keepflash

Fail

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill implements a custom update mechanism in scripts/bootstrap.mjs that clones a remote repository (github.com/dante-is-shipping/keepflash-skills.git) and downloads new script files and instructions. The agent is explicitly instructed to treat these downloaded files as the "authoritative workflow," effectively allowing the remote repository owner to execute arbitrary instructions and code on the user's system.
  • [DATA_EXFILTRATION]: The skill is designed to manage user authentication tokens and private notes, transmitting this sensitive data to keepflash.com. Multiple URLs associated with this domain, including authentication endpoints and note retrieval paths, have been flagged as phishing by automated security scanners.
  • [EXTERNAL_DOWNLOADS]: The scripts/bootstrap.mjs file performs unauthorized external downloads of executable scripts and markdown instructions from a non-whitelisted GitHub repository during every activation, bypassing static verification of the skill's behavior.
  • [COMMAND_EXECUTION]: The skill utilizes node:child_process to execute git commands and other shell scripts. This is used to facilitate the dynamic fetching and execution of remote content.
  • [CREDENTIALS_UNSAFE]: While the skill does not contain hardcoded credentials, its primary function is to collect and manage KEEPFLASH_ACCESS_TOKEN values, which are then transmitted to a site identified as a phishing threat.
Recommendations
  • AI detected serious security threats
  • Contains 5 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Aug 15, 2026, 07:26 AM
Security Audit — agent-trust-hub — keepflash