Show HN: Sharing Agentic Stream of Consciousness

There needs to be a universal format to share conversations with AI that captures tokens from multiple sources in one stream of conversation.

Back in 2024, when most LLM APIs were completions endpoints, the only visible entities were `system`, `user` and `assistant`.

Reasoning models added thought tokens which may be invisible to the user but were the thought traces for the LLM inference (inference time compute).

Now with agents making tool calls, and there are other entities also participating and influencing the flow of a conversation.

I took a stab at vibe-composing a `SKILLS.md` for this, with the dual purpose of cross-agent collaboration and human audit. - Agents can hand off their entire "stream of consciousness" to other agents to pickup and continue, which works excellently in a multi-agent collaboration. - Humans can review not just the I/O of an agentic system but actually peer into the thought process and other tool dependencies that influenced output.

Summary

The article discusses the concept of Stream of Consciousness, a literary technique that aims to capture the natural flow of a character's thoughts, feelings, and perceptions without the limitations of traditional narrative structure. It explores how this technique can be applied in various forms of creative writing to provide readers with a more immersive and authentic experience.

Story

Show HN: Sharing Agentic Stream of Consciousness