image-to-github-repo
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted image content (screenshots, card-news, SNS posts) to extract project metadata, which represents a potential surface for indirect prompt injection.
- Ingestion points: User-provided images and captions analyzed in Step 1 of SKILL.md.
- Boundary markers: Absent in the initial extraction instructions.
- Capability inventory: Includes network access via
WebFetchand thescripts/resolve_repo.pytool, as well as file system read/write via Bash tools. - Sanitization: The
scripts/resolve_repo.pyscript provides a layer of sanitization by normalizing text, filtering keywords against a static stopword list, and usingurllib.parse.quotefor API parameter construction. - [SAFE]: The skill interacts with the official GitHub API (
api.github.com) to verify repository existence and metadata. - Evidence: The
scripts/resolve_repo.pyscript queriesapi.github.comto resolve project names and URLs. - Security Measure: The script safely handles
GITHUB_TOKENby ensuring it is only transmitted toapi.github.comand utilizes a customSameHostRedirecthandler to prevent token leakage through malicious redirects. This follows security best practices for API client implementation.
Audit Metadata