skills/mukul975/anthropic-cybersecurity-skills/building-ransomware-playbook-with-cisa-framework/Gen Agent Trust Hub
building-ransomware-playbook-with-cisa-framework
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The
scripts/agent.pyscript ingests data from a user-provided JSON file via theassesscommand. While the script compares keys against a hardcoded checklist, the values provided in the JSON are interpolated into the script's output without sanitization. If this output is subsequently processed by an AI agent, it could influence the agent's context through maliciously crafted input values. - Ingestion points:
scripts/agent.pyreads a JSON file supplied as a command-line argument. - Boundary markers: None are present in the script's output to delimit external data.
- Capability inventory: The script has the capability to write files to the local directory (
ransomware_playbook_*.md). - Sanitization: The script lacks validation or escaping for the content of the JSON values.
- [METADATA_POISONING]: The skill contains inconsistent authorship and date metadata. The
SKILL.mdfile lists the author as 'mahipal', whereas theLICENSEfile attributes the work to 'mukul975'. Additionally, the license contains a future copyright date (2026). While these discrepancies do not appear to be malicious, they indicate a lack of metadata integrity.
Audit Metadata