ai-dev-issue
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions include phrases typically associated with prompt injection (e.g., "ignore previous instructions"), but these are used defensively. The agent is instructed to identify these patterns in external data and treat them as findings to be reported rather than instructions to be followed.
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from source code, issues, and templates, creating an attack surface for indirect prompt injection. This is mitigated by the "Untrusted Content" section in
SKILL.md. - Ingestion points: Source files, documentation, issue templates, and existing issue bodies retrieved from project trackers.
- Boundary markers: The instructions explicitly separate read content from authority-bearing project configuration files and user messages.
- Capability inventory: File writes (within project scope), network API calls (GitHub/GitLab), and local execution of
gh,glab,python3, andcurlcommands. - Sanitization: The skill mandates that any commands or directives discovered in external data must be reported and ignored.
- [CREDENTIALS_UNSAFE]: The skill handles sensitive tokens for GitHub and GitLab APIs. It implements a dedicated security protocol in
references/forge-credentials.mdthat requires explicit user confirmation before credentials can be sent to any host not already trusted or derived from the repository's git origin. - [DYNAMIC_EXECUTION]: The skill uses
python3 -cone-liners for utility tasks like URL encoding and JSON parsing. These are used to safely prepare data for API requests and do not execute untrusted external code. - [COMMAND_EXECUTION]: The skill invokes standard CLI tools (
gh,glab,curl) to manage issues. It includes strict constraints on writing to system paths and requires user approval for writing outside the working directory.
Audit Metadata