汉语

This repository is organized around a single-file Agent Skills manifest, skvlt.yaml, for Skills Vault.

It is intended for people who want a reviewed, portable snapshot of approved skill sources and installed skill names.

• skvlt.yaml: the source-of-truth manifest
• MANIFEST_POLICY.md: the agent-facing policy for additions, replacements, rejections, and escalations

The manifest records:

• approved upstream skill sources
• installed skill names under each source
• integrity fields such as total_sources, total_skills, and per-source count
• the current manifest scope, global

Minimal example:

total_sources: 21
total_skills: 137
scope: "global"

sources:
  "openai/skills":
    count: 27
    skills:
      - "openai-docs"
      - "slides"
      - "sentry"

Today, the manifest covers a broad set of real-world development tasks, including general engineering workflows, AI, cloud platforms, security, testing, documentation, design, deployment, and research-oriented work.

It is especially useful for:

• developers working across a wide range of modern software tasks
• teams that want a shared, reviewed skills baseline across domains
• Skills Vault users who do not want to assemble trusted sources one by one

This repository is a maintained manifest source intended to be used with Skills Vault.

Recommended workflow:

1. Fork this repository and clone it.
2. Run the restore flow through Skills Vault.

Example:

git clone https://github.com/xixu-me/skvlt.git
cd skvlt
bunx skvlt restore --all

This repository is not a standalone installer. Its role is to provide a maintained manifest for the Skills Vault toolchain.

This repository now includes a lightweight Prettier-based formatting toolchain for Markdown and YAML files.

Local commands:

bun install
bun run format:check
bun run format

GitHub Actions workflows:

• Format Check: runs formatting validation on pull requests and pushes to main
• Format Fix: manual workflow_dispatch workflow that formats the selected branch and pushes a chore: apply formatting commit only when changes are needed

Because this manifest is global rather than project-specific, changes are handled conservatively.

High-level rules include:

• prefer small, auditable changes
• prefer adding skills under existing trusted sources over introducing source churn
• reject skills tied to a specific repository, agent, or runtime unless explicitly allowed
• escalate ambiguous overlap, process-shaping changes, and source-level edits to a human reviewer

See MANIFEST_POLICY.md for the complete decision model.

In most cases, manifest changes should be requested through an agent rather than edited directly in skvlt.yaml.

Recommended flow:

1. Fork this repository and clone it.
2. Give an agent access to the cloned repository and request a manifest change.
3. Provide the candidate skill or source and the change you want.
4. Let the agent evaluate the request against MANIFEST_POLICY.md and prepare a patch if the change is justified.
5. Review the agent's rationale and diff before accepting the change.
6. Commit the result.

If the case is ambiguous, high-impact, or source-level, the agent should normally escalate to a human instead of applying the change automatically.

Released under the MIT License. See LICENSE for details.