@blundergoat/goat-flow
Version:
AI coding agent harness and local dashboard for Claude Code, OpenAI Codex, Google Antigravity, and GitHub Copilot - setup audits, guardrails, structured skills, deny hooks, and persistent learning loops.
33 lines (22 loc) • 1.41 kB
Markdown
goat-flow-reference-version: "1.12.1"
# Skill Quality Testing
Short index for full-depth skill-authoring work. Load only the topical file(s)
needed for the current phase; do not pre-load the whole pack unless the task
genuinely spans TDD iteration, review-class hardening, and deployment.
## Availability Check
Non-runnable authoring methodology - no CLI check applies. Load when creating or hardening a goat-flow skill, then open the topical file named in the table below.
## Which file to load
| File | Content | Load when |
|------|---------|-----------|
| `tdd-iteration.md` | Iron law, RED/GREEN/REFACTOR loop, pressure scenarios, rationalisations, bulletproofing | Creating or hardening any skill. Load first. |
| `adversarial-framing.md` | Cynical-reviewer role, zero-findings HALT, parallel reviewer pattern, finding schema | Authoring or hardening review-class skills. |
| `deployment.md` | Deployment checklist, verification claim evidence, consumer/API skill guardrails, STOP rule | Finalising before merge. |
## The iron law (always-loaded anchor)
> **No skill without a failing test first.**
This applies to NEW skills AND to EDITS of existing skills. See
`tdd-iteration.md` for the full methodology.
## Cross-references
- `.goat-flow/skill-docs/skill-preamble.md` - Proof Gate and evidence standard
- `.goat-flow/skill-docs/skill-conventions.md` - conventions and task tracking