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/davesheffer/hunch/claude-mdgit clone --depth 1 https://github.com/davesheffer/hunchWrote 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/davesheffer/hunch/claude-md)<a href="https://agentmods.dev/instructions/davesheffer/hunch/claude-md"><img src="https://agentmods.dev/badge/instructions/davesheffer/hunch/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 | $0.02899 | $0.02899 |
| Opus 5 | $0.01450 | $0.01450 |
| Sonnet 5 | $0.00580 | $0.00580 |
| Haiku 4.5 | $0.00290 | $0.00290 |
Grade A, and why
hunch 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 today.
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 — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
What this is
Hunch is an engineering-memory and architectural-conformance layer: a CLI + MCP server that builds a persistent, git-native reasoning graph (decisions, bugs, constraints, components), delivers the relevant evidence to coding assistants, and checks their changes deterministically. Published as @davesheffer/hunch (hunch binary). Pure TypeScript ESM, Node ≥22.13, no build step at dev time (run via tsx).
Hunch keeps memory true along two spokes: graph≠code (Architectural Conformance / intent-conformance — does the code still satisfy recorded intent?) and, as of v0.39.0, doc≠graph (decision-grounding — does the prose still match the live decision?). The doc≠graph spoke extends to markdown itself (src/core/docanchors.ts): a <!-- hunch:topic <topic> [dec_id] --> marker in AGENTS.md/CLAUDE.md/docs grounds the pre-edit hook with that topic's current decision, and a pinned marker whose decision gets superseded fires deterministic doc-anchor-stale drift (CI-gated in hunch drift, healed via hunch heal). A decision can carry an optional topic anchor (drift-detection key; defaults null, no schema bump, existing graphs load unchanged) with a current/history/rejected query contract. Read-time grounding surfaces a file's topic-anchored decisions on the pre-edit hook (doc-precedence framing: follow the graph, not a stale doc — including what each decision rejected), and the deterministic anchor-stale drift kind fires when a file is still anchored to a SUPERSEDED decision while a current one exists for its topic.
Commands
npm run dev -- <args> # run the CLI from source via tsx (e.g. npm run dev -- doctor)
npm run hunch -- <args> # alias for the above
npm run build # clean + tsc -> dist/ (the published artifact; bin = dist/cli/index.js)
npm run typecheck # tsc --noEmit
npm test # tsx --test over test/*.test.ts
tsx --test test/check.test.ts # run a single test file
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.
- today Changed · -11 tokens per session 335ae056b2ad
- 4d ago First seen · 90 lines · 2,910 tokens per session scan A 70011247b571
hunch CLAUDE.md is an instructions file published in the GitHub repository davesheffer/hunch (9 stars, last pushed today), licensed Apache-2.0. It adds 2,899 tokens to every session, about $0.0145 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 2026-08-31.
Other instructions, from other repositories
opensrc AGENTS.md
Instructions for vercel-labs/opensrc, covering agents.md, monorepo structure, cli (packages/opensrc), docs (apps/docs) and releasing.
memorix CLAUDE.md
Claude Code instructions for Tibu142/memorix, covering memorix — agent instructions for claude code, when to search memory (memorixsearch), when to store memory (memorixstore), when to check retention (memorixretention) and best practices.
plan-forge context-fuel.instructions.md
Context management for AI agents — recognize context degradation, prioritize instructions, bridge sessions with memory.
lerim AGENTS.md
AGENTS.md instructions for nablo-io/lerim, covering lerim, summary, current architecture, trace format and parsing and how to query existing context.
claude-context AGENTS.md
AGENTS.md instructions for zilliztech/claude-context, a project described as: Code search MCP for Claude Code. Make entire codebase the context for any coding agent.
capsule AGENTS.md
Instructions for dawitlabs/capsule, a project described as: Markdown context packs that help coding agents reduce repeated repository-discovery tokens.