From devflow
Persistent learning memory — curated runbook of mistakes, corrections, and patterns that work. Always active, every session.
npx claudepluginhub nexuz-sys/devflow --plugin devflowThis skill uses the workspace's default tool permissions.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Searches prompts.chat for AI prompt templates by keyword or category, retrieves by ID with variable handling, and improves prompts via AI. Use for discovering or enhancing prompts.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Persistent learning memory for agents. Based on blader/napkin v6.0.0.
The napkin is a continuously curated runbook, NOT a chronological log. It's a live knowledge base for execution speed and reliability.
No trigger required. Every session, unconditionally:
.context/napkin.md before doing anythingFile: .context/napkin.md (one per project, inside dotcontext directory).
.context/ to exist (Full or Lite mode only).context/ does, create it using the template belowWhen creating a new napkin:
# Napkin Runbook
## Curation Rules
- Re-prioritize on every read.
- Keep recurring, high-value notes only.
- Max 15 items per category, max 7 per agent section.
- Each item includes date + "Do instead".
## Execution & Validation
(empty)
## Shell & Command Reliability
(empty)
## Domain Behavior Guardrails
(empty)
## User Directives
(empty)
## Agent-Specific Notes
<!-- Sections appear on demand: ### agent-name -->
Every entry MUST follow this format:
1. **[YYYY-MM-DD] Short rule**
Do instead: concrete repeatable action.
[YYYY-MM-DD]Do instead: line with concrete actionFour fixed categories (max 15 items each, sorted by importance descending):
| Category | What goes here |
|---|---|
| Execution & Validation | Test failures, build gotchas, CI surprises, TDD patterns |
| Shell & Command Reliability | CLI quirks, command flags, PATH issues, tool-specific behavior |
| Domain Behavior Guardrails | Business logic traps, API contracts, data format surprises |
| User Directives | Explicit user preferences that affect repeated behavior |
Sub-sections under ## Agent-Specific Notes for domain-specific learnings. Max 7 items per agent. Sections appear on demand — no empty placeholders.
## Agent-Specific Notes
### security-auditor
1. **[2026-04-05] Rule title**
Do instead: action.
### database-specialist
1. **[2026-04-05] Rule title**
Do instead: action.
On every read (session start, post-compact):
Napkin and Claude Code's auto-memory (MEMORY.md) are independent:
| Mode | Available? | How |
|---|---|---|
| Full | Yes | SessionStart hook injects; dotcontext MCP can read .context/napkin.md |
| Lite | Yes | SessionStart hook injects; direct file read |
| Minimal | No | No .context/ directory — skill inactive |