skills/garrytan/gstack/spec/Gen Agent Trust Hub

spec

Pass

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes multiple local binaries from the gstack vendor suite (e.g., gstack-config, gstack-redact, gstack-telemetry-log) to manage configuration, state, and safety checks.
  • [DATA_EXFILTRATION]: Transmits anonymized usage telemetry (skill name, duration, and outcome) to an external endpoint. This behavior is subject to a user-configurable opt-in setting prompted during first use.
  • [REMOTE_CODE_EXECUTION]: Utilizes an external AI service (codex exec) to perform an automated quality review and executability score of the generated specification draft.
  • [COMMAND_EXECUTION]: Automates the creation of a new git worktree and the spawning of a secondary Claude Code agent (claude -p) to implement the specification.
  • [PROMPT_INJECTION]: Acts as an indirect prompt injection surface by ingesting untrusted user data to generate a specification that is subsequently processed by other AI models and agents. The skill implements mitigation strategies including a multi-tier redaction scan (gstack-redact), a semantic content review phase, and explicit instruction boundaries for external calls.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 26, 2026, 10:43 AM
Security Audit — agent-trust-hub — spec