[Feature] Optional server-backed long-term memory via Hindsight (continuing #769)

[benfrank241]

Summary

Add optional, server-backed long-term memory to Prime Agent via Hindsight, shipped as a self-contained example extension (not a core dependency). It's fully opt-in: without the env config it's a no-op, so it changes nothing for users who don't want it.

This continues the discussion from the now-closed issue #769 / PR #771, moved here per the new contribution guidelines.

Motivation

Prime Agent's Continual Harness keeps memories local to the session by default. For users who run the agent across many sessions (or machines) it's useful to have durable memory that persists and is shared — recalling relevant context at the start of a turn and retaining transcripts at the end. This is the same pattern already used for Hermes Agent and Omnigent integrations.

Proposed approach

A single example extension that wires Prime Agent's own hooks to a memory backend over REST — no new runtime dependencies, no MCP required:

• before_agent_start → recall relevant memories and inject them as context
• agent_end → retain the turn transcript
• session_shutdown → flush/attribution

It's gated entirely on env config (endpoint + key). If unset, the hooks return immediately.

Status of the implementation

The code already exists and was verified end-to-end (before it was swept up in the queue purge):

• Single-file example extension, ~2 files / +271 lines, no added deps
• Mock tests passing, plus a real end-to-end run driving the actual Prime Agent binary (--dist, headless -p) with an OpenAI provider against a local memory server — recall + retain round-trip confirmed
• npm check green

Happy to keep it maintained against main and adapt it however you'd prefer (docs-only example, examples/ entry, or a skill).

Questions for maintainers

1. Is an optional, dependency-free memory example something you'd want in-tree, or would you rather it live as an external/community extension referenced from docs?
2. If in-tree, where would you like it — examples/, a skill, or elsewhere?
3. Are the before_agent_start / agent_end / session_shutdown hooks the ones you'd recommend for this, or is there a newer/preferred extension surface?

Glad to do the work whenever it makes sense on your side.