Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/nikiforovall/scratchpad/claude-mdgit clone --depth 1 https://github.com/NikiforovAll/scratchpadWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/nikiforovall/scratchpad/claude-md)<a href="https://agentmods.dev/instructions/nikiforovall/scratchpad/claude-md"><img src="https://agentmods.dev/badge/instructions/nikiforovall/scratchpad/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.01374 | $0.01374 |
| Opus 5 | $0.00687 | $0.00687 |
| Sonnet 5 | $0.00275 | $0.00275 |
| Haiku 4.5 | $0.00137 | $0.00137 |
Grade A, and why
scratchpad CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.
What this is
scratch is a CLI-first tool that organizes agent knowledge into scratchpads: a folder + a scratchpad.json manifest, with a read-only visual viewer. Runs on the Bun runtime.
Core invariant (do not violate): the CLI never authors, copies, or moves file content. It only creates pad dirs, writes manifests, and tracks metadata. The user/agent writes files with their normal tools; scratch add just registers them. A pad is just a folder containing scratchpad.json — the folder path is its identity. There is no central store/index.
Two deliberate exceptions (don't "fix" them as violations), both in src/ui/launch.ts:
- Clicking a rendered GFM task checkbox in the viewer flips that single
[ ]/[x]marker in the source file (persistFileCheckbox). Line-addressed; verifies the line still IS a task marker before writing. - Saving from the viewer's in-place Excalidraw editor writes the edited scene JSON back to its
.excalidrawfile (persistExcalidrawScene). The user drew the content; the CLI only persists it — scoped to manifest-registered.excalidrawtargets and validated withparseScenefirst.
Commands
bun test # run the suite (bun's built-in test runner)
bun test test/cli.test.ts # single file
bun test --test-name-pattern "slugify" # single test by name
bun run scratch -- <args> # run the CLI in dev (e.g. bun run scratch -- ls)
bun run build # compile standalone → dist/scratch(.exe) + stage native host
bun link # expose `scratch` globally from source
bun run docs:dev # vitepress docs (docs/)
No separate lint/typecheck step is wired at the root; tsc runs via pi/ only (bun --cwd pi run typecheck). Tests are the gate (prepublishOnly: bun test).
Monorepo layout
Two independently-published npm packages plus a Claude Code plugin, all in one repo:
- root (
@nikiforovall/scratchpad) — the CLI. Entrysrc/cli.ts, bin namescratch. pi/(@nikiforovall/pi-scratchpad) — pi coding-agent package: skills +/scratch ui|export|stopcommands. Drives the samescratchCLI (does not reimplement it).plugins/scratchpad/— Claude Code plugin (skills);.claude-plugin/makes this repo a plugin marketplace.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 56 lines · 1,374 tokens per session scan A 9bd580b198ad
scratchpad CLAUDE.md is an instructions file published in the GitHub repository NikiforovAll/scratchpad (11 stars, last pushed yesterday), licensed MIT. It adds 1,374 tokens to every session, about $0.0069 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
eve AGENTS.md
AGENTS.md instructions for vercel/eve, covering agents.md, about eve, repository layout, git workflow and commands.
Backlog.md AGENTS.md
AGENTS.md instructions for MrLesk/Backlog.md, covering read this first, simplicity-first implementation rules, ui descriptions, commands and development.
reflect-open AGENTS.md
AGENTS.md instructions for team-reflect/reflect-open, covering purpose, what is reflect, product principles, agent workflow and development workflow.
reflect-open CLAUDE.md
Claude Code instructions for team-reflect/reflect-open: See AGENTS.md for the full project overview, tech stack, database tables, code conventions, and development cycle.
skill-based-architecture AGENTS.md
Instructions for WoJiSama/skill-based-architecture, covering agents.md, quick routing (survives context truncation), auto-triggers and red flags — stop.
kipi-system AGENTS.md
AGENTS.md instructions for assafkip/kipi-system, covering q entrepreneur os, project structure, conventions, commands and build and test.