The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze user-supplied code, which constitutes untrusted data. This interaction model presents a theoretical surface for indirect prompt injection where malicious instructions could be embedded within the code or comments being reviewed. However, the skill's instructions are explicitly defensive and focused on identifying and mitigating such vulnerabilities.
Ingestion points: Processes user-provided source code during implementation and review workflows as defined in SKILL.md and references/tier2-review.md.
Boundary markers: While it lacks strict technical delimiters for input, it requires the agent to state security assumptions and document simplified requirements, which acts as a logical boundary.
Capability inventory: The skill is capable of generating source code, providing technical reviews, and recommending architectural changes based on user input.
Sanitization: The instructions focus on the agent ensuring the output code contains sanitization and validation logic, though the agent itself relies on its core safety filters for processing the input.
[EXTERNAL_DOWNLOADS]: The skill references several reputable security tools in references/tier2-implementation.md (§2.5) as recommendations for development pipelines. These include well-known projects such as gitleaks, trufflehog, and detect-secrets. These references are purely informational and are provided to help users implement security best practices.

securing-code