zeref-os
SolidLocal-first memory engine for AI coding agents. Persistent project memory that survives context loss, shared across Claude, Cursor, and Codex. Privacy-first, evidence-graded, human-arbitrated.
Install
Plugins install via a marketplace, in two steps.
This plugin isn't listed in a marketplace we've indexed. Install it directly from its GitHub repository — the README has the setup steps.
View on GitHubBundles
Everything this plugin ships — skills, agents, commands, hooks, and MCP servers it bundles.
Skills (12)
caveman-handoff
Companion to handoff-compiler. Compresses session handoff into caveman-grammar payload (drop articles / filler / pleasantries / hedging; preserve technical substance, file paths, exact errors verbatim, code blocks unchanged). Survives cross-model + cross-harness switches with compressed handoffs; reduction varies by content (unmeasured; no benchmark backs a fixed ratio). Carries prompt-context-engine brief diff to satisfy R6 zero-context-loss.
contradiction-resolution
Detects conflicting claims in wiki state (flat memory/ layout), surfaces them via memory/CONFLICTS.md, supports snooze-until-/done, never silent-resolves. User arbitrates every conflict.
evidence-grader
Skill counterpart to evidence-curator agent. Provides on-demand grading for ad-hoc claims, source assessments, and review passes. Operates over flat memory/ layout.
handoff-compiler
Produces compact cross-model/cross-harness handoff package from flat memory/ layout. Invoked by handoff-orchestrator on /stop or model switch. References harness translation map.
memory-import-export
Imports/exports memory across tools, repos, and formats. Migration, onboarding, backup. Explicit user request only.
parent-sync
Stages approved summaries + decisions in memory/sync/outbound/, pushes to parent's memory/sync/parent/<child-id>/. Preserves provenance. Surfaces parent-side conflicts. Approval required every push.
pattern-to-skill
Drafts new skill files from candidates emitted by pattern-observer. Drafts land in skills/drafts/ — never auto-activate. User reviews via /review-skill to approve, edit, reject, or defer.
privacy-abstraction
Rewrites payloads to remove PII, internal paths, credentials per root REDACT.md classes and PRIVACY.md mode. Called by privacy-guardian before writes when mode = abstract.
prompt-context-engine
Auto-gate #3. Auto-classifies every incoming user prompt as STRUCTURED, SEMI-STRUCTURED, or UNSTRUCTURED. Rewrites UNSTRUCTURED prompts into a structured task brief with objective, deliverable, constraints, context, and success criteria before any execution begins. Strips redundant context, suggests prompt caching, and preserves all user intent with zero context loss (per `_shared/rules.md#R6`). Activates automatically on every major task without explicit user invocation.
skill-importer
Pull a skill from the user's broader skill directory into this Zeref project. Provenance-tracked, review-first — imports never auto-activate. Compatible with sandbox reference-only mode (see skills/imported/*/README.md).
skill-router
Auto-gate
wiki-maintenance
Maintains canonical wiki state (flat memory/ layout) — index consistency, page summaries, decisions log, risks register, open questions queue. Activates after writes and on consolidation requests.
Agents (6)
Commands (8)
Quality Score: 59/100
Details
- Author
- kanadhiayash
- Repository
- kanadhiayash/zeref-memory-engine
- Created
- 2 months ago
- Last Updated
- today
- Language
- Python
- License
- MIT