just-fucking-cancel
Pass
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill analyzes bank transaction CSVs (Apple Card, Chase, Mint) provided by the user. This involves processing highly sensitive financial data, creating a potential exposure surface for financial information.
- [COMMAND_EXECUTION]: The workflow involves using browser automation to interact with external service websites to perform subscription cancellations, which is a high-privilege operation.
- [EXTERNAL_DOWNLOADS]: Documentation in PUBLISH.md references the installation of the 'clawdhub' CLI tool via the npm package registry.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted transaction data from external CSV files. 1. Ingestion points: Bank CSV exports uploaded by the user (described in SKILL.md). 2. Boundary markers: Absent; no delimiters are used to separate transaction data from agent instructions. 3. Capability inventory: Browser automation and HTML report generation. 4. Sanitization: Absent; transaction data is used directly in the cancellation workflow without validation.
Audit Metadata