ow-vault
Maintain the private Orchestra workspace and its memory. Jump one note. Lasting files only. Record real outcomes, never synthetic ones. No secrets.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- ow-vault
- description
- Maintain the private Orchestra workspace and its memory. Jump one note. Lasting files only. Record real outcomes, never synthetic ones. No secrets.
03 SKILL.md 正文
Orchestra vault
BRAIN = MEMORY. The workspace is where Orchestra remembers; it is not a second control plane. Claude-Mem is not a second brain.
Root: the private vault on the operator's machine. It is not a path in this method repo. The vault MCP can read it. Do not paste that path into a public file.
Read
Do not open the whole workspace. routes.md → one file → the app repo. User preferences and stack stay in that vault. Catalog files only when asked to catalog.
Layout
| Path | What |
|---|---|
projects/<slug>/ |
One product. Starts with idea.md. Add design, spec, progress, review, ship-post only when they exist. |
templates/ |
Copy, then fill |
memory/decisions.md |
Dated architecture and product decisions |
memory/resource-memory.json |
Real resource outcomes only |
Preferences.md |
Taste, stack, how the human thinks |
kit/ |
Host adapters and sync scripts |
What memory is for
Three layers:
- Global — taste and stack in
Preferences.md; dated yes/no inmemory/decisions.md. - Project — one product's context in
projects/<slug>/idea.md. - Session — distilled at the end. Discarded, not filed.
Record after a real job: which resource combination improved the result, what failed, what the human liked or hated, and why a route was chosen.
A durable decision uses templates/memory/decision.md: date, project, provenance, confidence, and whether it is global taste or one product. Technology choices, visual preferences, failed approaches, and performance lessons go there only after they happened. Skill effectiveness and tool reliability belong in memory/resource-memory.json when a real run wrote them. A catalog name is not a result. Do not copy a private vault path or a private note into the public method repo.
Never write synthetic evaluations. Generated or duplicated rows in resource-memory.json are not learning — they are noise that will mislead future routing. Only an executed job writes a row.
Do not file chat transcripts. Do not restate facts the repository already shows.
Write hygiene
- Will this matter in a month? If not, say it in chat and do not file it.
- Prefer editing the existing note over creating a new one.
- One idea per note, self-contained, headed so a later reader can split it.
- A map points; it does not duplicate.
- Delete junk the same turn you notice it.
- Secrets never here. Never commit MCP configs,
.env, or tokens.
Learning
After a like, a hate, a finished task, or a genuinely useful find, append to Preferences.md in the same turn. Taste is allowed to change; prefer editing the existing bullet. Do not create a second taste file.
When a session ends, append one dated line to projects/<slug>/lessons.md. One line. What happened, and what to do next time. Do not paste the chat.
A new skill is a proposal, not an install. After the same lesson appears three times, and the human says yes, write skills/_proposed/<name>/SKILL.md from templates/custom-skill.md. Stop there. Do not copy it into skills/ and do not sync it. One job. A narrow description. A way to verify.
If the human rewrites a draft and asks to keep the change, append one line to Preferences.md naming the concrete edit. Do not rewrite old taste. Do not train a model.
Git
The workspace may have a private remote. This public template is a different repo. Do not flip a populated workspace public — its history contains personal files.
Preconditions
- The task names the files or the acceptance line.
Procedure
- The task's paths and acceptance line win.
- Open one note. Append a lesson only after the human says yes.
Output
The change, the command that checked it, and what that command printed.
Verify
Say the work is done only after that command has been run.
Skip
Skip when the human says skip orchestra. Do not add a second conductor.
04 同仓库更多 Skills
writing-skills
Use when creating new skills, editing existing skills, or verifying skills work before deployment
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
web-design-guidelines
Review UI code for Web Interface Guidelines compliance. Use when asked to "review my UI", "check accessibility", "audit design", "review UX", or "check my site against best practices". Do not load on Plan-only or backend-only jobs.
verification-before-completion
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
using-git-worktrees
Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback
unslop
Turns vibecoded, AI-looking UI into a designed interface with its own character. Finds a direction from the product itself, then fixes spacing, color, icons, copy, type, layout, details, visuals and motion in that order, and can generate a real hero visual with an image tool when one is available. Use when the user asks to unslop, polish, clean up or redesign a screen so it stops looking AI-generated, or when a UI shows the usual tells like purple gradients, glows, gradient text, emoji labels, "New ✨" badges, glossy blobs, uppercase tracked labels, pulsing elements or uneven spacing. Works on React, Tailwind, HTML and CSS, for dashboards, apps and landing pages.