openspec-grillme

A skill that merges two ideas:

• grilling — relentlessly interrogate every branch of a plan's design tree before building.
• openspec-agent-sdd — drive the OpenSpec spec-driven loop end-to-end from an autonomous agent, no human in the loop.

The twist: instead of grilling a human, the agent grills itself. It generates the hard design questions, answers each from the codebase / task / a recorded default, writes a decision log, then feeds that log into OpenSpec's non-interactive CLI loop to implement and archive the change — fully unattended.

Three phases, run back-to-back:

1. Self-grill — walk the design tree depth-first, resolving dependencies between decisions. Each open question is answered by (in order) exploring the codebase, deriving from the task, or a reasonable default. Every answer is logged with its source and rationale. Only a genuine blocker stops the run.
2. Freeze — transcribe the decision log into OpenSpec's four artifacts (proposal / specs / design / tasks). Behavioral decisions become requirements with #### Scenario: blocks.
3. Build — drive openspec status/instructions/validate/archive as a --json-first state machine: generate artifacts in dependency order, validate with --strict, implement tasks flipping - [ ] → - [x], and archive -y to merge deltas back into the specs.

The decision log is the bridge: phase 1 produces it, phases 2–3 consume it.

mkdir -p ~/.claude/skills
cp -r skills/openspec-grillme ~/.claude/skills/

mkdir -p ~/.codex/skills
cp -r skills/openspec-grillme ~/.codex/skills/openspec-grillme

The SKILL.md is written to be platform-neutral; it has been authored for Claude Code first.

In a new session:

self-grill then build: <fuzzy feature request>

or in Chinese:

自问自答把设计敲定再用 openspec 无人值守跑一遍：<需求>

• Node >= 20.19.0
• @fission-ai/openspec (the skill bootstraps it: npm install -g @fission-ai/openspec@latest)

• grilling / grill-me by @mattpocock
• openspec-agent-sdd by @Dominic789654
• OpenSpec by Fission-AI