hashloom CLAUDE.md

hashloom CLAUDE.md is an instructions file for coding agents from davet47/hashloom. It costs 1,643 tokens per session, scanned A, original, Apache-2.0.

Project instructions for Hashloom, a verification cache and contract store exposed through MCP, a standard way for AI agents to call tools. They define repository rules, scope limits, branching, and the expected definition of done.

In plain words
What is it for?
Use them when modifying Hashloom, planning changes, creating branches, deciding what belongs in an issue, or checking whether a task meets the project's verification and token-efficiency requirements.
Why use it?
They keep coding work consistent with the project's workflow and prevent unrelated features or unsafe Git changes from being introduced.

Instructions file

Install

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.

agentmods
npx agentmods add instructions/davet47/hashloom/claude-md
Clone the repo
git clone --depth 1 https://github.com/davet47/hashloom

Wrote 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.

agentmods badge for hashloom CLAUDE.md

README.md
[![agentmods](https://agentmods.dev/badge/instructions/davet47/hashloom/claude-md.svg)](https://agentmods.dev/instructions/davet47/hashloom/claude-md)
Your own site
<a href="https://agentmods.dev/instructions/davet47/hashloom/claude-md"><img src="https://agentmods.dev/badge/instructions/davet47/hashloom/claude-md.svg" alt="Measured on agentmods" height="20"></a>
Per session 1,643 This file is loaded in full into every session.
When invoked 1,643 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.01643 $0.01643
Opus 5 $0.00822 $0.00822
Sonnet 5 $0.00329 $0.00329
Haiku 4.5 $0.00164 $0.00164

Measured 3d ago against content hash 533f15116231, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

hashloom 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 3d 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.

CLAUDE.md · 127 lines

How it starts

The opening of the file, as written. The whole thing — 127 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Working rules for hashloom

Hashloom is a hash-keyed verification cache + content-addressed contract store, exposed over MCP. Contracts are warp (durable), code is weft (regenerable). These rules override default behavior — follow them.

Git is the user's

Commit and push only when the user explicitly asks — never on your own, and write no Co-Authored-By trailer when you do. The user owns the PR/merge flow and brings local back to a clean, synced main before each new piece of work.

A fresh branch per change. The first action on any new feature/change is git switch -c <scoped-name> off main — never commit to main, and never reuse a previous feature branch for new work. One piece = one branch = one PR.

Never git worktrees — too much machinery for this project; the user dislikes them. Don't reach for worktree isolation.

Scope discipline

Anything not on the current milestone is an entry in ISSUES.md — file it there, don't write the code. The named failure mode is "scope creep toward Loom." Keep the surface minimal: 5 MCP tools, 5 CLI commands. The README documents "the entire surface"; if a change would add to it, stop and confirm. (The shared-cache backend python -m hashloom.cache_server is a deliberate operational process, not a 6th CLI command — the client surface stays 5/5.)

Hashloom develops on hashloom

The repo is itself a hashloom project: contracts/ holds contracts for the stable seams (the five api.py functions, the contract.py hashing trio, impl_hash, verification_key, HashloomError, the Store Protocol, the LanguageAdapter seam behind the per-extension adapters, and the cache_protocol wire spec — spec-only, no impl to verify). The workflow from docs/getting-started.md applies here:

  • The hashloom MCP server is project-configured in .mcp.json (uv run hashloom serve). When its tools are present, prefer get_contract packets and get_dependents over reading a contracted seam's source file — the ~300-token packet is the token-frugal path this project exists to prove; fall back to file reads only for uncontracted code.
  • Before changing a contracted seam, check the blast radius (the get_dependents MCP tool when connected — it is not a CLI command); after touching one, uv run hashloom verify --radius <name> must return ok: true — that is the inner-loop gate.
  • A new stable seam gets a contract before its implementation. If you (the agent) derived the contract rather than the user specifying it, mark it status: inferred; the user flips it to confirmed on review. hashloom status lists the review queue.
  • Do not contract churning interiors. The cache server's wire protocol is contracted (cache_protocol — it became a compatibility surface when the v0.5 hosted-store theme shipped), but the module interiors behind it — remote.py, cache_server.py, shared.py — stay uncontracted: their store-facing obligations are Store's, the wire's are cache_protocol's. Helpers (tokens.py, project.py) are weft. Pinning interiors is the failure mode the README warns about.
  • The hashloom gate layers on the DoD — it never replaces it. Full uv run pytest and the benchmark below remain the definition of done; hashloom's cached verify must not be the only thing vouching for hashloom.

Read the full file on GitHub · 127 lines

Changes

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.

  1. 3d ago Changed · +3 lines · +43 tokens per session 533f15116231
  2. 4d ago First seen · 124 lines · 1,600 tokens per session scan A 864a3b2dd824

Subscribe to this mod's changes

hashloom CLAUDE.md is an instructions file published in the GitHub repository davet47/hashloom (2 stars, last pushed 2d ago), licensed Apache-2.0. It adds 1,643 tokens to every session, about $0.0082 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.