Borrowing it
Nothing to install: this file belongs to rmanish2000-del/warrant-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/rmanish2000-del/warrant-mcp/main/CLAUDE.mdgit clone --depth 1 https://github.com/rmanish2000-del/warrant-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/rmanish2000-del/warrant-mcp/claude-md)<a href="https://agentmods.dev/instructions/rmanish2000-del/warrant-mcp/claude-md"><img src="https://agentmods.dev/badge/instructions/rmanish2000-del/warrant-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/rmanish2000-del/warrant-mcp/claude-md"><img src="https://agentmods.dev/badge/instructions/rmanish2000-del/warrant-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 | $0.04139 | $0.04139 |
| Opus 5 | $0.02070 | $0.02070 |
| Sonnet 5 | $0.00828 | $0.00828 |
| Haiku 4.5 | $0.00414 | $0.00414 |
Grade D, and why
warrant-mcp CLAUDE.md scanned grade D 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 10d 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.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
junk-dir", not `sudo rm -rf /`. Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
junk-dir", not `sudo rm -rf /`. How it starts
The opening of the file, as written. The whole thing — 275 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
Guidance for Claude Code working in this repository. Written in the same
spirit as the sibling warrant repo's CLAUDE.md: decisions recorded once so
the next session does not re-derive them — or quietly undo them.
Product
warrant-mcp — a policy firewall for agent tool calls. A human writes policy in plain English; Claude compiles it once, off stage into numbered clauses backed by structured rules; deterministic code checks every intended action and refuses what the policy does not allow, citing the governing clause; a Claude Code PreToolUse hook makes that refusal a hard block.
Claude compiles policy. Deterministic code decides. The hook makes a DENY mean the action does not happen.
This is a skeleton by design: two verdicts (ALLOW / DENY), three action kinds, one MCP tool. No payments, no ESCALATE, no UI, no accounts.
Relationship to the warrant repo
C:\Push-to-Prod-2026\warrant is the proven deterministic authorization
engine for agentic commerce. It is reference only: read it to understand
the model (type-level guarantee, cached replay, clause citation, fail closed).
It has never been modified by this project and must not be — verify with
git status there, and with its own invariant git diff submission..HEAD -- src/.
What was reused: the thinking. What was deliberately diverged:
| warrant | warrant-mcp | why | |
|---|---|---|---|
| compile cache | gitignored | committed | the reviewed cache is the enforceable artifact; a fresh clone must enforce without an API key |
| compile failure | labelled stub fallback | no fallback, fail loud | a recording must survive an outage; a made-up enforcement policy is worse than none |
| verdicts | ALLOW / ESCALATE / DENY | ALLOW / DENY only | two values are what make the hook a hard veto |
Architecture
- Toolchain — TypeScript strict, no build step for development
(
--experimental-strip-types, Node ≥ 22.6).erasableSyntaxOnlyis load-bearing: source must stay strippable. - The published package is different, and has to be. Node refuses to strip
types under
node_modules(ERR_UNSUPPORTED_NODE_MODULES_TYPE_STRIPPING), so an installed copy would not run at all.prepackemitsdist/viatsconfig.build.jsonand the tarball ships that. BecauseerasableSyntaxOnlyis on, the emit is pure type erasure — there is no transform that could change a verdict.bin/warrant-mcp.mjsrunsdist/when it exists andsrc/when it does not, so a checkout and an install behave identically. src/engine/— pure evaluation. No clock, no I/O, no network, no process spawning. Everything arrives as a parameter.src/compiler/— the only place the Anthropic SDK is imported. The server and the hook never import it.src/hook/—pretooluse.ts(the stdin/stdout boundary) plus a pureadapter.tsthat maps Claude Code tool calls onto action kinds.src/server/— MCP stdio server (main.ts), purehandler.ts, presentation inpresent.ts.src/record/— the authorization record. Append-only by convention, not in fact: the only writer appends and nothing enforces it, there is no hash, chain or signature, and anyone who can reach the file can rewrite it. Do not describe it as an audit trail anywhere — README and SECURITY-SURFACE.md §4 both spell out the gap, and they are the wording to match.types.tsis pure (parsing, the policy fingerprint);store.tsis the only writer;observe.tsis what the two boundaries call. Never imported by the engine, handler, hook adapter or presentation — see invariant 6.src/report/—warrant-mcp report, pure:model.tsaggregates,redact.tsis screen safety both ways (rewrite what is private, scan for what is secret),client.tsholds the page's CSS and script as strings,html.tsrenders. No clock, no filesystem;src/cli/report.tsis the boundary.src/authoring/— review-time only: rules→English, behaviour-diff corpus, unmapped guidance, CLI phrase parsing. Never imported by the engine, handler or hook.src/cli/— system boundaries: the only files that read the clock, the environment, or spawn anything.src/config/— where things live and how warrant merges into files it does not own.settings.tsis pure JSON merging, so the promise that a user's settings survive is testable without touching a disk.
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.
- 10d ago First seen · 275 lines · 4,139 tokens per session scan D 886b766bcd6e
warrant-mcp CLAUDE.md is an instructions file published in the GitHub repository rmanish2000-del/warrant-mcp (1 stars, last pushed 16d ago), licensed MIT. It adds 4,139 tokens to every session, about $0.0207 per session on Opus 5. A static security scan graded it D with 2 findings (asks for root, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
codex AGENTS.md
AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.