context-restore

Pass

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Executes several specialized binaries located in ~/.claude/skills/gstack/bin/ (including gstack-skill-start, gstack-slug, gstack-paths, and gstack-skill-end) for state management and telemetry tasks.
  • [DATA_EXFILTRATION]: Transmits telemetry and operational data (such as session status, outcome, and insights) to the vendor's platform using the gstack-skill-end and gstack-learnings-log utilities.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes session transcripts (.jsonl) and checkpoint files (.md), which are potential sources of indirect prompt injection.
  • Ingestion points: Reads files from ~/.claude/projects/ and project-specific checkpoints directories.
  • Boundary markers: Contains explicit instructions to treat transcript contents as data rather than instructions and to flag any text attempting to command the agent.
  • Capability inventory: Employs shell execution via Bash and user interaction through AskUserQuestion.
  • Sanitization: References the use of gstack-redact to ensure data is cleaned before being quoted or used externally.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 16, 2026, 08:17 PM