objectstack-pm-dispatch

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSOBFUSCATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill automates repository state management by executing shell commands including gh label create, git log, gh pr list, git worktree add, and git push to handle task assignment and branch management.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue bodies and comments which is then interpolated into prompts for sub-agents.
  • Ingestion points: Fetches full issue bodies and all associated comments from the GitHub backlog.
  • Boundary markers: Uses explicit text delimiters like --- OPERATING PROCEDURE (binding) --- and --- END OPERATING PROCEDURE --- to isolate untrusted data.
  • Capability inventory: Possesses capabilities to create branches, push code, and manage Pull Requests across repositories.
  • Sanitization: Recommends using HTML entities (<, >) to prevent GitHub's platform sanitizer from deleting machine-readable markers.
  • [EXTERNAL_DOWNLOADS]: The developer agent template includes a command to install dependencies, which triggers the download and execution of packages from external registries based on the project being managed.
  • [OBFUSCATION]: The skill instructs agents to use HTML entities to encode certain characters in GitHub comments to prevent platform-level sanitization from modifying load-bearing markers, such as < and > in the dev-report marker.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 01:15 PM
Security Audit — agent-trust-hub — objectstack-pm-dispatch