skill-ledger
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on a local CLI tool,
agent-sec-cli, to perform its primary functions. It executes commands such ascheck,show,scan,decide, andcertifyto manage the security lifecycle of other agent skills. These operations are restricted to the local filesystem and the specific utility provided for security management. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface because its core purpose is to read and analyze files within external skill directories (
SKILL_DIR). A malicious skill could potentially include content designed to confuse the security scanner or the auditing agent. - Ingestion points: Files (scripts, markdown, and config) located within the directory targeted for scanning are read during the
skill-vetterprotocol execution. - Boundary markers: The skill implements a strict four-stage vetting framework (
skill-vetter-protocol.md) that explicitly lists prompt injection patterns to detect and block, creating a robust boundary against adversarial content. - Capability inventory: The skill uses
agent-sec-clifor state management and findings certification, and standard shell utilities (grep,ls,read) for file analysis. - Sanitization: Audit results are structured into a normalized JSON schema (
NormalizedFinding[]) before being recorded, preventing raw malicious content from being interpreted as agent instructions.
Audit Metadata