continuous-learning

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill creates a significant attack surface for Indirect Prompt Injection by synthesizing and saving new instruction sets (skills) from untrusted data sources.
  • Ingestion points: Conversation history, task context, and results from WebSearch are used to generate skill content.
  • Boundary markers: The system lacks structural delimiters to prevent embedded instructions in processed data from influencing the generated skill logic.
  • Capability inventory: The skill uses high-privilege tools like Write, Edit, and Skill to persist and load new instructions.
  • Sanitization: Relies on behavioral instructions for the AI rather than technical filtering of generated markdown/YAML.
  • [COMMAND_EXECUTION]: Installation requires the user to execute local shell commands and configure a persistent terminal hook script (continuous-learning-activator.sh). While the provided script is informational, the use of execution hooks for external scripts introduces a path for local code execution.
  • [EXTERNAL_DOWNLOADS]: The README directs users to download the skill from an external repository (github.com/blader/claude-code-continuous-learning-skill.git). This repository owner does not match the provided author context ('ckorhonen'), which is a metadata discrepancy.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 08:19 AM