Timmy/claw-agent
1
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects Releases Wiki Activity

[PHASE 1] Background compaction + promoted memory markers (Hermes-centered) #1

New Issue
Open
opened 2026-04-05 12:46:20 +00:00 by Timmy · 0 comments
Timmy commented 2026-04-05 12:46:20 +00:00
Owner
Copy Link

Parent context

  • timmy-home #408 — Claw Code clean-room roadmap
  • timmy-home #154 and child study tickets — Claude Code source study
  • External references: Claude Code patterns, Magic Context historian/dreamer/sidekick

Directive

Alexander's architecture decision is explicit:

Plug and play. Hermes always stays at the center.

This means claw-agent is not building a rival center.
It is building a clean-room capability slice that plugs into Hermes.

Phase 1 goal

Implement the first differentiated Claw capability:

background context compaction + promoted memory markers

What to build

1. Background historian

A background worker/service that reads completed or aging conversation state and produces compacted summaries without blocking the main coding loop.

2. Promoted memory markers

A schema for promoted summaries / markers that can be rehydrated later as project memory.

3. Rehydration interface

A minimal way for the main runtime to ask for relevant promoted summaries and inject them back into task context.

Architectural constraints

  • Hermes remains the center runtime/orchestrator.
  • claw-agent is the clean-room implementation lane.
  • timmy-config sidecar will own deployment and wiring later.
  • Do not copy code from claude-code-src.
  • Code from this issue must be clean-room authored from behavior/spec notes only.
  • Do not waste time on shell UX / full CLI theater in this issue.
  • Focus only on the differentiated service wedge.

Suggested deliverables

  • Minimal historian module/service in claw-agent
  • Summary/marker schema (JSON or SQLite-backed)
  • Rehydration API or helper function
  • Tiny end-to-end demo showing:
    1. raw session history enters
    2. background compaction runs
    3. promoted memory markers are produced
    4. next task can consume them
  • README note explaining how this plugs into Hermes later

Acceptance criteria

  • Main loop does not block on compaction
  • Promoted memory is durable and queryable
  • Rehydration path exists and is demonstrated
  • Implementation is clearly clean-room
  • Output is useful enough that timmy-config can later wire it into Hermes as a sidecar-owned feature

Non-goals

  • full Claude Code clone
  • replacing Hermes
  • direct shipping from claude-code-src
  • product naming/branding work

Filed by Timmy.

## Parent context - timmy-home #408 — Claw Code clean-room roadmap - timmy-home #154 and child study tickets — Claude Code source study - External references: Claude Code patterns, Magic Context historian/dreamer/sidekick ## Directive Alexander's architecture decision is explicit: > Plug and play. Hermes always stays at the center. This means `claw-agent` is **not** building a rival center. It is building a clean-room capability slice that plugs into Hermes. ## Phase 1 goal Implement the first differentiated Claw capability: **background context compaction + promoted memory markers** ## What to build ### 1. Background historian A background worker/service that reads completed or aging conversation state and produces compacted summaries without blocking the main coding loop. ### 2. Promoted memory markers A schema for promoted summaries / markers that can be rehydrated later as project memory. ### 3. Rehydration interface A minimal way for the main runtime to ask for relevant promoted summaries and inject them back into task context. ## Architectural constraints - Hermes remains the center runtime/orchestrator. - `claw-agent` is the clean-room implementation lane. - `timmy-config` sidecar will own deployment and wiring later. - Do not copy code from `claude-code-src`. - Code from this issue must be clean-room authored from behavior/spec notes only. - Do not waste time on shell UX / full CLI theater in this issue. - Focus only on the differentiated service wedge. ## Suggested deliverables - [ ] Minimal historian module/service in `claw-agent` - [ ] Summary/marker schema (JSON or SQLite-backed) - [ ] Rehydration API or helper function - [ ] Tiny end-to-end demo showing: 1. raw session history enters 2. background compaction runs 3. promoted memory markers are produced 4. next task can consume them - [ ] README note explaining how this plugs into Hermes later ## Acceptance criteria - [ ] Main loop does not block on compaction - [ ] Promoted memory is durable and queryable - [ ] Rehydration path exists and is demonstrated - [ ] Implementation is clearly clean-room - [ ] Output is useful enough that `timmy-config` can later wire it into Hermes as a sidecar-owned feature ## Non-goals - full Claude Code clone - replacing Hermes - direct shipping from `claude-code-src` - product naming/branding work --- Filed by Timmy.
kimi was assigned by Timmy 2026-04-06 03:31:35 +00:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
No items
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
Timmy
No Assignees kimi
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Timmy/claw-agent#1
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?