personal-mode

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill drafts new system-level instructions based on untrusted data from past user transcripts, creating a surface for potential instruction propagation.
  • Ingestion points: Analyzes recent transcripts across user projects (SKILL.md).
  • Boundary markers: Does not specify technical delimiters for the generated content, but relies on a mandatory human-in-the-loop confirmation step.
  • Capability inventory: Writes persistent skill files to the local configuration directory (${CLAUDE_CONFIG_DIR:-$HOME/.claude}/skills/).
  • Sanitization: Explicitly instructs the agent to exclude secrets and tokens, and requires a diff review before overwriting existing profiles.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 07:21 PM