Aligning Agent Workflows with the Grill Me Skill: Design, Mechanism, and Real‑World Demo

Core Mechanism

Grill Me skill aligns agent actions before execution by outputting a batch plan with dependency order, waiting for user confirmation, then proceeding. It merges independent decisions into a single batch, reducing interrogation rounds from dozens to under ten. Each decision includes a recommended option plus alternatives, enabling quick confirmation. At the end it outputs a complete decision table for a single final sign‑off. A default‑value shortcut (e.g., “use recommendation”) allows immediate exit from further questioning. The implementation consists of a single SKILL.md file, making it lightweight and easy to adapt.

Installation

One‑line installation via npx:

npx skills add flyeric0212/agent-skills

-g

installs globally to ~/<agent>/skills/ instead of the current project.

Manual installation (git clone and symlink):

git clone https://github.com/flyeric0212/agent-skills.git
mkdir -p .agents/skills
ln -s /path/to/agent-skills/skills/grill-me .agents/skills/grill-me
# Claude Code
mkdir -p .claude/skills
ln -s /path/to/agent-skills/skills/grill-me .claude/skills/grill-me

Effect Demonstration

Verified on multiple mainstream agents; example run on Codex CLI.

01 Summarize Rounds

02 Confirm Each Round

First round:

Second round:

Subsequent rounds follow the same pattern: the agent scans the current codebase, presents at least three options with a recommendation, waits for confirmation via the Answer tool, then proceeds.

03 Decision Summary

The run used Plan mode; the author recommends Build mode to obtain a full decision panorama table instead of the truncated Proposed Plan output. The two modes produce similar results, but Proposed Plan is longer and harder to capture fully.

Conclusion

The Grill Me skill enforces pre‑execution alignment by structuring batch questioning, providing recommendation‑plus‑option prompts, and allowing a quick default‑value exit. It is open‑source at https://github.com/flyeric0212/agent-skills and can be installed individually or via npx skills into mainstream agents.