๐—ช๐—ต๐˜† ๐—ฌ๐—ผ๐˜‚๐—ฟ ๐—ฆ๐˜‚๐—ฏ-๐—”๐—ด๐—ฒ๐—ป๐˜ ๐—™๐—ฎ๐—ถ๐—น๐˜€ ๐—ฌ๐—ผ๐˜‚๐—ฟ ๐— ๐—ฒ๐—บ๐—ผ๐—ฟ๐˜† ๐—ฅ๐˜‚๐—น๐—ฒ๐˜€

I gave a sub-agent a task. It committed code to my main branch. It skipped the PR. I spent 30 minutes fixing the mess.

I thought the agent knew my rules. I have 120 feedback files. I assumed the agent inherited these files.

I was wrong. Sub-agents work in a sandbox. They only know the brief you write. They do not see your personal memory index.

If a rule is not in the brief, it does not exist for the agent.

You must inline critical rules.

Memory does not happen by inheritance. It happens by explicit briefing. Put the discipline in the brief before you click delegate.

Source: https://dev.to/michelfaure/why-your-sub-agent-doesnt-load-the-same-memory-as-you-and-how-it-commits-to-main-behind-your-back-cgi