semantic-core-architect
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The Validation section in SKILL.md provides commands to run local Python scripts (lint_production_skill.py and lint_skill_cluster.py) for auditing skill edits. These scripts reference environment variables like $CODEX_HOME and $HOME, which are standard for internal development toolchains.
- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection. 1. Ingestion points: User-provided site goals, audience descriptions, and market constraints in SKILL.md. 2. Boundary markers: No delimiters or ignore instructions are present for the ingested data. 3. Capability inventory: The skill specifies local python3 command execution for linting and writes to local YAML files. 4. Sanitization: There is no explicit sanitization of external input before it is processed into templates.
- [SAFE]: The skill includes strong safety guidelines for data integrity, explicitly instructing the agent to mark metrics as unknown rather than inventing data, and it maintains clear scope boundaries by deferring final URL and canonical decisions to other skills.
Audit Metadata