sentinel CLAUDE.md

A set of instructions for working on the Sentinel repository, which contains a Go MCP server and a Python test harness. MCP is a standard way for software tools to communicate with AI clients.

In plain words
What is it for?
Use it when changing or testing the Sentinel broker or its conformance harness, especially work governed by the repository’s MCP specification.
Why use it?
It gives coding agents the repository’s rules and constraints, including how stateless behaviour, retries, and deterministic results must work.

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/patsypppe/sentinel/claude-md
Clone the repo
git clone --depth 1 https://github.com/patsypppe/sentinel
Per session 925 This file is loaded in full into every session.
When invoked 925 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 $0.00925 $0.00925
Opus 5 $0.00463 $0.00463
Sonnet 5 $0.00185 $0.00185
Haiku 4.5 $0.00093 $0.00093

Measured yesterday against content hash 39cc4702a610, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

sentinel 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.

CLAUDE.md · 70 lines

How it starts

The opening of the file, as written. The whole thing — 70 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Sentinel — working instructions

Two products in one repository:

  • broker/ — a Go MCP server built natively on the stateless MCP 2026-07-28 specification.
  • harness/sentinel, a Python conformance harness that grades any MCP server.

The spec is docs/HANDOFF.md (SN-HND-001). Where this repository and the MCP specification disagree, the spec wins.

The four rules

  1. Stateless. Handles are data, never credentials. No session, nothing keyed by connection. Cross-call state is a server-minted handle passed as an ordinary tool argument. Possession of a handle is not authentication — every resolution re-verifies principal and tenant.
  2. No server-initiated requests. MRTR only, and it must be idempotent. resultType: "input_required" + inputRequests; the client retries with inputResponses. Correlation is the sealed requestState, never the JSON-RPC id. A duplicate retry returns the recorded result and performs zero additional side effects.
  3. Deterministic where the spec asks for determinism. 100 tools/list calls → one SHA-256. Every list/read result carries ttlMs and cacheScope. Mcp-Method required on every POST; Mcp-Name on tools/call, resources/read, prompts/get only.
  4. Every invocation is audited; no token is trusted that was not issued for this server. Validate the audience exactly. Never forward an inbound token downstream. If the audit write fails, the invocation fails.

Never cut

resultType + serverInfo on every result · handle binding · MRTR idempotency · the audit row · the non-conformant fixture server.

Single-source files (audit the security posture by reading three files)

File Sole authority for
broker/internal/envelope/errors.go every error code
broker/internal/handles/resolve.go making a handle usable
broker/internal/authz/audience.go accepting a token

Read the full file on GitHub · 70 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. yesterday First seen · 70 lines · 925 tokens per session scan A 39cc4702a610

Subscribe to this mod's changes

sentinel CLAUDE.md is an instructions file published in the GitHub repository patsypppe/sentinel (0 stars, last pushed 6d ago), licensed MIT. It adds 925 tokens to every session, about $0.0046 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.