pstack-create-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill identifies and executes the repository's own build, launch, and health-check commands (e.g., from Makefiles, package scripts, or README instructions) to establish a functional environment for verification testing.
- [DYNAMIC_EXECUTION]: The skill generates a project-local verification skill (
SKILL.md) and then executes its instructions end-to-end—including server setup and UI/CLI drivers—to prove the generated logic works as intended. - [INDIRECT_PROMPT_INJECTION]: The skill's 'Interview' phase involves reading untrusted data from the analyzed repository (READMEs, code comments, configuration files), which is used to generate and execute shell commands.
- Ingestion points: Reads repository source files, documentation, and configuration metadata to understand application behavior.
- Boundary markers: None; the instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the ingested codebase.
- Capability inventory: High; the agent is authorized to execute arbitrary shell commands for building and running the app, as well as writing new skill files to the filesystem.
- Sanitization: None; the skill relies on the model's interpretation of the codebase without specific validation or escaping of the extracted commands.
Audit Metadata