Agent测试
- 作者仓库星标 291
- 作者更新于 实时读取
- 作者仓库 intent
- 领域
- 通用
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @TanStack · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- macOS · Linux · Windows
- 底层运行要求
- Node.js
- 文件与系统权限
-
- 只读
- 允许写入 / 修改
- 网络行为
- 允许外网请求
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: skill-staleness-check
description: > You are a coding agent. Your job is to evaluate whether intent skills are stale after upstream…
category: 通用
runtime: Node.js
---
# skill-staleness-check 输出预览
## PART A: 任务判断
- 适用问题:通用任务拆解、检查和交付。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“Inputs / Step 1 — Match changed files to skills / Using sync-skills.mjs”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于通用任务拆解、检查和交付,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Inputs / Step 1 — Match changed files to skills / Using sync-skills.mjs”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、写入/修改文件、会按任务需要访问外部网络、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、写入/修改文件;会按任务需要访问外部网络;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文没有稳定的斜杠命令要求。安装验证后通常全局生效,直接在对话里点名这个 Skill 并描述任务即可。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、写入/修改文件。
先用一个小任务确认它会围绕“Inputs / Step 1 — Match changed files to skills / Using sync-skills.mjs”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: skill-staleness-check
description: > You are a coding agent. Your job is to evaluate whether intent skills are stale after upstream…
category: 通用
source: TanStack/intent
---
# skill-staleness-check
## 什么时候使用
- 把通用方向的常用动作沉淀成 Agent 可调用的技能 适合处理通用任务拆解、检查、交付和复盘,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答。 把任务拆成可执行、可检查、可继续迭代的步骤;通常不需要额外…
- 面向通用任务拆解、检查和交付,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「Inputs / Step 1 — Match changed files to skills / Using sync-skills.mjs」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件;会按任务需要访问外部网络;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "skill-staleness-check" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Inputs / Step 1 — Match changed files to skills / Using sync-skills.mjs
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> Node.js | 读取文件、写入/修改文件 | 会按任务需要访问外部网络
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} Skill Staleness Check
You are a coding agent. Your job is to evaluate whether intent skills are stale after upstream source changes, and if so, update them and open PRs. You act autonomously end-to-end. PRs contain already-updated skill content, not suggestions.
If nothing needs updating, exit silently. No PR, no notification.
Inputs
Webhook payload from an upstream package repo merge to main:
{
"package": "@tanstack/query",
"sha": "abc123",
"changed_files": ["docs/framework/react/guides/queries.md", "src/query.ts"]
}
Step 1 — Match changed files to skills
Read all SKILL.md files under packages/intent/skills/. For each skill,
extract sources from the frontmatter.
Match changed_files from the webhook against sources entries across all
skills. Source references use the format Owner/repo:relative-path and
support glob patterns.
A skill is a candidate if any of its sources entries match a changed
file.
If no skills match, exit silently.
Using sync-skills.mjs
The repo includes scripts/sync-skills.mjs for programmatic staleness
detection. For a given library:
node scripts/sync-skills.mjs <library>
This checks:
- Source file SHA drift (compares stored SHAs in
sync-state.jsonagainst current remote SHAs via GitHub API) - Library version drift (frontmatter
library_versionvs current published version) - Tree-generator changes (whether the meta skill has been updated since last sync)
Use --report to write a structured staleness_report.yaml:
node scripts/sync-skills.mjs <library> --report
The report classifies skills as needing regeneration (source changed) or version bump only.
Step 2 — Evaluate each candidate
For each matched skill:
- Read the current SKILL.md content
- Fetch the file diff from the triggering commit in the source repo
- Classify the change:
| Classification | Criteria | Action |
|---|---|---|
| No impact | Diff is typo fix, comment change, test-only, or internal refactor with no API/behavior change | Skip — no update needed |
| Version bump only | Diff changes version numbers, dependency ranges, or metadata but no documented behavior | Bump library_version in frontmatter |
| Content update | Diff changes API shape, behavior, defaults, types, or patterns that the skill documents | Rewrite affected sections |
| Breaking change | Diff removes, renames, or fundamentally changes an API the skill documents | Rewrite + add old pattern as Common Mistake |
Two-pass classification
Pass 1 — Quick scan: Read the diff summary (files changed, insertions, deletions). Identify which skill sections could be affected.
Pass 2 — Detail evaluation: For each potentially affected section, read the full diff hunks and compare against the skill content. Determine if the change actually affects what the skill documents.
This prevents over-updating. A 200-line diff to a source file may only affect one line of one skill, or none at all.
Step 3 — Update stale skills
For skills classified as needing content updates:
- Load the skill-generate meta skill
- Provide it with:
- The existing SKILL.md content
- The source diff
- The current source documentation (fetch the updated file)
- Use regeneration mode (surgical update, not full rewrite)
- Validate the updated skill against all checks
For version bump only:
node scripts/sync-skills.mjs <library> --bump-version <new-version>
This updates library_version in all frontmatter for the library and
records the new version in sync-state.json.
Step 4 — Check cross-skill references
After updating skills in Step 3, check for cross-skill staleness:
- For each skill that was updated, read its
name - Scan all other skills for
requiresentries orsourcesthat reference the updated skill - For each skill that references an updated skill, evaluate whether the update makes the referencing skill stale or inconsistent
- If stale → update using the same process as Step 3
- If not → skip
This cascade is bounded to one level. Skills that reference a second-order dependency are not automatically re-checked.
Step 5 — Mark skills as synced
After updating, mark the affected skills as synced so future staleness checks have a clean baseline:
# Mark specific skills
node scripts/sync-skills.mjs <library> --mark-synced <skill1> <skill2>
# Mark all skills for a library
node scripts/sync-skills.mjs <library> --mark-synced --all
This updates sync-state.json with current source file SHAs, the
tree-generator SHA, and the sync timestamp.
Step 6 — Open PRs
For each skill (or group of skills) that was updated:
- Create branch:
skill-update/<skill-name>-<short-sha> - Commit updated SKILL.md file(s)
- Open PR with structured body
PR format
Title: skill: update <skill-name> (<package>@<short-sha>)
Body:
### Triggered by
Changes to: <list of source files that matched>
### What changed in the source
<summary of the diff — 2–3 sentences max>
### What changed in the skill
<summary of skill edits — which sections were updated and why>
### Cross-skill impact
<list any downstream skills checked; note if PRs were opened for them>
### Review checklist
- [ ] Skill content is accurate
- [ ] Code examples are complete and copy-pasteable
- [ ] No other skills need corresponding updates
- [ ] Under 500 lines
Grouping PRs
- If multiple skills for the same library are affected by the same commit, group them in a single PR
- If a cross-skill update is needed (Step 4), open a separate PR for the downstream skill to keep review scopes clean
- Never mix skills from different libraries in the same PR
No-op behavior
Exit silently (no PR, no notification, no issue) when ANY of these are true:
- No changed files match any skill's
sources - All matched diffs are classified as "no impact" in Step 2
- The sync-skills.mjs report shows all skills are current
Operational notes
GitHub API usage
The sync-skills.mjs script uses the gh CLI for GitHub API access. It
requires:
ghCLI installed and authenticated- Read access to upstream TanStack package repos (query, router, db, form, table)
- Write access to the intent repo for creating branches and PRs
Rate limiting
When checking multiple libraries or many source files, the script makes
one API call per source file per skill. For large batches, the GitHub API
rate limit (5000 requests/hour for authenticated users) may apply. The
script does not currently batch or cache API responses — if this becomes
an issue, add caching at the getRemoteFileSha level.
Manual triggering
Maintainers can run staleness detection manually:
# Check a specific library
node scripts/sync-skills.mjs db
# Check and write a report
node scripts/sync-skills.mjs db --report
# After reviewing and regenerating, mark as synced
node scripts/sync-skills.mjs db --mark-synced --all
Constraints
| Rule | Detail |
|---|---|
| Silent when nothing changes | No noise — exit cleanly if no updates needed |
| Surgical updates over full rewrites | Only change sections affected by the diff |
| One cascade level | Cross-skill checks go one level deep, not recursive |
| PRs scoped to one library | Never mix libraries in a single PR |
| Version bumps are separate from content updates | A version-only bump doesn't require regeneration |
| Commit messages include co-author | Include the coding agent's co-author tag |
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核