Borrowing it
Nothing to install: this file belongs to DanFashauer/signalgrid-mcp. 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/DanFashauer/signalgrid-mcp/main/CLAUDE.mdgit clone --depth 1 https://github.com/DanFashauer/signalgrid-mcpWrote 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/danfashauer/signalgrid-mcp/claude-md)<a href="https://agentmods.dev/instructions/danfashauer/signalgrid-mcp/claude-md"><img src="https://agentmods.dev/badge/instructions/danfashauer/signalgrid-mcp/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/danfashauer/signalgrid-mcp/claude-md"><img src="https://agentmods.dev/badge/instructions/danfashauer/signalgrid-mcp/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 | $0.01687 | $0.01687 |
| Opus 5.5 | $0.00675 | $0.00675 |
| Sonnet 5.5 | $0.00337 | $0.00337 |
| Haiku 4.5 | $0.00169 | $0.00169 |
Grade A, and why
signalgrid-mcp 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 18d 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.
How it starts
The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — signalgrid-mcp
Guidance for owner-authorized agents and humans working in this repository. Read it before your first change. These rules override default behaviour.
What this is
An MCP server that exposes macOS-native device trust signals — the facts
about a Mac that no Linux container or cloud runner can gather. It is one
source feeding SignalGrid's decision core; the core and the canonical posture
contract live in the SignalGrid-Review-Hub repository. README.md must list
every registered tool; RUNBOOK.md says how it is run and verified.
Role: read the device and report honestly. Almost every tool only reports
posture. One piece here does decide: src/signalgrid_mcp/tools/verdict.py folds
a posture report into a single-device verdict (two or more criticals deny, one
restricts, unmanaged / auto-update off / any unknown steps up, else allow). It
ships as the tool signalgrid_trust_verdict, the resource signalgrid://verdict
and the posture report's include_verdict. Its own _note says it is a local,
single-device verdict that the SignalGrid fabric fuses with identity, custody and
other signals for the final decision. The golden rules below bind it.
Both SignalGrid lanes (Mac and cloud) have opened pull requests here. Work arrives as an explicitly selected pull request with its reason in the body.
Golden rules — do not break these
- Strictly read-only. No tool mutates the device, ever: no writes to files, settings, profiles, users, services or the network. Every tool carries the read-only annotation; a tool without it does not ship. A new tool that needs a write is a different product and needs a decision record in the Review-Hub first.
- Unknown is unknown, and never loosens an answer. A value this server
could not determine is reported as null/unknown, never as "off", "disabled"
or "safe". Downstream, unknown raises the bar; a collector that guesses
lowers it, which is the one failure SignalGrid refuses. This binds
verdict.pytoo:compute_verdictstays pure and deterministic (no clock, randomness or I/O), and a control it cannot read raises the answer to at leaststep_up, neverallow. Some probes legitimately read unknown without elevation or Full Disk Access (profiles list,systemsetup,tmutil latestbackup; README "Permissions & elevation", RUNBOOK step 3 item 2), by design. An unknown on a real Mac is a discrepancy to chase, neither a fact nor automatically a bug: if the raw command shows a value, the parser is wrong. Never "fix" an unknown by running elevated or by inferring a value. - The contract is owned elsewhere. The canonical posture contract is
lib/integrations/src/integrations/macos-posture/contract/posture-report.contract.jsonin the Review-Hub.tests/test_posture_contract.pyreads it throughSIGNALGRID_CONTRACT_PATHonly when that variable is set, which the Review-Hub'spnpm run verify:alldoes. Without it (this repository's CI) the test checks a copy of the core shape built into the test file. Keep that copy equal to the contract, never ahead of it: a shape change is a Review-Hub change first, proven withverify:all, then this repository follows. - Truthful on CI. Ubuntu CI proves the code, not the signals.
verify.shinstalls, runspytest -q(on a Mac that includes the live collector tests), thentools/inspect_stdio.py, which rejects crashes, missing tools, MCP tool errors and missing read-only markers. Honest unknown results can still pass protocol health. Itsvalue/unknowncolumn uses text-marker heuristics and is not proof that a signal is live. Thetest-macosCI job runspytest -qon GitHub's macOS runner. Live assertions are intests/test_live_collectors.pyand Darwin-only checks intests/test_smoke.py; these do not prove every collector reads a value. A green ubuntu run is never evidence that a collector works. - Nothing ships by assertion.
README.mdmust list and describe every registered tool (git grep 'name="signalgrid_' src/is the list;EXPECTED_TOOLSintests/test_smoke.pymust match it).CHANGELOG.mdis meant to describe what shipped, so do not use it as the tool list. A capability is claimed only after successful, signal-specific returned data or a corresponding live-collector assertion establishes it on a real Mac. An inspector label, registration or green ubuntu run is not that proof.
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.
- 18d ago First seen · 116 lines · 1,687 tokens per session scan A 53714bbf11f6
signalgrid-mcp CLAUDE.md is an instructions file published in the GitHub repository DanFashauer/signalgrid-mcp (0 stars, last pushed 19d ago), licensed MIT. It adds 1,687 tokens to every session, about $0.0067 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 2026-10-30.
signalgrid-mcp CLAUDE.md compared with instructions from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| signalgrid-mcp CLAUDE.mdDanFashauer/signalgrid-mcp | Instructions file | 0 | 19d ago | MIT | A | 1,687 | Claude Code |
| vscode buildNext.instructions.mdmicrosoft/vscode | Instructions file | 196k | today | MIT | A | 9,719 | GitHub Copilot |
| next.js AGENTS.mdvercel/next.js | Instructions file | 144k | today | MIT | A | 8,130 | Codex, OpenCode |
| vscode oss-third-party-notices.instructions.mdmicrosoft/vscode | Instructions file | 196k | today | MIT | A | 5,685 | GitHub Copilot |
| langchain AGENTS.mdlangchain-ai/langchain | Instructions file | 150k | today | MIT | A | 4,680 | Codex, OpenCode |
| deepseek-harness AGENTS.mddeepseek-ai/deepseek-harness | Instructions file | 283k | today | MIT | A | 4,342 | Codex, OpenCode |
| vscode copilot-instructions.mdmicrosoft/vscode | Instructions file | 196k | today | MIT | A | 3,153 | GitHub Copilot |