How to build an agent approval layer
Permission Slip was built over eight weeks with unnamed AI coding agents and human architecture and review. It gates structured account actions through configured rules and phone approvals.
news.ycombinator.comSuggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Help build [approval service] between [agent] and [account services]. Validate structured actions, apply configurable approval rules and keep credentials local. Add phone approval requests and a CLI. Present architecture and code for my review, then test in [development environment].
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- chiedo
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The iPhone app is live and Google and Slack integrations are the most developed; the project remains a beta with rough edges.
- Human involvement
- The author explicitly handled architecture and review and has run the project only in local development and a hosted beta.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Built a schema-validated action approval layer with AI coding agents.
- Developed service connections, with Google and Slack most fleshed out.
- Released the iPhone app and ran local development and a hosted beta.
The author built Permission Slip as an approval layer between an AI agent and connected accounts. They originally made it for OpenClaw, but say it also works with local Claude Code and other agents. The coding agents used to build the project are not named. Over eight weeks, those agents did most of the typing while the author directed architecture and review, often working from a phone.
The implementation accepts structured, schema-validated actions and decides whether to allow them according to configured rules. Read actions can be automatically approved, while destructive or irreversible actions can require a phone tap. Credentials stay on the box. Connections use OAuth for services such as Google, Microsoft and Slack, or an API key, and an agent uses the Permission Slip CLI to request actions.
The author also built an iPhone app, which they report is live. Google and Slack are the most developed integrations. They have run the system in local development and a hosted beta environment; deploying their own Raspberry Pi instance remains a future milestone. The hosted version is described as a beta playground where users should not put sensitive information.
Who it's for: Builders who want explicit human approval around an agent's account actions and can inspect and operate a self-hosted service.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · chiedo
Frequently asked questions
Which coding agent built Permission Slip?
The author says AI coding agents did the typing while they handled architecture and review, but does not identify those coding agents. OpenClaw is the intended client.
Was Raspberry Pi hosting tested?
No. The author had only run local development and the hosted beta environment. Their own Raspberry Pi instance was the next milestone, and its performance had not been tested.
More details
About this account
Unnamed AI coding agents did most of the typing during the eight-week side project.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47874972. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: chiedo on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
AI coding agents · Permission Slip CLI · OAuth · Permission Slip iPhone app
Agent approvals · Self-hosting · OAuth
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.