Epismo
A platform to share and manage powerful human-AI workflows
Start new thread
Hiroki

Epismo Context Pack - Portable memory for agent workflows

by
Epismo
•
Context Pack is portable memory for agent workflows. Turn prompts, plans, decisions, project context, and hard-won know-how into reusable packs you can fetch across agents and threads. Keep them private, share them with your team, or publish them for the community, so others can reuse proven context instead of starting from scratch. Works across MCP and CLI, with support for cloud agents, local setups, Slack, and Discord.

Add a comment

Replies

Best
Hiroki
Maker
📌
We built Context Pack because valuable context keeps getting trapped inside one chat, one tool, or one moment. That leads to a lot of manual work: moving context between agents, re-explaining the same project in new threads, pasting old prompts again, or rewriting good discussions into docs just to share them. Context Pack makes that reusable. A pack is a title + content set. You can use it for prompts, plans, decisions, project context, or hard-won know-how. You can fetch titles first, load full content only when needed, and use your context window more efficiently. One of the most exciting parts is that packs are not limited to private use. You can keep them private, share them with your team, or publish them for the community. That means Context Pack is not just about saving your own memory. It is also a way to reuse proven context from others. AI power users can publish the memory behind their workflows, prompts, research habits, and playbooks, and others can build on that instead of starting from scratch. To get started, you can simply tell your agent: `Set up Epismo access and load the Skills from https://github.com/epismoai/skills` The Skills are designed for both MCP and CLI, so you can use Context Pack with cloud-based agents like ChatGPT or Claude, local setups like Claude Code or Codex, and even Epismo agent on Slack or Discord. For a first example of how to share a Context Pack: `/context-pack @hirokiyn/context-pack` Would love to hear how you’d use it.
swati paliwal

@hirokiyn Congrats. How would you recommend structuring a shared pack for a team's 'personal branding playbook' to make it plug-and-play for Claude or ChatGPT agents?

Hiroki
Maker

@swati_paliwal Thanks! Honestly, agents can handle a lot of this.

I’d keep a few core sections stable, then let Claude or ChatGPT agents interpret and apply the pack for each task

Abhra Das

The part about reusing context from other people's workflows is interesting. If I load someone's published context pack, does it just give me their prompts or does it actually carry over the decisions and reasoning behind why they built it that way?

Hiroki
Maker

@abhra_das1 Not just prompts.

A Context Pack can carry the surrounding context too, like decisions, rationale, project background, conventions, and other working knowledge behind the workflow.

The goal is to reuse understanding, not only reuse prompt text.

Mykola Kondratiuk

how do you handle context conflicts when multiple agents write to the same memory pack?

Hiroki
Maker

@mykola_kondratiuk Good question!

We keep track of how often a pack is used and favorited, so higher-value context naturally stands out over time while lower-value entries get pruned. A good pattern on top of that is to have agents periodically run an organize workflow to clean up, merge, and reconcile overlapping context.

Mykola Kondratiuk

That makes sense. Usage + favorites is basically organic quality filtering - the useful stuff surfaces itself. Works well as long as the initial pool stays manageable.

Hiroki
Maker

@mykola_kondratiuk there might be better filtering in the future, although usage + favorites can work as organic quality filtering for humans.

For agents, I think the better model is ranking context by things like reuse success, consistency, freshness, and task relevance, so the most useful context gets selected more often over time.

Mykola Kondratiuk

reuse success and consistency are the right signals for agents. freshness I'd weigh lower - stale context often still works if it's stable.

Manash Pratim

Re explaining context in every workflow gets tiring fast. This idea of carrying memory across sessions feels useful. Keeping that memory updated over time is the tricky part.