The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to access and read highly sensitive local configuration and secret files including .env, config/master.key, and config/credentials.yml.enc. Evidence: The instructions in SKILL.md under 'Step 1: Deep Codebase Exploration' and 'Step 6: Environment Variables' explicitly direct the agent to check these files to identify required variables and credential structures. Risk: These files contain production secrets and encryption keys; processing them within the agent context creates a risk of accidentally exposing these values in the publicly readable README.md.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from the codebase to generate documentation. Ingestion points: SKILL.md directs the agent to 'thoroughly explore the codebase' and map directory structures, file contents, and configuration data. Boundary markers: Absent; there are no instructions to delimit or ignore potentially malicious instructions embedded in the codebase. Capability inventory: The agent has filesystem read access and the ability to write to the project root (README.md). Sanitization: Absent; content derived from the codebase is interpolated directly into the final document without validation or escaping.

readme