update-agent-learnings

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes session data to extract learnings, creating a surface for indirect prompt injection into persistent configuration files.\n
  • Ingestion points: Current session conversation history.\n
  • Boundary markers: The skill includes a mandatory user confirmation gate via the AskUserQuestion tool before any file updates are executed.\n
  • Capability inventory: Read and write access to agent definition files in .claude/agents/*.md and the central learnings file .claude/learnings/agent-learnings.md.\n
  • Sanitization: Relies on human-in-the-loop verification to validate extracted learnings before they are persisted.\n- [COMMAND_EXECUTION]: The skill performs internal state verification using standard shell commands.\n
  • Evidence: The skill uses grep -l "^## Learnings" .claude/agents/*.md | wc -l to verify the state of agent definition files within the local project directory.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 01:30 PM