memq-planning

name	memq-planning
description	Durable plan state management for multi-step AI agent workflows. Checkpoint, resume, and coordinate complex tasks across sessions using MemQ's hosted governed plan state engine.
version	1.0.0
homepage	https://multinex.ai/memq
tools	["plan_state_read","plan_state_write","plan_state_checkpoint","plan_state_resume","bridge_sync","reflection_handoff","slice_project","hybrid_retrieve"]

MemQ Planning Skill

Durable plan state management for multi-step AI agent workflows. Use this skill to persist, checkpoint, and resume complex plans across sessions — ensuring no work is ever lost.

Prerequisites

MemQ MCP is hosted at https://mcp.multinex.ai/mcp/v1 and governed through Billing Manager OAuth/API-key access. An active subscription with contract tools is required: MemQ signup.

Core Concepts

Plan State

A plan is a durable state object that tracks:

• Status: draft → active → checkpointed → completed/failed
• State patch: Arbitrary JSON representing current progress
• Summary: Human-readable description of current state
• Checkpoints: Named snapshots you can resume from
• Tags: Searchable labels for organization

Plan Lifecycle

┌─────────┐     ┌─────────┐     ┌──────────────┐     ┌───────────┐
│  draft   │ ──▷ │  active  │ ──▷ │ checkpointed │ ──▷ │ completed │
└─────────┘     └─────────┘     └──────────────┘     └───────────┘
                     │                   │
                     ▽                   ▽
                ┌─────────┐        ┌──────────┐
                │  failed  │        │  resumed  │
                └─────────┘        └──────────┘

Workflow Patterns

Pattern 1: Simple Task Tracking

For single-session tasks that benefit from durable state:

# Start
plan_state_write(
  plan_id: "refactor-auth-middleware",
  status: "active",
  summary: "Refactoring auth middleware to support OAuth2 + API key dual auth",
  state_patch: { files_modified: [], tests_passing: false },
  tags: ["refactor", "auth"]
)

# Update progress
plan_state_write(
  plan_id: "refactor-auth-middleware",
  summary: "Completed OAuth2 flow, working on API key validation",
  state_patch: { files_modified: ["middleware.ts", "oauth.ts"], tests_passing: true }
)

# Complete
plan_state_write(
  plan_id: "refactor-auth-middleware",
  status: "completed",
  summary: "Auth middleware refactored: OAuth2 + API key dual auth with 100% test coverage"
)

Pattern 2: Multi-Session with Checkpoints

For complex tasks that span multiple conversations:

# Session 1: Start work
plan_state_write(
  plan_id: "billing-overhaul",
  status: "active",
  summary: "Phase 1: fixing checkout flow",
  state_patch: { phase: 1, completed_items: ["checkout-api", "webhook-handler"] }
)

# Checkpoint before ending session
plan_state_checkpoint(
  plan_id: "billing-overhaul",
  label: "phase-1-complete",
  summary: "Checkout flow working, webhook handler deployed"
)

# Session 2: Resume from checkpoint
plan_state_resume(
  plan_id: "billing-overhaul",
  checkpoint_id: "<checkpoint-id-from-above>",
  resume_reason: "Continuing with Phase 2: status dashboard"
)

Pattern 3: Execution Bridge Sync

Synchronize plan state with execution outcomes:

# Before executing a deployment
bridge_sync(
  plan_id: "production-deploy",
  phase: "pre_execution",
  outcome: "success",
  summary: "All tests passing, ready to deploy"
)

# After deployment
bridge_sync(
  plan_id: "production-deploy",
  phase: "post_execution",
  outcome: "success",
  summary: "Deployed to production, health checks passing",
  request_checkpoint: true
)

# On failure
bridge_sync(
  plan_id: "production-deploy",
  phase: "post_execution",
  outcome: "failure",
  summary: "Deployment failed: health check timeout after 5 minutes",
  request_reflection_handoff: true
)

Pattern 4: Intelligent Retrieval

Use slice projection and hybrid retrieval to pull relevant context:

# Get focused context for a specific objective
slice_project(
  objective: "Fix the race condition in the checkout handler",
  projection_mode: "focused",
  sources: ["plan_state", "journal", "vector"],
  max_tokens: 4000
)

# Broad retrieval across all sources
hybrid_retrieve(
  query: "How did we handle webhook retries last time?",
  sources: ["journal", "vector", "temporal"],
  fusion_strategy: "memory_first",
  top_k: 10
)

Pattern 5: Reflection Handoff

When a plan needs deeper analysis or has failed:

reflection_handoff(
  plan_id: "billing-overhaul",
  summary: "Checkout flow has intermittent 500 errors under load. Suspect connection pool exhaustion.",
  tags: ["performance", "database", "connection-pool"],
  force: true
)

Reading Plan State

Always read existing plan state before resuming work:

plan_state_read(
  plan_id: "billing-overhaul",
  include_messages: true,
  include_artifacts: true
)

Best Practices

1. Use descriptive plan IDs: fix-auth-middleware not task-1
2. Checkpoint often: Before session end, before risky operations, after phase completion
3. Write meaningful summaries: Future sessions depend on these for context recovery
4. Tag consistently: Use domain/action tags matching your team conventions
5. Bridge sync on execution boundaries: Pre/post deployment, pre/post test runs
6. Request reflection on failures: reflection_handoff distills failure patterns into learnable memories