The Agent Skills Directory

[Prompt Injection] (MEDIUM): The skill is vulnerable to indirect prompt injection via local file ingestion.
Ingestion points: SKILL.md Step 2 reads the contents of DESIGN.md from the project directory.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands, using only a simple markdown header.
Capability inventory: The skill is granted Read and Write permissions, allowing it to access local files and write results to the filesystem.
Sanitization: There is no evidence of content sanitization for the ingested design data before interpolation.
[External Downloads] (MEDIUM): The skill installation utilizes a source organization not included in the trusted list.
Evidence: npx add-skill google-labs-code/stitch-skills found in README.md.
Status: The organization google-labs-code is not an exact match for the trusted google or google-gemini organizations.

enhance-prompt