bear-notes-hardened

Warn

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's configuration specifies the installation of a binary tool from an external GitHub repository (github.com/tylerwince/grizzly) using the Go package manager.
  • [REMOTE_CODE_EXECUTION]: The installation process involves downloading and executing code from an unverified remote source, which could lead to arbitrary code execution if the source is compromised.
  • [CREDENTIALS_UNSAFE]: The skill is designed to access a sensitive API token stored in the user's home directory at ~/.config/grizzly/token. Accessing such files is a sensitive operation involving credentials.
  • [COMMAND_EXECUTION]: The skill relies on executing shell commands with arguments derived from potentially untrusted inputs, such as note IDs and note content.
  • [PROMPT_INJECTION]: The instructions include defensive prompts designed to restrict agent behavior under specific conditions, such as requests for bulk data enumeration or token exposure.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from Bear notes (via grizzly open-note in SKILL.md) without boundary markers or sanitization, which could influence the agent's behavior through instructions embedded in the notes themselves.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 21, 2026, 06:30 PM
Security Audit — agent-trust-hub — bear-notes-hardened