GitHub Release Analyzer

以 prepare→summarize→render→deliver→commit 固定流水线分析 GitHub 仓库的 release：prepare 确定待分析版本，summarize 按 summary-contract 生成汇总，render 输出最终消息。manual 模式只看最新正式版且不读写状态，cron 模式做增量跟踪并在交付成功后提交状态。

stoner-byte

热度 53.87

最近更新 2026-08-19 12:39:41

分类：数据依赖：代码执行 / 第三方 API / 读写文件 / 内部调用大模型应用场景：摘要总结 / 报告撰写 / 自动化脚本 / DevOps ...

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add github-release-analyzer

Metadata

name: github-release-analyzer
version: 0.0.5
description: Analyze GitHub repository releases and summarize release notes. Use when the user asks to analyze, summarize, review, or track the latest release or recent releases for a GitHub repo, including phrases like "分析最新 release", "latest release", "release 更新了什么", "汇总仓库 release", or requests to build a cron that tracks GitHub releases. Supports manual one-off analysis and cron-driven incremental tracking.
metadata: {"openclaw":{"emoji":"📦","requires":{"bins":["python3"]}},"hermes":{"tags":["github","releases","monitoring","cron"]}}

GitHub Release Analyzer

Use this skill as a fixed pipeline:

prepare -> summarize -> render -> deliver -> commit

Only summarize is free-form. prepare, render, and commit are deterministic script steps.

Inputs

Repo input: https://github.com/<owner>/<repo> or <owner>/<repo>
Default stateKey: {owner}__{repo}
Default template: default
Default behavior:
- manual: latest formal release only, no state read/write
- cron: formal releases only, first run selects latest only, later runs select unprocessed only

Workflow

Before running any scripts/... command, set the command working directory to the directory containing the loaded SKILL.md (the Skill root). Do not run these commands from the OpenClaw workspace root. This does not require a shell SKILL_DIR environment variable: OpenClaw uses the loaded Skill's installed directory, and Hermes uses the skill_dir returned by skill_view.

Read references/execution-modes.md and choose manual or cron.
From the Skill root, run:

python3 scripts/run.py prepare --repo <repo> --mode <manual|cron>

Useful flags:

--state-key <key>
--limit <n>
--include-prerelease
--output-template default
--initial-cron-behavior latest-only

If status=has_updates, read references/summary-contract.md and produce one summary string per selected release. Keep the summary language aligned with the primary language of the invocation instruction.
Render the final message:

python3 scripts/run.py render --repo <repo> --mode <manual|cron> < payload.json

When building the render payload, preserve the release fields required by render, especially tag_name, published_at, and html_url.

In cron mode, after successful delivery, commit state:

python3 scripts/run.py commit \
  --repo <repo> \
  --mode cron \
  --processed-tags <comma-separated-tags> \
  --latest-release-id <id> \
  --latest-published-at <iso8601>

Rules

Treat prepare output as the source of truth for selection, ordering, first-run behavior, repo normalization, and state path.
The output language must follow the primary language of the invocation instruction.
The final outgoing message must come from render.
In cron mode:
- if status=no_update or status=no_release, reply with NO_REPLY
- if status=has_updates, emit only the final rendered message
Update processed state only after successful delivery.

Hard prohibitions

Do not skip prepare.
Do not bypass render.
Do not reorder or silently drop releases selected by prepare.
Do not call commit in manual mode.
Do not call commit before delivery succeeds in cron mode.
Do not replace the summary contract with ad hoc prose outside the required summaries[] string-array shape.
In cron mode, when status=no_update or status=no_release, reply exactly NO_REPLY.
In cron mode, do not emit progress chatter, intermediate JSON, step labels, or workflow narration.

Files to read

references/execution-modes.md
references/summary-contract.md
references/templates.md
references/state-format.md