skill
Pass
Audited by Gen Agent Trust Hub on Feb 20, 2026
Risk Level: SAFE
Full Analysis
- [DYNAMIC_EXECUTION] (LOW): The skill utilizes 'Empathic Templates' (.tmpl files) to generate new skill documentation and metadata. This involves the LLM interpolating values into predefined structures, which is a core function of a meta-skill.
- [EXTERNAL_DOWNLOADS] (LOW): Documentation in EXPORTS-PROTOCOL.md provides instructions for downloading and installing skill bundles from GitHub. As the source is a known repository and these are user-facing instructions rather than automated background scripts, the risk is minimal.
- [INDIRECT_PROMPT_INJECTION] (LOW): The skill defines a 'Multi-Tier Persistence' model that involves reading data from logs (LOG.md) and data islands. Ingestion points: LOG.md and YAML data islands. Boundary markers: Uses #object-id addressing for data islands to isolate specific YAML blocks within logs. Capability inventory: Uses read_file and write_file; defines the protocol for child skills to request run_terminal_cmd. Sanitization: Relies on the Delegation Object Protocol (DOP) for structured resolution, though explicit data sanitization logic is not detailed.
- [COMMAND_EXECUTION] (LOW): The EXPORTS-PROTOCOL.md and SKILL.md.tmpl files describe workflows involving shell commands like tar, python3, and grep. These are documented procedures for packaging or specialized terminal skills, not hidden execution paths within the meta-skill itself.
Audit Metadata