The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script claim-verifier.py using command-line arguments derived directly from user input ({content} and {evidence_file}). If the execution environment does not properly sanitize or escape these variables, an attacker could potentially execute arbitrary shell commands by including shell metacharacters in the provided content or file path.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data in the form of marketing text and JSON evidence files.
Ingestion points: The process steps in SKILL.md take {content} and {evidence_file} as inputs to the verification process.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the ingested data as untrusted or to ignore any embedded instructions within that data.
Capability inventory: The skill has access to the local file system (reading and potentially writing to ~/.claude-marketing/) and can execute local Python scripts.
Sanitization: No sanitization or validation logic for the content or JSON structure is described in the skill definition, making it susceptible to instructions embedded within the marketing claims being analyzed.

verify-claims