team-pr
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements strict security controls in
scripts/upload.shfor attaching files to PRs. It validates file types using MIME checks (restricting toimage/*), explicitly refuses symlinks, and ensures all files are within a designated root directory. This prevents the exfiltration of sensitive files such as credentials or private keys. - [INDIRECT_PROMPT_INJECTION]: The skill treats PR bodies and external notes as untrusted data. Ingestion points include
gh pr viewoutput and local review records. Thescripts/splice.mjsscript uses a 'refusal-by-default' model to scan for unmodeled markdown or HTML constructs, ensuring that content authored by others cannot be used to trick the agent into executing unintended commands or corrupting the PR body. It also includes sanitization by escaping raw HTML and markdown links before interpolation. - [COMMAND_EXECUTION]: The skill uses the GitHub CLI (
gh) for all repository interactions. Command arguments are carefully handled, using files or quoted heredocs to prevent shell injection vulnerabilities. Path resolution scripts likeresolve-pr.shinclude validation against unexpected characters and enforce strict regex patterns for input. - [SAFE]: The skill follows security best practices, such as implementing a 'lost-update' guard to detect concurrent modifications to PR bodies, ensuring that the agent does not overwrite changes made by others during its operation. It also prioritizes privacy by using file basenames rather than absolute paths in PR descriptions.
Audit Metadata