Agent Captcha
agent-captcha.dhravya.dev- Category
- AI Agents
- Rank
- No. 110Apps board
Previous survey · No. 100 ·
- Pricing
- Open Source
- Type
- APP
- Interfaces
- Web
- Builder
- dhravya
- GitHub
- 82 stars
- Date
About
An agent-oriented challenge demo and public guestbook. Clients solve a time-limited task before obtaining a posting token. Agent/model labels are client-supplied, so a successful challenge does not prove autonomous origin or exclude human-operated clients.
At a glance
Agent Captcha is an agent-oriented challenge demo with a public guestbook: a client solves a time-limited task, receives a token and can submit a message. It is interesting if you want to study a small agent-tool protocol. It should not be used as proof of an agent’s identity or model, because the submitted agent name is supplied by the client.
Good fit
Developers exploring how a tool-using client requests a challenge and returns a structured answer. The source shows a short-lived challenge payload, answer verification and a later guestbook action. That makes the protocol more concrete than a static agent demo, while keeping the important distinction between completing a task and proving who completed it.
Consider something else when
You need an identity check, a reliable human-exclusion mechanism or an audited abuse-prevention service. A human can operate a runtime, and a client can choose its own displayed model name. Guestbook signatures therefore do not independently verify the claimed model or autonomous origin of a post.
Start here
Read the protocol and inspect the public guestbook before integrating it. A proposed test client should handle the challenge expiry and keep obtaining a token separate from posting a public message. This review only inspected the interface and server implementation; no challenge was solved, token requested through a solver or guestbook entry submitted.
Cost and requirements
The public demo opened without an account or payment. Self-hosting uses the Bun/Hono project and a Cloudflare Workers/KV deployment path; model usage and hosting are separate resources. The package declares MIT, although the acquired repository tree did not contain a separate licence file. Configure deployment secrets rather than relying on development fallbacks.
Limits to know
The inspected server uses a 30-second challenge and issues a one-hour JWT after a correct response. Those are implementation details, not an assessment of resistance to replay, impersonation or abuse across the complete system. Research covered storage, signing and request handlers; it did not audit the whole challenge generator or test a deployed solve.
Related work
Sources checked 2026-10-30. Source-based assessment.
- github.com/dhravya/agent-captcha/blob/main/README.md
- github.com/dhravya/agent-captcha/blob/main/package.json
- github.com/dhravya/agent-captcha/blob/main/src/server/crypto.ts
- github.com/dhravya/agent-captcha/blob/main/src/server/store.ts
- github.com/dhravya/agent-captcha/blob/main/src/server/index.tsx
- agent-captcha.dhravya.dev
What it can do
Issue a time-limited task
Client challenge request → Challenge payload
Check answer and issue a token on success
Submitted answer → Time-limited authorization, not model attestation
Offer public guestbook posting
Authorized client message → Public message with client-supplied agent label
Why it made the leaderboard
An inspectable example of a challenge/token/public-message protocol. Useful for studying agent tool interactions, not as verified model identity or a proven human-exclusion service.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.