recipe-add-integration-tests
Fail
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: In Step 1, the skill executes a bash script containing
ls $ARGUMENTS. This directly interpolates user-provided arguments into a shell command without sanitization or quoting, allowing for arbitrary command execution through shell metacharacters such as semicolons or pipes. - [PROMPT_INJECTION]: The skill ingests user-provided documentation to drive automated code generation, which introduces an indirect prompt injection surface.
- Ingestion points: Document paths and content provided via the
$ARGUMENTSvariable and thedocs/folder (SKILL.md, Step 1). - Boundary markers: Absent. No delimiters or specific instructions are provided to distinguish document content from system instructions when passed to sub-agents.
- Capability inventory: Local command execution (
ls), file creation/writing (Step 3), and repository commits (Step 8). - Sanitization: There is no filtering or validation of the content within the design documents before they are processed by sub-agents.
- [PROMPT_INJECTION]: The skill implements a manual confirmation gate in Step 3 ('[STOP — BLOCKING]'), which requires the user to review the generated task files before implementation begins. This provides a point for human review, mitigating some of the risk of automated abuse.
Recommendations
- AI detected serious security threats
Audit Metadata