MOLTRON
What it is
Section titled “What it is”Self-Evolving AI Agent Framework. Deploy autonomous AI agents that code their own upgrades. Uses Skills.md to enable agents to build, evolve, and compound their capabilities on the fly.
What we kept
Section titled “What we kept”- Self-evolving capability via worker-tagged learnings. Our
TRICK:convention inraw.mdis the structural equivalent. Workers flag generalisable patterns; the curator promotes them toidentity/worker.mdafter recurrence.
What we dropped
Section titled “What we dropped”- Agents codes their own upgrades. We don’t let workers modify their own role prompt. Capability evolution happens via the curator’s promotion of identity entries, not by workers rewriting
prompts/worker.md. This is a deliberate guard against runaway prompt engineering.
Differences in philosophy
Section titled “Differences in philosophy”| MOLTRON | Papercup |
|---|---|
| Workers can write their own Skills.md | Workers append TRICK: entries; curator promotes |
| Capability compounding is the agent’s job | Capability compounding is the curator’s job |
Honest take
Section titled “Honest take”The “self-evolving” framing is provocative. The conservative reading — workers flag patterns, a separate role triages — is what we built. If you want capabilities that the agent itself proposes and adopts, MOLTRON’s model is more aggressive.
We chose the conservative version because we want determinism: a mission’s prompts are stable for the duration of the mission, and capability changes are auditable curator commits.