@companion — Shared Agent Feedback / Memory System

Problem

Feedback learned in coworker-agent sessions (Claude Code, lilith-platform.live) lives in two disconnected places:

lilith-platform.live/.claude/projects/.../memory/ — Claude Code auto-memory, loaded in CC sessions only
agents/coworker-agent/memory/ — coworker nag loop memory, loaded by the loop agent

Neither feeds @companion. When @companion ships, it will have no awareness of lessons already learned across dozens of coworker sessions.

Specific example

Learned Apr 11: never infer the current time from session context — always run date. Wrote to CC memory and coworker memory, but @companion will start blind.

What's needed

A shared feedback/memory store that:

coworker-agent reads and writes
@companion reads at session start
Claude Code sessions can contribute to

Options

Single canonical path — e.g. ~/Code/@projects/@companion/data/agent-memory/ — all agents read from here, coworker-agent writes here, CC memory points here
quinn-my API — add a feedback resource to the MCP server; all agents read/write via API (survives across machines, accessible from VPS)
Hybrid — local file store for dev, quinn-my API for prod

Recommendation

Option 2 (quinn-my API) is cleanest long-term — single source of truth, accessible everywhere, queryable. Short-term, use a shared local path until the API endpoint exists.

Priority

Before @companion v1.1 (memory milestone). Not blocking v1.0.

1.5 KiB Raw Permalink Blame History