SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsauto-updater

auto-updater

Agents & models

View repository

Updated 10-15 20:29

About this skill

/auto-updater
  1. Load ~/.claude/plugins/config/claude-for-legal-zh/legal-builder-hub/CLAUDE.md → installed skills + auto-update preference.
  2. Use the following workflow.
  3. Check each installed skill's source for a newer version.
  4. Per preference: apply / notify / show diff.

Purpose

Community skills improve. This skill notices when improvements happen, shows what changed, and applies updates only after explicit approval.

Trust posture

Installed skills are code running in your privileged legal environment. Upstream repositories can be compromised, transferred to new owners, or have their behavior changed in ways you don't want. This skill is designed to ensure thatno update is ever applied without you reading the diff and approving it. This is not a preference—it is by design.

Load context

~/.claude/plugins/config/claude-for-legal-zh/legal-builder-hub/CLAUDE.md → installed skills (with version/commit SHA), update preference (notify / manual).

Workflow

Step 1: Check each installed skill

For each skill in the installed list:

  • Fetch the current commit SHA from the source registry (exact commit, not a tag or branch head—tags are mutable and can be retroactively rewritten by the publisher; only commit SHAs are immutable)
  • Compare with the SHA pinned at install time
  • If different: an update is available

Step 2: Diff and trust review

For each update, show the full diff:

# [技能名称] — [已安装 SHA] → [最新 SHA]

## SKILL.md 变更
[统一差异]

## hooks/hooks.json 变更
[统一差异 — 标记：hooks 可执行任意代码]

## .mcp.json 变更
[统一差异 — 标记：MCP 服务器以你的凭据运行]

## 其他文件
[附差异的新增/删除/修改文件列表]

Then run trust checks:

  • hooks/hooks.json changed? Hooks can execute arbitrary shell commands. Prominently show the diff and ask the user to confirm they understand what the new hooks do.
  • .mcp.json changed? New or changed MCP servers can access your environment. Handle the same way.
  • allowed-tools or tools frontmatter expanded? New tool access is a privilege escalation.
  • Are there new network calls, file writes outside the skill directory, or command execution in SKILL.md? Flag them.
  • Has the skill's description or stated purpose changed? A skill that claimed to "review NDAs" and now claims to "send contracts" has changed its purpose.

Step 2.5: Rescan the new version (GlassWorm gate)

Before applying an update, re-run the fullnewversion scan on the skills-qa . A skill that was clean at v1.0 may carry harmful content at v1.1 (GlassWorm pattern: trusted publisher, established skill, minor version bump carrying harmful code). Trust at install time does not transfer to updates.

Rules:

  1. Do not update when a regression is found. If the new version produces findings the old version did not—in any skills-qa Step 1.5 category—deny the update by default and explain why.
  2. Security-surface diffs require human approval regardless of the assessment result. Any change involving hooks/hooks.json、.mcp.json、allowed-tools/tools frontmatter, new Bash/WebFetch/WebSearch access, new external URLs, new file write paths outside the skill directory, or description frontmatter differences always require a mandatory human approval prompt and cannot be bypassed by a clean LLM scan.
  3. Read-only scan context. The scan reads attacker-controlled text (the new SKILL.md). Run it in a read-only subagent where available. The installing agent receives the subagent's report; it only gains write permission after human approval.
  4. Reject updates that now fail the scan. If the new version hits REFUSE tier patterns, do not offer a "still apply" option.

Step 2.6: Freshness-triggered re-verification

Check not only new submissions. Also check whether installed skills have passed their freshness window.

For each installed skill, read the verified last_verified、freshness_window and freshness_category markers from the install log.

If the active window has passed and there are no newer submissions: offer options: (a) check the source yourself, (b) flag it to the registry maintainer, (c) disable the skill until re-verification.

If the active window has passed and there are newer submissions: always re-verify, do not apply silently.

Step 3: Handle by preference

Notify (default): show the full diff and trust check. "An update is available. Review the diff above. Apply? [y/n]"

Manual: list only which ones have available updates. Run when the user is ready /legal-builder-hub:auto-updater --apply [技能名称]。

There is no "automatic" mode. Updates to code running in your legal environment always require a human to read the diff.

Step 4: Apply (after explicit approval)

Replace the installed skill file with the new version. Update the commit SHA in the installed list. Back up the old version first for rollback.

Rollback

If an update breaks some functionality:/legal-builder-hub:auto-updater --rollback [技能名称] restore from the backup.

What this skill does not do

  • Automatically apply updates. Never. Every update has a diff and approval.
  • Update skills not installed through the center (manually placed skills are managed by the user themselves).
  • Trust tags, branches or version numbers. Pin only commit SHAs, because only commit SHAs are immutable.