August 2026
Coding agents are surprisingly good at learning undocumented APIs. The problem is they usually forget what they learned when the session ends.
For one user it’s easily solved with an AGENTS.md
or
folder of notes. For a team just commit those notes to Git.
But what if thousands of people are independently exploring the same undocumented API with their agents?
I ran into this problem while building Modmixer, a coding agent for modding games. Modding involves reverse-engineering binaries, finding injection points, and learning undocumented runtime behavior through experimentation.
With many people modding the same games, I wanted a discovery made by one person’s agent to become useful to everyone. Instead of each user spending tokens and compute rediscovering the same things, the community should gradually build up a shared memory of everything its agents have learned.
In practice, this means turning discoveries into small, reusable findings that make sense for the domain. In Modmixer, each finding has a title that says what was learned, a recipe with the working code or steps, and a “why it’s tricky” line describing the plausible wrong turn an agent would otherwise take.
Once you have a useful unit of knowledge, the rest of the system is simple:
The server has two database tables: submissions
and
curated
.
When an agent makes a discovery, it gets a unique ID, is uploaded to
submissions
, and is cached locally so the agent can use it immediately.
Submissions are untrusted. An agent skill periodically reviews them
and inserts accepted findings into the curated
table. It can:
Each curated finding records the IDs of the submissions it was derived from, allowing clients to remove superseded findings from their local cache.
The result is a shared memory that gets better as more people use it.