claude

meldwerkes

Meldwerkes: multi-agent cognitive architecture that builds a personalized decision model through PAHF loops, distillation, and multi-brain orchestration.

4 on GitHubby Darrell Rossv0.7.0MITupdated 2 days ago
/

Plugin meldwerkes from marketplace eukota-claude-marketplace in eukota/claude-marketplace. Commands read from the repo's manifest — yesterday.

This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.

What it installs

Skills (10)

  • meldwerkes-bootstrap

    Pretrain minds from existing Claude Code chat history in two phases — survey what domains are present and propose a set of minds, then create them and import. Use when starting from scratch and wanting the model seeded from prior work rather than empty.

  • meldwerkes-calibrate

    Test how well a mind predicts its owner — proposes questions, predicts each answer from existing principles before the user responds, then scores agreement with correlation, calibration error, and a per-domain breakdown. Use when the user wants to check whether the model actually knows them, or to s

  • meldwerkes-capture

    Turn Meldwerkes learning on or off, and report whether it is currently recording. Use when the user wants to pause capture — NDA work, someone else at the keyboard, a throwaway experiment — or asks whether the plugin is currently learning.

  • meldwerkes-export

    Export a brain to a portable JSON file for backup, sharing, or transfer.

  • meldwerkes-history

    Checkpoint, retract, and restore the accumulated model — remove a time range of learning (a bad day, a paused period that got recorded anyway), list snapshots, and roll back to any of them. Use when the user wants to undo learning, remove recent decisions, or recover from a change that made the mode

  • meldwerkes-import

    Import a brain from a portable JSON export. Supports provisional trust (confidence discounted) or full trust.

  • meldwerkes-report

    Generate a Meldwerkes status report — brains, memory state, decision stats, principles, conflict history.

  • meldwerkes-review

    Review accumulated principles with the user and apply reinforcement — confirm what is still true (resetting temporal decay), weaken what is partly wrong, reword what is clumsily stated, retire what was never right. Use when the user asks to review, prune, reaffirm, or check what the model believes a

  • meldwerkes-setup

    Set up Meldwerkes — create the first brain, configure settings, verify the memory store is ready.

  • pahf-compress

    Run the distillation loop — extract principles from recent decisions and corrections, detect brain split signals.

Hooks (2)

  • SessionStart

    cat "$HOME/.small-brain/session-context.json" 2>/dev/null || true

  • UserPromptSubmit

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/capture_prompt.py" 2>/dev/null || true

View source on GitHub

More from eukota/claude-marketplace

All 7 plugins →

interview-skills

claude

Brief interview skills for candidates in AI-assisted technical interviews: configure, clarify, plan, direct, review, verify, build integration tests with a dashboard, and narrate.

8 skills

4

pahf

claude

Personalized Agents from Human Feedback — PAHF loop for Meldwerkes: pre-action clarify, ground in memory, post-action correct, compress to principles.

4 skills

4

plan-usage

claude

Tracks Claude account rate-limit usage (5-hour and 7-day windows) from the status line, and reports whether your plan is the constraint on your work.

1 skill · 2 commands

4

plugin-builder

claude

Tools for building Claude marketplace plugins — choosing primitives, checking name availability, smoke-testing hooks, scaffolding structure, and following conventions.

2 skills

4

context-setup

claude

Personal project context management system for engineers – structured Markdown + YAML that gives your AI assistant persistent, organized memory across sessions and machines. Auto-maps each project's workspace folders into per-project permissions.

1 skill

4

local-dev-standards

claude

Enforces standard local development practices — git worktrees, containerized package installs, Makefile start/stop targets, and hot reload.

4

Maintain meldwerkes?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins7 plugins