Waggle 开源：用 Claude 智能体对代码仓库做只读审计并生成修复计划

Formerly accel-scope. Like a scout bee's waggle dance, it goes out into your code and comes back with what matters.

Point an AI auditor at your repositories (and, optionally, your warehouse) and get an evidence-backed report plus a fix plan your coding agent can run.

Waggle runs on your machine. It clones the code you select read-only, lets a team of Claude agents investigate it, verifies every claim against the code, and writes two reports: a short Leadership brief (what matters and what to decide) and an Execution report (every finding with evidence, the fix, how to verify it, and a REMEDIATION.md you can hand to a coding agent).

What it does

Quick Ask — ask one question about your code or data ("Where do we compute monthly revenue, and do the two dashboards agree?"). One agent investigates and answers with cited evidence. Typically $0.30–1.
Full Scan — a whole-repo audit. A comprehension agent decides which expert lenses apply (security & supply chain, architecture & code health, release engineering, API stability, business logic, data trust, mobile, multi-tenancy, …); each lens poses falsifiable hypotheses and measures them against the code and connected data. Claims are re-checked by an adversarial audit before they reach the report. Typically $5–30.
Incremental re-scan — scanning the same code again reuses everything the changes did not touch and reports what is new, fixed, or still open since the last scan.
Memory — durable notes about your projects (metric definitions, known traps, conventions) that later runs recall as prior context. Stored locally; editable and revertable.
Read-only, always — agents can read and query; they cannot write to your repos or data sources.

Quick start

Requirements: Node.js 22.7+ and git.

git clone https://github.com/ai-theresa-lab/waggle.git && cd waggle
npm ci
ANTHROPIC_API_KEY=sk-ant-... npm start

Open http://localhost:4317. (You can also leave the key out and paste it in Settings → API keys.)

With Docker:

docker build -t waggle .
docker run --rm -p 127.0.0.1:4317:4317 -v waggle-data:/data -e ANTHROPIC_API_KEY=sk-ant-... waggle

API keys and cost

All model usage bills your own API keys. Keys are only ever sent to the provider's API.

Key		Used for
`ANTHROPIC_API_KEY`	required	Every agent in the pipeline (Claude Agent SDK). A Claude plan token from `claude setup-token` (`sk-ant-oat…`) also works, and without any key an existing local `claude` login is used (on macOS it is found in the Keychain).
`OPENAI_API_KEY`	optional	Report writing, quality-check judges and cross-report reconciliation try OpenAI first (and the `codex` CLI when installed) and fall back to Claude without it.

Set them in the environment, in a .env file in the working directory, or in Settings → API keys (memory only, unless you tick Save keys on this machine, which writes <data dir>/keys.json with mode 0600).

Every run has a spend cap, $20 by default — change it in Settings → Spending or with THERESA_RUN_BUDGET. The run page shows live spend against the cap. The cap stops a run from starting new work; a step already in flight finishes, so the final cost can pass the cap by a small amount (the run page says so).

Your first scan

Connect → Public GitHub URL: paste https://github.com/<owner>/<repo> (no token needed for public repos).
Full Scan → New full scan: tick the repo, optionally write what you care about in the brief ("is our auth safe?", "why do these two numbers disagree?"), and press Run diagnosis.
Watch the pipeline live. When it completes, open the Leadership report first; the Execution report has the details and the REMEDIATION.md export.

Private code: connect a GitHub personal access token with read-only access, or a local folder on this machine (or upload one from the browser).

Connectors

All connectors are read-only.

Source	How
Public GitHub repos	Paste URLs. Cloned without credentials.
Private GitHub repos	A read-only personal access token.
Local folders	A path on this machine, or a folder uploaded from the browser. Local secrets (`.env` files other than templates such as `.env.example`, private keys, Waggle's own data folder) are left out of the scan.
BigQuery	A service-account key (JSON) with a read-only role. Queries are `SELECT`-only with a per-query scan cap.
SQL warehouse / Redis / any MCP server	The URL of a read-only MCP server (and an optional bearer token). The agent may call every tool the server exposes unless you list the allowed tools, so connect read-only servers or read-only database users.

Connection credentials (tokens, service-account keys, and credentials embedded in a connection URL) stay in memory unless you choose to remember them on this machine; they are never written to state.json or shown back in the console. Data planes are mounted only into runs you tick them for.

Privacy and telemetry

Your code, data, questions and reports stay on your machine; they are sent only to the model providers whose keys you configured, as part of the prompts the agents need.

Waggle sends anonymous usage telemetry so we can see how it is used. It is on by default, the console tells you so on first run, and you can turn it off at any time:

THERESA_TELEMETRY=0 (or DO_NOT_TRACK=1) in the environment, or the switch in Settings → Telemetry;
npm start -- --print-telemetry prints every payload exactly as it is sent.

What is sent (and nothing else):

Field	Content
`installId`	A random ID generated once (reset when you turn telemetry off).
`version`, `os`, `arch`, `node`	App version, OS family, CPU architecture, Node major version.
`event`	`start`, `quick-ask`, `full-scan` or `incremental`.
`outcome`, `durationSec`	Complete / error / stopped, and the run's duration.
`costBucket`	Spend as a range: `0`, `<1`, `1-5`, `5-10`, `10-30`, `30-100`, `100+` (USD).
`bundles`	IDs of the built-in expert lenses that ran (e.g. `appsec`).
`connectorKinds`	The kinds of sources a run used (e.g. `github`, `warehouse`).
`findings`	Counts by severity and by lens.
`degradedStages`, `failedStages`	Names of built-in pipeline stages that fell back or failed.
`keyProvider`	Which kind of key was configured (`anthropic`, `anthropic+openai`, `claude-subscription`, `claude-subscription+openai`).
`ts`	The time, rounded to the hour.

Never sent: repository names or URLs, code, file paths, questions or briefs, finding text, reports, keys, emails, hostnames or usernames. The collector does not record IP addresses. Its source is in telemetry-collector/.

Security

The console has no login: it is a single-user app. It listens on 127.0.0.1 by default, and the Docker example publishes the port on 127.0.0.1 only. Anyone who can reach the port can run scans on your keys and read your reports, so do not expose it to a network without putting your own authentication in front of it. The server also refuses requests that a web page you visit could make on your behalf: a cross-site request (CSRF), or a request under another host name (DNS rebinding).

Reports are generated by language models from code you scan, so they are served under a strict Content-Security-Policy in a sandboxed frame. See SECURITY.md to report a vulnerability.

Configuration

Variable	Default
`PORT` / `HOST`	`4317` / `127.0.0.1`	Where the console listens.
`THERESA_ALLOWED_HOSTS`	none	Extra host names the console answers to (comma-separated), e.g. behind your own reverse proxy. Only `localhost`, `127.0.0.1` and `[::1]` are served otherwise.
`THERESA_DATA_DIR`	`./.data`	Run history, reports, memory, settings.
`THERESA_RUN_BUDGET`	`20`	Per-run spend cap in USD (overrides the Settings value).
`THERESA_TELEMETRY`	on	`0` turns telemetry off.
`THERESA_LOCAL_AUDIT`	on	`0` disables scanning local folders by path.
`THERESA_CODEINTEL`	off	`1` enables the code-intelligence plane (needs repowise, AGPL-3.0, on `PATH` or `THERESA_REPOWISE_BIN`).
`THERESA_OSV`	off	OSV vulnerability lookups run automatically for public-only scans; `1` also allows them for private code (package names and versions are sent to api.osv.dev).
`THERESA_MEMORY`	on	`off` disables memory recall and writes.

Limitations

Requires an Anthropic key (or a Claude login); an OpenAI-only setup is not supported yet.
One heavy run at a time; further Full Scans queue.
Findings are produced by language models and verified against the code, but they can still be wrong. Each one carries its evidence and a "how to verify" step — check before you act.
Scanning very large monorepos is slow and expensive; select the repos or folders that matter.

Contributing

Issues and pull requests are welcome — see CONTRIBUTING.md. Run npm run typecheck, npm test and npm run check:english before opening a PR.

License

Waggle is licensed under the Apache License 2.0. It depends on the Claude Agent SDK, which is installed from npm and governed by Anthropic's own terms. See NOTICE for third-party components.