act-or-decide

Pass

Audited by Gen Agent Trust Hub on Aug 3, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill contains explicit, robust instructions to treat external data (backlog tasks, emails, issues) as content only. It includes a specific prohibition against laundering injected instructions into user decisions.
  • [DATA_EXFILTRATION]: All database operations are performed through a scoped helper (db.mjs) that automatically limits queries to the current user's ID. Network operations are restricted to specific, defined tools for research and status updates.
  • [COMMAND_EXECUTION]: The skill uses a shared script (ensure-deps.sh) to manage dependencies. While this executes bash, it targets a known idempotent script within the plugin root for environment consistency.
  • [EXTERNAL_DOWNLOADS]: The skill downloads dependencies via standard package managers (npm via ensure-deps.sh) and fetches configuration from defined local workspace paths (.proactive-jupi/config.local.json). No untrusted remote execution patterns were found.
  • [REMOTE_CODE_EXECUTION]: The skill strictly separates the 'planner' (this skill) from the 'executor' (execute-action). This skill only queues actions and does not perform side-effects itself, reducing the risk of unintended code execution.
  • [CREDENTIALS_UNSAFE]: No hardcoded secrets were detected. Connection strings and user IDs are read from a local configuration file that is explicitly managed outside of the skill's own code.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 3, 2026, 12:18 PM
Security Audit — agent-trust-hub — act-or-decide