/install codebase-argus
Codebase Argus
Use this portable agent playbook for evidence-first codebase review across GitHub PRs, CI failures, and downstream fork integration work. It works as an OpenClaw/ClawHub skill, Codex project instruction, Claude Code project instruction, or a plain Markdown playbook for any coding agent that can run local shell commands.
Both upstream and downstream workflows can send the same evidence package to one provider or a multi-agent tribunal.
Fast Path
From a Codebase Argus checkout:
npm install
npm run argus -- review owner/repo#123
For private repositories or higher GitHub API limits:
GITHUB_TOKEN=\x3Cread-only-token> npm run argus -- review owner/repo#123
Do not print tokens. Do not write tokens to files.
Upstream PR Review
Default review is deterministic and rule-based:
npm run argus -- review owner/repo#123 --format markdown
Use a policy file when the repository has local rules:
npm run argus -- review owner/repo#123 --policy .codebase-argus.yml
Use API providers when credentials are available in the environment:
OPENAI_API_KEY=\x3Ckey> npm run argus -- review owner/repo#123 --provider openai-api
ANTHROPIC_API_KEY=\x3Ckey> npm run argus -- review owner/repo#123 --provider anthropic-api
GEMINI_API_KEY=\x3Ckey> npm run argus -- review owner/repo#123 --provider gemini-api
Use local CLI providers only in trusted local workspaces:
npm run argus -- review owner/repo#123 --provider codex-cli
npm run argus -- review owner/repo#123 --provider claude-cli
npm run argus -- review owner/repo#123 --provider gemini-cli
For multi-agent review:
npm run argus -- review owner/repo#123 --tribunal openai-api,claude-cli,codex-cli
Treat output as review assistance. Do not approve, merge, push, or post comments automatically unless the user explicitly asks.
CI Failure Review
When the user provides a failing job log or local log file, review it through the same provider system:
npm run argus -- ci-log logs/failure.txt
npm run argus -- ci-log logs/failure.txt --provider codex-cli
npm run argus -- ci-log logs/failure.txt --tribunal codex-cli,claude-cli,gemini-cli
When the user points at a GitHub PR with failing Actions checks, fetch the job logs directly:
GITHUB_TOKEN=\x3Cread-only-token> npm run argus -- ci-github owner/repo#123
GITHUB_TOKEN=\x3Cread-only-token> npm run argus -- ci-github owner/repo#123 --provider codex-cli
Focus on the first failing command, the most likely root cause, and the smallest fix that can be verified locally.
Autofix Plan
Use autofix-plan when the user asks for suggested fixes, safe automatic repair,
or a branch plan for narrow mechanical failures:
npm run argus -- autofix-plan owner/repo#123
The plan covers gated lanes such as npm lockfile refreshes, snapshot updates, and formatter/linter fixes. Treat it as a command plan; do not execute, push, or open a PR unless the user explicitly asks.
GitHub App Webhook Review
For automatic PR review, the deployed Next.js server exposes:
POST /api/github/webhook
GET /api/github/app-manifest
Required environment:
GITHUB_WEBHOOK_SECRET=\x3Csecret>
GITHUB_APP_ID=\x3Capp-id>
GITHUB_APP_PRIVATE_KEY=\x3Cpem-or-escaped-pem>
Use GITHUB_APP_PRIVATE_KEY_BASE64 if storing multiline PEM is awkward. The
webhook verifies X-Hub-Signature-256, reviews opened, reopened,
ready_for_review, and synchronize, ignores draft PRs, posts COMMENT reviews
only, and can add inline comments when ARGUS_WEBHOOK_INLINE_COMMENTS=true.
Supported PR comment commands:
/argus help
/argus review
/argus ci
/argus autofix
/argus pause
/argus resume
/argus pause applies argus:paused; automatic review skips PRs with that label.
/argus resume removes it.
Do not approve, request changes, merge, push, or post comments outside this configured webhook path unless the user explicitly asks.
What To Look For
Prioritize findings with concrete evidence:
- failing or pending checks;
- source changes without matching tests;
- workflow edits, especially
pull_request_target; - auth, token, webhook, payment, signature, or route-handling changes;
- dependency and lockfile changes;
- large PRs that exceed policy gates;
- stacked PRs targeting non-default base branches;
- merge queue states such as blocked, behind, dirty, or unstable;
- agreement between multiple providers.
Low-confidence model-only claims need manual verification before reporting them as facts.
Downstream Fork Sync And Integration Review
For long-lived fork maintenance, use the downstream CLI first:
npm run argus -- downstream owner/upstream me/fork
npm run argus -- downstream owner/upstream me/fork --upstream-branch main --fork-branch feature/demo
For AI CLI review of merge/rebase risk:
npm run argus -- downstream owner/upstream me/fork --fork-branch feature/demo --provider codex-cli
npm run argus -- downstream owner/upstream me/fork --fork-branch feature/demo --tribunal codex-cli,claude-cli,gemini-cli
When the user explicitly asks the agent to perform the downstream integration,
use sync. It prints a dry-run plan unless --execute is present:
npm run argus -- sync owner/upstream me/fork --mode merge --fork-branch feature/demo --test "npm test"
npm run argus -- sync owner/upstream me/fork --mode rebase --fork-branch feature/demo --test "npm test" --execute --push --create-pr
Execution rules:
- run
downstreamwith a provider or tribunal first for risky branches; - prefer a sync branch such as
sync/upstream-main; - push only with explicit
--push; - open a PR only with explicit
--create-pr; - never push directly over the user's original target branch;
- report failed commands, conflicts, and test output without hiding them.
The downstream prompt must consider both integration paths:
- merge upstream into the fork, using
git merge-treeconflict evidence; - rebase the fork on upstream, using temporary worktree rebase simulation;
- patch-equivalent cleanup candidates from
git cherry -v; - semantic patch movement from
git range-diff; - backup, test, and force-with-lease gates before any push.
Use the local dashboard when a human needs to inspect the same evidence visually:
npm run dev
Open the Downstream Fork Sync, Downstream Merge/Rebase Risk, and Downstream Agent Workflow panels. The local analyzer works in .cache/repos and temporary worktrees, and must not push or force-push by itself. Actual sync execution belongs to the CLI sync command and only runs after explicit flags.
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install codebase-argus - 安装完成后,直接呼叫该 Skill 的名称或使用
/codebase-argus触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
Codebase Argus 是什么?
Portable Codebase Argus agent playbook for evidence-first multi-agent review of GitHub pull requests, CI failures, GitHub Actions logs, GitHub App webhook re... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 22 次。
如何安装 Codebase Argus?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install codebase-argus」即可一键安装,无需额外配置。
Codebase Argus 是免费的吗?
是的,Codebase Argus 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Codebase Argus 支持哪些平台?
Codebase Argus 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Codebase Argus?
由 Yu Zhang(@aaronz345)开发并维护,当前版本 v0.1.0。