The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted code snippets provided by users, which is a standard surface for this category given the skill's primary purpose.
Ingestion points: The skill processes code provided by the user in chat or uploaded files as described in SKILL.md.
Boundary markers: The skill uses a structured reporting format to separate analysis from the user content, although it does not define explicit delimiters for input data.
Capability inventory: The skill frontmatter in SKILL.md allows the use of Bash, Read, Grep, and Glob to facilitate codebase inspection.
Sanitization: The skill performs static analysis and does not instruct the agent to execute or evaluate the ingested code, mitigating the risk of injection becoming an execution attack.
[DATA_EXPOSURE]: Reference files such as references/security-checklist.md and SKILL.md contain example strings for hardcoded API keys (e.g., pk_live_abc123xyz). These are explicitly labeled as "BAD" examples for educational training and to help the agent recognize similar patterns in user code; they are not active or sensitive credentials.
[SAFE]: All external URL references target well-known and trusted security organizations, such as OWASP, for the purpose of developer education.

code-review