budget-optimizer
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: Accesses the local filesystem at
~/.claude-marketing/to retrieve brand profiles, guidelines, and SOPs. While these appear to be application-specific configuration files, accessing paths in the user home directory is a sensitive operation.\n- [COMMAND_EXECUTION]: Executes a local Python script (budget-optimizer.py) and passes user-provided channel data and budget amounts as command-line arguments. This pattern relies on the execution environment to properly sanitize and escape input to prevent command injection.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external data from brand profiles and compliance rules which could contain instructions that influence agent behavior.\n - Ingestion points: Reads
profile.json,_manifest.json, and compliance rules from the~/.claude-marketing/directory structure.\n - Boundary markers: No explicit delimiters or boundary instructions are mentioned to separate the ingested brand data from the agent's core instructions.\n
- Capability inventory: Capability to execute shell commands via a Python script and read various configuration files from the local filesystem.\n
- Sanitization: The instructions do not specify any validation or sanitization of the brand data before it is loaded into the agent's context.
Audit Metadata