Borrowing it
Nothing to install: this file belongs to Supa-Media/context. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/Supa-Media/context/main/CLAUDE.mdgit clone --depth 1 https://github.com/Supa-Media/contextWrote 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/supa-media/context/claude-md)<a href="https://agentmods.dev/instructions/supa-media/context/claude-md"><img src="https://agentmods.dev/badge/instructions/supa-media/context/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/supa-media/context/claude-md"><img src="https://agentmods.dev/badge/instructions/supa-media/context/claude-md.svg" alt="Reviewed on agentmods" width="80" 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 | $0.04516 | $0.04516 |
| Opus 5.5 | $0.01806 | $0.01806 |
| Sonnet 5.5 | $0.00903 | $0.00903 |
| Haiku 4.5 | $0.00452 | $0.00452 |
Grade A, and why
context 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 yesterday.
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 — 310 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Instructions — Context
Free your context. Share your context.
Context is one MCP endpoint a person adds to every AI client, backed by a
markdown bucket — one they own, or one we run for them on the paid plan and
they can take away at any time. Read README.md first for the product shape.
This file is the part you read every session. The reasoning behind the rules —
argued through once, not to be re-litigated — lives in
docs/decisions/, split by area. Read the
file for the area you are touching, and write new durable decisions there.
Non-negotiables
These are the product, not implementation details. If a task seems to require breaking one, stop and say so rather than working around it.
- The customer owns the content, and can always leave with it. Note bodies and attachments live in a bucket dedicated to one workspace. Names, folders and privacy live in the control plane, on each workspace's tree rows (decided by the owner, 2026-10-15, replacing the bucket-as-truth rule; notes-by-id), alongside accounts, bindings, grants, audit, and each note's rebuildable facts (links, front matter, first heading and paragraph; convex-facts). The control plane never holds a note's body. Leaving is Download as folders: a complete, decrypted copy with real names, free on every plan, still available after cancelling, never gated, degraded or behind a paywall — gate it and this is a different product. Moving to storage of their own is free too. Cancelling puts a context back on the free plan, still writable and exportable; it never deletes (decided by the owner, 2026-09-29). Credentials never live in Markdown, in the bucket, in logs, in URLs, or on a device: encrypted at rest, decrypted only in the gateway at request time. The one exception is the texting assistant's autofill vault, sealed in the person's own bucket with the key held outside it and never shown to a model (decided by the owner, 2026-10-06; texting-assistant). See storage-and-credentials.
- Tenancy is bucket-level, never prefix-level. Do not namespace keys inside
a bucket — no
tenants/<id>/, noworkspaces/<slug>/. A note's bytes live atnotes/<noteId>.md(other files atfiles/<noteId>.<ext>), and its name and folder are rows in the control plane; Context may move or rewrite any key in a bucket it is connected to, a customer's own included (decided by the owner, 2026-10-15). One workspace maps to one bucket (optionally plus a fixed root prefix the customer chose, applied at the adapter boundary). Managed buckets are one per workspace, named from the immutable workspace id, in a Cloudflare account that holds customer data — those buckets, the per-context search databases, and a live note's in-flight keystrokes while people are typing it together — and nothing of ours. That third one was added deliberately (decided by the owner, 2026-09-20): two people editing one note need a shared place for characters that are seconds old. It is bounded — one room per note, append-only, deleted when the last person leaves, replayed only to a socket that passedcanSeeat join, never the only copy of anything, and the flush to the bucket is continuous. See storage-and-credentials. A store's per-account resource ceiling is a constraint on the product: R2 allows a million buckets, and anything low forces prefix tenancy. - Note bodies stay plain Markdown. A Context-managed bucket may hold each
object's body sealed at rest (key outside the bucket), because every read
through the product and every export is plain; a customer's own bucket
never is (decided by the owner, 2026-09-29;
managed-encryption).
Collaboration history under
.context/collaboration/is customer-owned essential data, needed to merge offline edits; accepted edits may briefly precede their Markdown rendering, and a completed save or export includes them (collaboration). Search indexes, caches and embeddings are disposable derivatives, rebuildable from the notes and the rows. The bucket layout isnotes/,files/and.context/; there is noprivacy.md, and nothing outside the app writing into the bucket is picked up (decided by the owner, 2026-10-15). No compatibility with syncing a bucket to Obsidian is kept; Obsidian plugins running inside Context are. - One person or workspace is one security boundary. Every workspace has its
own identity, storage binding, privacy manifest, audit trail and connector
grants. Never extend the legacy shared-token model (
PRIVATE_TOKEN/TEAM_TOKEN/PUBLIC_TOKEN) to multiple customers — it is single-tenant by construction and exists only to keep the original single-tenant deployment running. MCP access uses OAuth with per-client revocable grants; token-in-URL is a compatibility fallback and never the security boundary. Prove isolation with tests: one tenant must not enumerate, read, or infer the existence of another. teamnever means public. A row's access isprivate(owners) orteam(members), plus any mix of groups and named people, inside the workspace or not, each with view or edit (decided by the owner, 2026-10-15). Every grantee is a person with an account or an email awaiting one. No setting publishes a context, a folder, or a visibility class to the internet, and nothing here is indexed. A link an owner mints and can revoke is the single exception, and it is always a share row, never a value of a row's access. A link covers one note, or one folder and the subtree beneath it — and a folder link narrows: every path under it is still re-derived through the live rows atteamscope with no granted names, so a note held back by name, a private subfolder and a note pointed at a group are all absent through it. It publishes what the folder already published to the workspace and never more, which is why this is a wider locator and not a wider tier. The whole context is never the subject of one. See privacy-and-sharing. A workspace's website is this exception forwebsite/and the folders its pages name withfolder:: turning it on is the owner minting that locator, and every page, referenced notes included, resolves the same way, so frontmatter can narrow a page but never publish one the rows hold back, and no page may name the whole context (websites). The gate that held an AI client's widening for the person's yes was removed by the owner on 2026-10-17; since 2026-10-18 a small judge model lets a widening run only when the person's own words ask for it, with nothing to tap. What still waits for a yes, checked by the gateway and never by the model, is a step nobody can take back: an email, a calendar change, a handle claim, a vault key's first use (agent-egress).
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.
- yesterday Changed · -1 lines · -22 tokens per session 0e14338afcf8
- 3d ago Changed · +2 lines · +40 tokens per session 16b63ec11fa0
- 4d ago Changed · -15 lines · -232 tokens per session 5e96130d7ea2
- 5d ago Changed · +13 lines · +265 tokens per session 66eb8b96f2ec
- 7d ago Changed · +8 lines · +146 tokens per session ed489b21b422
- 9d ago Changed · +2 lines · +38 tokens per session 5fe15e583f96
- 10d ago Changed · +6 lines · +116 tokens per session a7802a6618b3
- 13d ago Changed · +5 lines · +98 tokens per session b0de484beb77
- 20d ago Changed · +6 lines · +106 tokens per session cb6cb50517f6
- 23d ago First seen · 284 lines · 3,961 tokens per session scan A 0e8e20c0075f
context CLAUDE.md is an instructions file published in the GitHub repository Supa-Media/context (3 stars, last pushed today), licensed MIT. It adds 4,516 tokens to every session, about $0.0181 per session on Opus 5.5. 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 2026-08-31.
Other instructions, from other repositories
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, cache eligibility and rollout and applying the notice (cutover).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
vscode copilot-instructions.md
Copilot instructions for microsoft/vscode, covering vs code copilot instructions, project overview, root folders, core architecture (src/ folder) and built-in extensions (extensions/ folder).
deepseek-harness AGENTS.md
AGENTS.md instructions for deepseek-ai/deepseek-harness, covering agents.md, pre-stable apis and released session data, repository layout, commands and host sandbox failures.