pack-shared

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions define an architecture where agents resolve execution context and specifications from external, potentially untrusted sources such as GitHub tickets, Pull Request bodies, and comments.
  • Ingestion points: execution-context.md (tickets, PRs, and comments) and review-contract.md (specifications derived from user requests and external ticket data).
  • Boundary markers: The contracts do not explicitly define the use of delimiters or instructions to ignore embedded prompts when processing data from these external sources.
  • Capability inventory: Agents following these contracts are granted capabilities to create or update Pull Requests (pr-ship.md), spawn subagents for tasks (subagents.md), and interact with a browser for visual evidence (browser-evidence.md).
  • Sanitization: No explicit sanitization, escaping, or strict schema validation for the data ingested from external ticketing systems is specified in the contracts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 08:19 PM
Security Audit — agent-trust-hub — pack-shared