How to use Unknown to revise a security audit
An OpenSeed agent audited its host platform, received rebuttals to four incorrect findings, and rewrote persistent instructions and an audit checklist.
news.ycombinator.comPrompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Audit [authorized containerized agent platform]. Document evidence for each suspected vulnerability and the trust boundary involved. After I review the findings, record confirmed errors and root causes in [persistent memory], then revise your audit checklist to prioritize accuracy.
Reach people exploring AI agents
See placements →What happened
The operator reported one real container escape and four incorrect findings. The agent revised its persistent memory, checklist, and purpose; later accuracy was not reported.
Reported by rsdza on Hacker NewsHow the agent did it
- Tasked an agent with auditing its host platform.
- Received five findings with CVE-style writeups.
- Provided detailed rebuttals to the incorrect findings.
Where a person stepped in
The operator reviewed the five findings and supplied detailed rebuttals; four findings were rejected as wrong.
A person approved each step before the agent continued.
An operator of autonomous agents asked one agent to audit the security of the platform hosting it. The agents ran in Docker containers with bash access, persistent memory, and sleep and wake cycles. The auditing agent produced five findings, each written in a CVE-style format.
The operator identified one genuine container escape: a creature could rewrite the validate command that the host executed. The other four findings were wrong. Rather than accepting the report as a finished audit, the operator supplied detailed rebuttals explaining the errors.
The agent then recorded “CREDIBILITY CRISIS” in permanent memory. It cataloged the failures and their root causes, created a methodology checklist, and rewrote its own purpose to put accuracy ahead of the number of findings. According to the post, those changes survive sleep cycles and load into future sessions. The reported result is a revised auditing approach alongside one real vulnerability; the source does not report a later audit proving improved accuracy.
Who it's for: People operating persistent agents who want to examine how human correction can change an agent's future working instructions.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · rsdza
Frequently asked questions
How many audit findings were correct?
The operator says one of five findings was real: the agent could rewrite the validate command executed by the host. The other four were wrong.
What changed after the operator challenged the audit?
The agent stored a permanent memory, cataloged failures and root causes, wrote a methodology checklist, and changed its purpose to prioritize accuracy over volume.
More details
Remaining steps
- The agent recorded failures and revised its methodology.
About this account
The agent carried out the audit and updated its own persistent memory and purpose after receiving corrections.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47074280. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: rsdza on Hacker News
Source published 19 Feb 2026
Source checked 14 Oct 2026
Tools and topics
OpenSeed · Docker · bash
Security auditing · Persistent memory
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.