Jules skills · tiesa.tech

Recipes for using Google Jules as a read-only reviewer and, later, more roles. Human-readable pages with raw Markdown twins for agents.

What this is

Reusable recipes for working with Google Jules as one worker among several coding agents. Every skill has a human twin (rendered page) and an AI twin (raw SKILL.md). Rule shared by all of them: Jules never merges, publishes or opens PRs on its own; people decide.

1skills published
3review lanes (correctness · quality · security)
5live boundary sessions logged

Skills

skill

Jules adversarial review

Use when an implementer agent (Codex, Claude, Grok, Gemini or a human) has pushed a commit and you want an independent, read-only second opinion from Google Jules before merge: correctness, quality or defensive security review lanes via the Jules REST API, with plan vetting, refusal handling, findings normalisation and read-only compliance checks. Not for having Jules write code or open PRs.

Open recipe →AI twin: /skills/jules-adversarial-review/SKILL.md
planned

More Jules skills

Room for future Jules skills (for example implementation and orchestration via the same adapter) side by side with this one.

Diagrams and evidence

Adversarial review: pipeline + session lifecycle

Side-by-side diagram: review lanes, scanner fallback, evaluator gate, and one Jules session's guardrails.

Boundary tests, 8 Oct 2026

What Jules accepted, refused and changed across five read-only review sessions, plus API quirks.