Handlebar Python — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Handlebar Python (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
[Handlebar] is a runtime control layer for your AI agents.
Enforce deterministic rules on your agents as they act, so you can guarantee they don't violate your team's policies.
| Without Handlebar | With Handlebar |
|---|---|
| "Whoops the agent deleted prod DB" | Deterministically block dangerous tool actions. Full auditability into what your agent _tried_ to do. |
| "Our costs are ballooning with no way to control them" | Track token usage and USD spend, and set hard limits on your agents. When the limit is reached, Handlebar can block the agent from taking further actions. |
| "Someone convinced the agent to leak another user's emails" | Limit tool permissions to the user. |
| "The agent is going off-the-rails and spamming heavy APIs" | Set rate limits on tool use and prevent runaway actions |
| "We can't be sure the agent isn't leaking sensitive data" | Enforce hard data boundaries between tools and your output. Filter PII before it leaks through agent context |
send_email(internalAddress) -> PASS | send_email([email protected]) -> BLOCK)get_pii -> send_slack_message -> BLOCK: risk of data exfil)You will need:
>= 3.11)This repository is a monorepo containing installable packages for different Python agent building frameworks. We provide some pre-built wrappers for agent frameworks, with more on the way soon. If your agent is not directly supported, you can still easily plug Handlebar into your agent.
| Framework | Install command | Where to read more |
|---|---|---|
| google-adk | uv add handlebar-google-adk | Integration guide |
| Langchain (Python) | Soon... | |
| Other frameworks + custom agents | uv add handlebar-core | Custom integration guide coming soon |
| Javascript agents (Langchain, Vercel ai etc.) | bun i @handlebar/core | Checkout the Handlebar JS codebase |
The client SDKs interact with the Handlebar API to emit agent telemetry and event data it collects, and to evaluate your configured policies.
Sign up at [https://app.gethandlebar.com][platform].\ If you are waitlisted, get in touch with us to get access.
Once on the platform, create an API key and activate your agent by setting the HANDLEBAR_API_KEY environment variable in your agent codebase.
On the [platform] you can create policies from simple templates: usage limits, dangerous tool use, GDPR, finance agents, and more.
Alternatively, run the Handlebar claude code skill to generate rules custom to your agent, by running:
npx skills add gethandlebar/agent-skillsGo to the skill repository for full instructions.
This repo uses [mise][mise] for command utilities
mise r format # Formats code
mise r ci # Lints, checks types, runs testsPlease open an issue if you have any feedback, suggestions, or requests for framework support. Alternatively, [book a call][calendar] to talk to us about how Handlebar could help to protect your team's agents.
The SDKs defined under packages/ are currently licensed under Apache 2.0 LICENSE.
[handlebar]: https://www.gethandlebar.com [platform]: https://app.gethandlebar.com [calendar]: https://calendly.com/arjun-handlebar/30min [docs]: https://handlebar.mintlify.app [mise]: https://mise.jdx.dev/
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.