learning-log

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to use local shell commands (ls, grep, mkdir, cat) to manage a learning-log/ directory. These operations are essential for the skill's intended file-management functionality and are restricted to a specific local path.
  • [PROMPT_INJECTION]: The skill ingests untrusted conversation data to generate log entries. This creates a surface for indirect prompt injection, but the agent's instructions to transform and distill the content into a fixed Markdown schema limit the risk.
  • Ingestion points: Analyzes active conversation history (SKILL.md, Step 1).
  • Boundary markers: None specified in the workflow.
  • Capability inventory: Includes directory creation, file searching, and file writing to the local file system.
  • Sanitization: Instructions to 'distill key concepts' and 'cut conversational filler' serve as semantic filtering of the input.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 10:21 AM