continual-learning
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Memory Persistence Infrastructure: The skill describes a system for persisting agent 'learnings' in database files (
learnings.db) located in both project-local directories and the user's home directory (~/.copilot/). While accessing the home directory is a sensitive operation, in this context it appears to be used for the legitimate purpose of maintaining a global cross-project memory for the agent within the vendor's standard application data paths. - Hook Installation Pattern: The instructions include a manual step to copy a hook directory into the project's
.github/hooks/folder. This mechanism allows the agent to monitor tool outcomes and refine its behavior. Users should verify the contents of any such hooks before installation as they become part of the local development environment's execution flow. - Indirect Prompt Injection Surface: The skill establishes a workflow where the agent reads instructions and 'learnings' from external files like
conventions.mdandlearnings.db. This creates a potential surface for indirect prompt injection; if these files were to be populated with untrusted or adversarial content, they could influence the agent's future logic and decision-making processes.
Audit Metadata