The Agent Skills Directory

[SAFE]: No malicious patterns or significant security risks were identified during the audit.- [EXTERNAL_DOWNLOADS]: The skill automates the installation of required utilities including Bun, jq, and yq, as well as the vendor's own CLI tool. These are fetched from official project sites and the author's own GitHub repositories, adhering to the trust policy for vendor-owned resources.- [REMOTE_CODE_EXECUTION]: Installation scripts for the Semantius CLI and Bun are executed directly via shell pipes (e.g., curl | bash). These operations are considered safe as the sources are the primary vendor (semantius) or recognized well-known services (bun.sh).- [COMMAND_EXECUTION]: The skill executes local administrative commands using the semantius CLI to query, list, and back up the semantic model. These actions are consistent with the skill's stated purpose.- [DATA_EXFILTRATION]: The skill manages the SEMANTIUS_API_KEY by prompting the user and storing it in a local .env file. This follows established safe practices for credential management in the execution environment.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted blueprint and specification data from URLs or workspace files. Evidence chain: 1. Ingestion points: SKILL.md Step 2 and Step 6.1. 2. Boundary markers: Absent. 3. Capability inventory: Shell execution (bash), file system writes, and network operations (curl). 4. Sanitization: Absent. Although this represents an attack surface, it is inherent to the skill's function and the severity is assessed as low.

semantius-admin