disclosure of the target application's entire source tree and configuration to external entities.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze untrusted data (project source code) via LLMs, which introduces a vulnerability to indirect prompt injection. Adversarial code or comments within the audited project could be crafted to influence the behavior of the 'Attacker' or 'Reviewer' agents.
Ingestion points: Recursively scans all .php, .twig, .yaml, and .xml files within the target project directory.
Boundary markers: The skill definition does not specify the use of clear delimiters or instructions to the LLM to ignore potentially adversarial commands embedded in the scanned source code.
Capability inventory: The skill executes file system reads and performs network requests to third-party AI APIs.
Sanitization: No explicit sanitization, filtering, or escaping of the project content is described before it is passed to the AI agents.

symfony-security-auditor