You know the “stuck” moments: a decision only one person can make, but nobody has time for a sit-down; a setup step that requires a human to click through a web console while the agent can only watch; a plan that looks airtight until you start building and realize it was full of assumptions. On August 29, Hermes shipped an official optional skill for each of the three: decision-questionnaire (turn the decision into an async questionnaire), setup-wizard-generator (generate interactive setup wizards), and grill-me v2.0.0 (upgraded to frontier-rounds plan interrogation). All three are Hermes-native ports from the MIT-licensed mattpocock/skills repo (240k★), installable with one hermes skills install command.

What optional skills are

A Hermes skill is a document with frontmatter and a procedure — it teaches the agent which steps to follow in a specific situation. Alongside the default built-in skills, the official repo ships a batch of optional skills: installed on demand, no default-context cost. August 29 brought three new ones in a single sweep.

1. decision-questionnaire: turn “waiting for a decision” into a form

When to use it: a decision is blocked on one person — you’re the implementer, but choosing A or B, or how much budget to approve, belongs to your boss, client, or partner. Instead of waiting for a “convenient chat,” Hermes drafts a structured questionnaire the other person can fill in async, or during a meeting gap.

The clever part: it interviews the sender, not the subject. Two exchanges — who’s the recipient, and what must come back — then it drafts gap-targeted questions, most important first, each with an answer stub, written to decision-questionnaire-<slug>.md.

"I need the CTO to decide on the API gateway — generate a questionnaire"
→ Hermes confirms: who is it for? what decision must come back?
→ Output: decision-questionnaire-api-gateway.md (questions ranked by importance + answer templates)

2. setup-wizard-generator: automate the human-only steps

When to use it: part of the flow can only be done by a human — grabbing an API key from a web console, provisioning CI secrets, running a one-off migration. Previously your only option was writing the steps down in a doc while the agent watched. Now Hermes can generate an interactive bash wizard that walks a person through it, stage by stage.

The generated wizard: opens each URL → tells the human exactly what to click and copy → captures values (hidden input for secrets) → writes them idempotently to .env or GitHub Secrets → shows progress at every stage. Generation itself is a four-step procedure: scope → map the journeys → author the stages → static verification.

"Generate a wizard to walk me through configuring a Grafana API key"
→ Hermes produces a setup wizard script (built on the 204-line template.sh wizard library)
→ You run it, follow the prompts, click and paste — environment variables land on disk

3. grill-me v2.0.0: interrogate the plan before writing code

When to use it: you suspect the plan has unexamined assumptions. grill-me is an “adversarial plan interview” skill: before any code is written, the agent plays a rigorous interviewer and grills your plan end to end.

The headline change in v2.0.0 is the frontier-rounds mechanic (from Matt Pocock’s grilling skill): instead of one question per turn at snail speed, each round asks the whole current frontier — every question whose prerequisites are settled — in dependency order, each with a recommendation. Facts are the agent’s job (it looks things up in the codebase and beyond); decisions stay with you. The session ends only when no branch is being silently assumed — every unexamined corner of the plan gets surfaced.

# Install (official skill identifier)
hermes skills install official/software-development/grill-me

"Run grill-me over my refactoring plan"
→ agent asks in rounds: technical decisions → edge cases → synthesis, everything eligible per round
→ the interview only ends when no implicit assumption remains

One naming note: the skill was briefly renamed “plan-interrogation,” but the follow-up fix on the same day (PR #97872) kept the name grill-me with the content upgrade intact — so what you install now is grill-me v2.0.0 with frontier-rounds. It originates from community contributor Rafael Zendron’s original design (#43694), and the credits now include both him and Matt Pocock.

Discovering and installing more optional skills

The hermes skills command family is built for exactly this:

hermes skills browse          # browse the official skill directory
hermes skills search questionnaire
hermes skills install official/productivity/decision-questionnaire
hermes skills install official/devops/setup-wizard-generator
hermes skills list            # see what's installed

That’s it — they’re active immediately, and the agent picks the right procedure automatically when the situation comes up.

Caveats

• All three landed August 29, currently on main only — not in any release tag (v0.20.6 was tagged August 27); run hermes update before installing;
• They’re ports from a third-party repo, but each passed the official skill authoring-standards test suite, including a de-upstreaming residue check;
• setup-wizard-generator targets human-in-the-loop steps — if a real API exists for the task, the normal tools are still the better path.

Summary

One skill handles “waiting for a decision,” one handles “clicking through a console,” one handles “finding the holes before you build” — three of the most common human-in-the-loop blockers in an agent workflow, now covered by official optional skills. Installation is one command, and they show up automatically when the situation calls for them. To dig into the skill system itself, see our project-local skills guide and skill combinations post; the full hermes skills command reference lives on the command docs page.