slice-and-spine-review

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill's architecture relies on subagents to ingest and summarize the entire codebase, which creates a significant surface for indirect prompt injection if the repository contains malicious strings or comments.
  • Ingestion points: Phase 1 (Fan-out) instructs the agent to fan out subagents to digest every file and its associated tests within the repository.
  • Boundary markers: The instructions do not specify any delimiters or explicit 'ignore embedded instructions' warnings to prevent subagents from being influenced by data content.
  • Capability inventory: The skill uses subagent management tools (Workflow/Agent), performs broad file reads and code searches, and writes findings to the local file system in both report and wiki formats.
  • Sanitization: There is no requirement or mechanism mentioned for escaping, validating, or filtering repository content before it is processed by the AI subagents.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 06:08 AM
Security Audit — agent-trust-hub — slice-and-spine-review