vouch AGENTS.md

A set of instructions for coding agents working on vouch, a Git-based knowledge base for language-model agents. It explains what vouch is, where to start reading, how to install it, and its approval and trust rules.

In plain words
What is it for?
Use it to orient agents in the vouch repository, follow its installation and reading order, and understand its MCP or JSONL connection, configuration, and safety boundaries.
Why use it?
It tells an agent which guidance applies when changing vouch itself versus using it in another project. It also clarifies that proposed knowledge-base writes require human approval and that certain operations are logged or restricted.

Instructions file for CodexOpenCode

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/vouchdev/vouch/agents-md
Clone the repo
git clone --depth 1 https://github.com/vouchdev/vouch

Made for: Codex, OpenCode.

Per session 1,886 This file is loaded in full into every session.
When invoked 1,886 The same file — it is already loaded in full.
Security scan C 2 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 $0.01886 $0.01886
Opus 5 $0.00943 $0.00943
Sonnet 5 $0.00377 $0.00377
Haiku 4.5 $0.00189 $0.00189

Measured 2d ago against content hash 86f14ed68b77, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

vouch AGENTS.md scanned grade C with 2 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 2d 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.

Downloads and executes remote codehighSupply chain

curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.

curl -fsSL https://raw.githubusercontent.com/vouchdev/vouch/main/install.sh | sh

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

curl -fsSL https://raw.githubusercontent.com/vouchdev/vouch/main/install.sh | sh
AGENTS.md · 182 lines

How it starts

The opening of the file, as written. The whole thing — 182 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Agents working with vouch

Entry document for any AI coding agent reading this repository — Cursor, Codex, OpenClaw, Aider, Continue, JetBrains AI, an LLM fetching the raw URL, anything that isn't Claude Code. (Claude Code reads CLAUDE.md instead; the two files complement each other.)

If your task is to work on the vouch codebase itself, read this file and then CLAUDE.md. If your task is to use vouch from inside another project, read README.md for install + concepts and docs/getting-started.md for the agent-side loop.

If you're an OpenClaw plugin loader, the plugin manifest is at the repo root: openclaw.plugin.json. It declares vouch's MCP wiring, the four slash commands, the trust boundary (write tools review-gated, lifecycle ops audit-logged, remote-caller filesystem confined), and the config schema (kb_path, agent, transport — no secrets). No additional wiring is required to surface vouch's kb.* surface inside an OpenClaw deployment.

What vouch is, in one paragraph

Vouch is a git-native, review-gated knowledge base for LLM agents. Agents propose writes via an MCP server (or a JSONL pipe); a human approves each proposal with vouch approve. Approved artifacts land as YAML claims and markdown pages under .vouch/ — plain files that diff cleanly in PRs and travel as a tarball bundle. The CLI is vouch; the PyPI distribution is vouch-kb; supported Python versions are 3.11, 3.12, 3.13.

Install (1 minute)

curl -fsSL https://raw.githubusercontent.com/vouchdev/vouch/main/install.sh | sh

Or, deterministically from a clone (the path you want when contributing):

git clone https://github.com/vouchdev/vouch.git
cd vouch
python3 -m venv .venv && . .venv/bin/activate
pip install -e '.[dev]'

Read in this order

  1. This file — entry, install, trust boundary, common tasks.
  2. CLAUDE.md — orientation for working on the repo: architecture, conventions, ship rules, voice. Read even if you aren't Claude Code; the conventions are universal.
  3. README.md — the user-facing pitch, install, quick start, full CLI surface, MCP / JSONL method list.
  4. SPEC.md — the canonical protocol description: .vouch/ layout, object model, kb.* method shapes, review-gate state machine. Authoritative when in doubt.
  5. ROADMAP.md — what's planned for 0.2 and 0.3. Don't propose features that are already scoped here.

Read the full file on GitHub · 182 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. 2d ago First seen · 182 lines · 1,886 tokens per session scan C 86f14ed68b77

Subscribe to this mod's changes

vouch AGENTS.md is an instructions file published in the GitHub repository vouchdev/vouch (92 stars, last pushed 9d ago), licensed MIT. It adds 1,886 tokens to every session, about $0.0094 per session on Opus 5. A static security scan graded it C with 2 findings (downloads and executes remote code, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.