[ Idea ]Found 12 Nov
Exorcist
Deterministic approval guard for coding agents
Claude Code, Codex, and Kiro already ship built-in approval prompts and permission rules, so the guard adds little.
1 products · 0 sources13 Nov
Deterministic approval guard for coding agents
A single wrong command can delete work, leak secrets, or break a build, and developers often only notice after the damage is done. Teams that want to use agents safely end up approving every step by hand, which removes most of the benefit.
Exorcist sits between the agent and your machine and applies fixed approval rules to every action before it runs.
- 1Define which commands, files, and paths the agent may touch
- 2The agent proposes an action and Exorcist checks it against your rules
- 3Allowed actions run automatically, and blocked or unclear ones wait for your approval
[ The six checks ]Failed at check 1 of 6
1Not built in alreadyFailed here
2An AI agent can't do itNot checked
3Nothing similar existsNot checked
4People pay for thisNot checked
5Strong scoresNot checked
6Real sourcesNot checked
1 · Not built in alreadyFailed here
2 · An AI agent can't do itNot checked
Why it failed
Check 1 · Already built inNatively solved
Who paysDevs
Who uses itDevs
What already exists
1 products checkedProductFixed, repeatable rule checksChecks actions before executionAsks you on uncertain actionsRecords every decision madeOverlap with idea
This ideaWhat Exorcist plans to do
4 / 4
From Notion notesAuto-approves agent actions via Claude Code auto mode and Kiro Approve for me, with partial coverage of the idea.
[ What would make it pass ]3 steps
- 01
Find a gap native permissions miss, like one policy across Claude Code and Codex.
- 02
Interview five heavy agent users on what their current permission setup gets wrong.
- 03
Only proceed if three of them say they would pay for a fix today.
[ Inspired by ]What it borrows, and what's different
- -
-