instruction-ledger

Pass

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a mechanism to extract 'durable instructions' from session history based on keyword patterns, creating a surface for indirect prompt injection.
  • Ingestion points: Session history and instruction signals specified in SKILL.md.
  • Boundary markers: Absent. There are no defined delimiters or instructions to ignore embedded commands within the processed session data.
  • Capability inventory: The skill writes instruction metadata to .claude/audits/intent-ledger.jsonl and .claude/audits/intent-ledger.md, which the agent uses for project memory and future requests.
  • Sanitization: Absent. No evidence of validation or filtering of the extracted text before it is logged to the durable ledger.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 19, 2026, 01:53 PM
Security Audit — agent-trust-hub — instruction-ledger