Skip to content
Back to search

Auto-Optimize

An Agent Skill for vibe coding and AI workflow development。Built for when you hit these walls:Vague requirements → no baseline, endless rework、Only one prompt phrasing → can't tell if the problem is you or the model、Model says "looks good" → but features are missing 、great output last time, no idea how to reproduce it.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/gilbertwuu--auto-optimize/badge.svg)](/listings/gilbertwuu--auto-optimize)

HTML

<a href="/listings/gilbertwuu--auto-optimize"><img src="/api/listings/gilbertwuu--auto-optimize/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

1. Keep this repo at your project root (or merge `SKILL.md`, `program.md`, `docs/`, `scripts/` into an existing repo). 2. Optional: add host-specific rules (e.g. `.cursor/rules/`, OpenClaw config) to enforce the workflow. 3. Ask for features in your host; the agent should follow the phases when this Skill is loaded. 4. Verify from **repository root**:

FINAL_TRACK_FILE=tracks/prompt-a/r05.html DELIVERY_FILE=index.html bash scripts/skill-verify.sh

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 61 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: auto-optimize
  3. description: >-
  4. Host-agnostic Agent Skill: double-layer workflow for AI-assisted delivery—requirement
  5. specification, multi-prompt outer loop, on-disk inner iterations, phase gates,
  6. optional byte-level verify. For any IDE, CLI, or Skill-capable tool (e.g. OpenClaw).
  7. Read program.md and docs/verifiability.md before editing delivery artifacts unless waived.
  8. ---
  9. # Double-layer auto-optimization (Skill entry)
  10. This file is the **Skill package entry** (`SKILL.md`). Normative text lives under `docs/`. **Not tied to a specific IDE**—any host that loads this Skill can use it.
  11. ## Strict execution (hard constraints)
  12. - The agent **must** follow **`program.md`**, **`docs/verifiability.md`**, **`docs/phase-01-spec.md`** through **`phase-04-output.md`**, and **`docs/execution.md`** in full; **must not** shorten the workflow to save tokens, time, or because it “feels enough”.
  13. - **Each** `tracks/prompt-*/` inner loop **must** satisfy **`docs/phase-03-inner.md`**: **gates** (e.g. at least **r01**, **r02** per scheme) and **main termination** (**all items checked** in **`tracks/phase-01-acceptance.md`**; or **5** rounds max). Self-score “≥8” **must not** alone justify stopping (**`verifiability.md` R3**). **Forbidden**: claiming a track is “deliverable” and moving to phase 4 before the acceptance checklist passes.
  14. - **Forbidden**: using “reduce workload” or other **reasons not in the Skill** to deviate; **only** when the user **explicitly waives in the current turn** may the full workflow be skipped.
  15. ## Quick start (agent)
  16. 1. Read **[program.md](program.md)** (shortest path) or the checklist below; **do not skip** **[docs/verifiability.md](docs/verifiability.md)** (R1–R5).
  17. 2. Open phases in order: **`docs/phase-01-spec.md`** → **`phase-02-prompts.md`** → **`phase-03-inner.md`** → **`phase-04-output.md`**; read **[docs/execution.md](docs/execution.md)** for cost/order reminders.
  18. 3. **Before** phase 1 gates pass, do not rewrite delivery code; phase 1 **must** write **`tracks/phase-01-acceptance.md`** (**R3**); phase 2 **must** write **`tracks/phase-02-consistency-check.md`** (check + scheme–dimension table); **`r01` for `prompt-b`/`prompt-c` must not** be a full-page copy of `prompt-a`; waivers **must** be explicit in the **current** user message.
  19. 4. Before claiming delivery matches the final track file, in a class-A environment actually run **`scripts/skill-verify.sh`** from the **repository root** (see [program.md](program.md)).
  20. ## Execution checklist
  21. ```text
  22. - [ ] Read docs/verifiability.md (R1–R5)
  23. - [ ] Phase 1: spec written, feature list non-empty; tracks/phase-01-acceptance.md exists with ≥1 item
  24. - [ ] Phase 2: outer prompts + consistency check; tracks/phase-02-consistency-check.md written (with table)
  25. - [ ] Phase 3: each tracks/prompt-*/ has at least r01.* and r02.*; non-a r01 is independently generated
  26. - [ ] Phase 4: compare & output; at least one track fully checked before final delivery (R3; self-score reference only)
  27. - [ ] Delivery claim: ran skill-verify.sh and recorded command + output (R2/R4)
  28. ```
  29. ## Resource index
  30. | Resource | Path |
  31. |----------|------|
  32. | Shortest entry | [program.md](program.md) |
  33. | R1–R5 authority | [docs/verifiability.md](docs/verifiability.md) |
  34. | Phases 1–4 | [docs/phase-01-spec.md](docs/phase-01-spec.md) etc. |
  35. | Execution | [docs/execution.md](docs/execution.md) |
  36. | Layout & verify | [reference.md](reference.md) |
  37. | Example | [examples.md](examples.md) |
  38. | Human guide | [SKILL-GUIDE.md](SKILL-GUIDE.md) |
  39. | Script | `scripts/skill-verify.sh` (run from repo root) |
  40. ## User waiver
  41. Must be **written in the current task message**, e.g.:
  42. `This task waives the Skill / skips phases`
  43. (Synonyms allowed if equally explicit; the agent **must not** infer a waiver.)
  44. ## High-risk domains
  45. Does not replace regulation, clinical procedures, or human review; model self-scores and scripts are not sole safety evidence.
README.md · README · 41 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Auto-optimize — Agent Skill
  2. **Double-layer auto-optimization Agent Skill** — **host-agnostic**: use with any IDE, CLI, or Skill-capable tool (OpenClaw, editor integrations, etc.), not tied to a single vendor.
  3. ## One-liner
  4. An **Agent Skill** workflow (entry: root `SKILL.md` + `docs/`): lock requirements first, iterate **multiple Prompt phrasings** in parallel, write **each round to disk** under `tracks/prompt-*/`, use an **acceptance checklist** (not self-score alone) as the delivery bar, and optionally **hash-compare** delivery vs the chosen final track file.
  5. ## Why use it?
  6. - Vague requirements → churn without a baseline
  7. - Only one phrasing → no A/B on “how to ask”
  8. - “Five iterations” with **no files** → not auditable
  9. - Self-score 8/10 while features missing
  10. - **No proof** delivery matches the “final” track file
  11. This Skill addresses that with **four phases** and **five verifiability rules (R1–R5)**.
  12. ## Quick start
  13. 1. Keep this repo at your project root (or merge `SKILL.md`, `program.md`, `docs/`, `scripts/` into an existing repo).
  14. 2. Optional: add host-specific rules (e.g. `.cursor/rules/`, OpenClaw config) to enforce the workflow.
  15. 3. Ask for features in your host; the agent should follow the phases when this Skill is loaded.
  16. 4. Verify from **repository root**:
  17. ```bash
  18. FINAL_TRACK_FILE=tracks/prompt-a/r05.html DELIVERY_FILE=index.html bash scripts/skill-verify.sh
  19. ```
  20. ## Layout
  21. | Path | Role |
  22. |------|------|
  23. | `SKILL.md` | Skill entry (YAML `description` for discovery) |
  24. | `program.md` | Shortest path for agents |
  25. | `docs/` | R1–R5 and phase 1–4 norms |
  26. | `scripts/skill-verify.sh` | Optional checks (run from repo root) |
  27. ## License
  28. MIT — see [LICENSE](LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.