
from orchestkit196
Automated nightly memory consolidation: prunes stale memory files, merges duplicates, resolves contradictions, and rebuilds MEMORY.md indexes with deterministic
Dream performs deterministic memory maintenance for agent memory directories. It discovers MEMORY.md files, scans associated memory entries, detects stale references and missing symbols, merges duplicate entries, resolves contradictory decisions (keeping the newer), and rebuilds tidy MEMORY.md indexes. Changes can be previewed in dry-run mode before any deletions.
Use Dream when agent or project memory files have accumulated over many sessions and you need to clean up stale or duplicate entries, rebuild the index, or resolve conflicting decisions. Good for periodic housekeeping (nightly runs) and before audits of long-running agents.
Best for agents using Claude Code-style tooling and filesystem access (skills that can Run Read/Write/Glob/Grep/Bash).
Dream is a well-architected memory consolidation skill for Claude Code that deterministically prunes stale memory entries, merges duplicates, resolves contradictions, and rebuilds MEMORY.md indexes. No bundled scripts — all logic is expressed as detailed pseudocode within the SKILL.md itself. The skill has strong safety features including --dry-run mode, consult-gate protection against pruning recently-used memories, and clear error handling. It follows the skill spec excellently with comprehensive frontmatter and progressive disclosure.
No scripts to execute — static analysis only. The consult-gate feature (checking memory-consult logs before pruning) is a thoughtful safety addition. Steps 7 and 8 (plugin housekeeping, stale project state) are Claude Code specific and well-scoped with version guards.
Cross-Project Analytics
Query local, privacy-safe cross-project analytics to report on agent, skill, hook, and team performance; replay sessions and estimate token costs.
Task Dependency Patterns
Patterns and workflows for creating, sequencing, and managing task dependencies using TaskCreate/TaskUpdate/TaskGet/TaskList tools to coordinate multi-step impl
Review PR
Deep pull-request code review using parallel specialized agents (quality, security, tests, architecture, performance) and synthesized review reports for approve
Write PRD (Product Requirements Document)
Generate an 8-section Product Requirements Document with user stories, INVEST-formatted acceptance criteria, and go/no-go criteria to turn product vision into e
Telemetry Inspect
Run a read-only health check of OrchestKit's telemetry pipeline: writer activity, file sizes, schema lock coverage, orphan files, and growth trends.
Design Ship
Turn a Claude Design handoff URL or bundle into a reviewable GitHub PR with generated Storybook stories, Playwright tests, and before/after verification.
Scope-Appropriate Architecture
Prevents over-engineering by right-sizing architectural patterns to the project's actual scope using a 6-tier classification system.
Responsive Patterns
Practical patterns for responsive UI: container queries, fluid typography, intrinsic layouts, and foldable-device support for React components.
Verify
Run comprehensive verification: parallel agents for tests, security, coverage, grading and improvement suggestions; produces evidence-backed verification report