← ClaudeAtlas

yindeelisted

Token-efficient development harness. Deterministic scripts map the repo, scope context, tier risk, verify and review, so the model never rescans it. Use for any real development task in a codebase — feature work, bug fixes, refactors, reviews, and Git/GitHub workflow (issues, branches, PRs, CI). Cooperates with existing CLAUDE.md/AGENTS.md. Trigger with /yindee.
yindeejs/skill-yindee · ★ 1 · Code & Development · score 64
Install: claude install-skill yindeejs/skill-yindee
# /yindee — adaptive development harness Scripts answer what scripts can answer. You spend tokens on judgement, not on rediscovering the repository. **This file is a router: it names the step, the command prints the policy.** `Y` = `node <this skill's directory>/scripts/yindee.mjs`. Every command takes `--repo <dir>` (default: current repo root), `--json`, and `--verbose`. **Language:** talk to the user in **Thai**. Everything internal — reasoning, code, commit messages, PR text, subagent prompts — stays **English**. ## The loop In order. Do not skip to implementing, and **do not stop at the first failed verification.** | # | Step | How | | --- | --- | --- | | 1 | Understand | Read the task. If it names an issue: `gh issue view <n>`. | | 2 | Orient | `Y map` — layout, packages, commands, CI. Cached. | | 3 | Scope | `Y context "<task>"` — the packages, rules and files this task needs. `--reference <dir>` when the task names another repo. | | 4 | Load minimum | Read **only** what step 3 named, plus the rule files it lists. | | 5 | Implement | Edit those files. Match the surrounding code. | | 6 | Classify | `Y impact` — changed files → affected packages → risk tier → verify plan. | | 7 | Verify | `Y verify` — runs that plan, prints failures only. | | 8 | Fix → re-verify | Fix the cause, re-run **7**. Repeat until green. | | 9 | Review | `Y review` — bounded diff + path-scoped checklist + failure evidence. | | 10 | Broaden | If review changed anything, or tier is `broad`/`c