Use this skill when context compression is imminent, when resuming a session, or when preserving critical decisions across long tasks. Provides a durable notepad that survives compaction and lets any agent pick up exactly where the previous context left off.
npx skills add https://github.com/vibeeval/vibecosystem --skill notepad-system
Structured storage for decisions, task state, and critical context that must survive
context window compression. Extends pre-compact-state.md from manual discipline to
a concrete file-based protocol.
When context compresses, the following is lost:
Without this, the recovered context either restarts from scratch or makes assumptions
that contradict earlier decisions.
| Type | Persistence | When to Use |
|------|-------------|-------------|
| Priority | Permanent until explicitly deleted | Architectural decisions, user confirmed preferences, hard constraints |
| Working | Session-scoped, cleared at session end | Current task state, WIP progress, in-flight findings |
| Scratch | Cleared after use (one-time calc) | Quick calculations, temp values used once |
~/.claude/projects/<project-hash>/notepad.md
The project hash is derived from the working directory path.
If the directory does not exist, create it before writing.
# Notepad
## Priority Notes
### [2026-03-29 14:22] Decision: Auth approach
JWT with refresh tokens. User confirmed.
Reason: Stateless, scalable, mobile-friendly.
Rejected: Session-based (server state, scaling issues).
### [2026-03-29 15:10] Constraint: Node version
Must stay on Node 18. Server has LTS pinned, no upgrade path in scope.
## Working Notes
### [2026-03-29 14:45] Task: API refactor — Phase 2
Completed: /users, /products, /categories endpoints
Remaining: /orders, /payments endpoints
Blocked on: Stripe webhook secret (user to provide)
Next action: Start /orders, skip payment until secret arrives
## Scratch
### [2026-03-29 15:30] Token estimate
Average response: 256 tokens. 4x safety margin = ~1024 tokens per request budget.
USED — safe to delete.
When context compression is imminent (context window at ~80% or system warning appears):
Step 1 — Dump active task state to Working Notes
What task is in progress?
What has been completed so far?
What remains?
Are there any blockers?
What is the immediate next action when context resumes?
Step 2 — Promote any pending decisions to Priority Notes
Any architectural choice made this session → Priority Note
Any user-confirmed preference → Priority Note
Any external constraint discovered → Priority Note
Step 3 — Record modified files
List every file touched this session.
Note what changed in each (one line per file is enough).
Note whether changes are committed or uncommitted.
Step 4 — Save pending decisions with their options
If a decision was not yet made, record the question and the options.
This prevents re-deriving the same options after compression.
Do this BEFORE continuing work. State preservation takes priority.
When a new or compressed context begins:
Step 1 — Read notepad.md
Path: ~/.claude/projects/<project-hash>/notepad.md
Step 2 — Inject Priority Notes into active context
These are always relevant. Treat them as if the user stated them at session start.
Step 3 — Summarize Working Notes for the user
"Resuming: API refactor, Phase 2.
Completed: /users, /products, /categories.
Remaining: /orders, /payments.
Blocked: Stripe webhook secret needed.
Next: starting /orders."
Step 4 — Clear Scratch notes
They were single-use. Remove them now.
Step 5 — Resume from last known state
Do not ask the user to re-explain context that is in Working Notes.
notepad delete priority <title>file.ts:line insteadRead the notepad:
cat ~/.claude/projects/<hash>/notepad.md
Clear working notes (session end):
# Remove the Working Notes section content, keep the headers
Clear scratch notes (after use):
# Remove individual scratch entries after they have been applied
pre-compact-state.md — This skill is the concrete implementation of that rule.
The rule defines what to save; this skill defines where, how, and in what format.
compass agent — The compass agent performs context recovery. It should read
notepad.md as its first action and use the contents to construct the "Nerede kalmistik?" summary.
Any agent — Any agent starting work on a project should check notepad.md for
Priority Notes that constrain their approach. An architect should not propose a
solution that contradicts a Priority Note.
Working on a large refactor across 12 files. Context at 80%.
Pre-compact dump:
### [2026-03-29 16:00] Task: Auth system refactor — 7/12 files done
Completed: user.service.ts, auth.controller.ts, token.service.ts,
session.middleware.ts, logout.handler.ts, login.handler.ts, refresh.handler.ts
Remaining: password.service.ts, oauth.handler.ts, 2fa.handler.ts,
admin.auth.ts, tests/auth.test.ts
Next: start password.service.ts — method signature unchanged, internals only
After compression, any agent reads this and knows exactly which 5 files remain.
User asked about caching strategy but has not confirmed yet.
Scratch note:
### [2026-03-29 16:15] Pending: Caching strategy
Options: (a) Redis — fast, requires infra; (b) in-memory — simple, no persistence
Awaiting user confirmation on whether Redis is available in prod.
This prevents re-deriving the same question after compression.
User mentioned early in a long session that they cannot use PostgreSQL — MySQL only.
Priority note created immediately:
### [2026-03-29 10:05] Constraint: Database — MySQL only
No PostgreSQL. Server is MySQL 8.0. All queries and schema must be MySQL-compatible.
Reason: Hosting provider limitation.
Every subsequent agent reads this and never proposes PostgreSQL.
Complete development kit for Microsoft 365 Copilot declarative agents with three comprehensive workflows (basic, advanced, validation), TypeSpec support, and Microsoft 365 Agents Toolkit integration
Format and structurally validate local treatment-plan documentation after clinical decisions have already been supplied and verified by authorized licensed professionals. Use for source traceability, clinician-authored intervention records, goals and checkpoints, shared-decision records, reconciliation handoffs, and release gates—not for clinical decision-making.
> provider/change budget/修改卖家/修改预算/draft/草稿/我的任务/my tasks/what am I working on/关闭/取消任务/决策列表/decision list/指定服务商/browse (sender.role = COUNTERPARTY, not you); (3) literal "Read the okx-ai skill" (or legacy "Read the okx-agent-task skill") in the envelope.
Automate payer review of prior authorization (PA) requests. This skill should be used when users say "Review this PA request", "Process prior authorization for [procedure]", "Assess medical necessity", "Generate PA decision", or when processing clinical documentation for coverage policy validation and authorization decisions.
Expert in designing and building autonomous AI agents. Masters tool use, memory systems, planning strategies, and multi-agent orchestration.
Autonomous agents are AI systems that can independently decompose goals, plan actions, execute tools, and self-correct without constant human guidance. The challenge isn't making them capable - it's making them reliable. Every extra decision multiplies failure probability.
Orchestrates design workflows by routing work through brainstorming, multi-agent review, and execution readiness in the correct order.
Structured persuasion for tech leads, PMs, and founders—not activity logs. Five scenarios (kickoff, status update, wrap-up, investor pitch, solution selling) on one 5-part framework (Hook→Context→Proposal→Evidence→Ask). AI prompts for missing materials and audience context; pre-submit checklist. Claude Code plugin; Cursor, Codex, and chat via prompts.
Take vibeeval/notepad-system from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.