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.
npx agentmods add instructions/eclipsesource/review-guard/agents-mdgit clone --depth 1 https://github.com/eclipsesource/review-guardWhat 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 | $0.01854 | $0.01854 |
| Opus 5 | $0.00927 | $0.00927 |
| Sonnet 5 | $0.00371 | $0.00371 |
| Haiku 4.5 | $0.00185 | $0.00185 |
Grade A, and why
review-guard AGENTS.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 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.
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 — 35 lines — stays where its author put it; the contents beside it link to each section on GitHub.
For everything not covered here, the human-facing docs are authoritative: see
README.md for modes, tools, flags, and usage, and CONTRIBUTING.md for the
development setup, the script list, and the release process.
Project overview
ReviewGuard MCP (npm package @eclipsesource/review-guard-mcp, binary
review-guard-mcp, MCP server name review-guard) is a TypeScript MCP server
that lets AI agents work on GitHub PR reviews behind a safety boundary. In the
default pending mode agents read PR discussion context and create draft
(pending) reviews but cannot submit them. In the opt-in submit mode
(--allow-submit) an agent may also submit the review, restricted to an allowed
action set and always prefixed with a fixed disclaimer. Two transports: stdio
(for IDE-managed lifetime) and HTTP (remote server).
Architecture
src/github.ts:GitHubReviewClientclass. The ONLY file that touches GitHub APIs. Uses GraphQL for resolved review-thread context and mutations, and REST for review summaries and general PR comments. Review-thread comments (GraphQLreactionGroups) and conversation comments (RESTreactions) expose a normalizedreactionsmap (content -> count, e.g.THUMBS_DOWN) so agents can see downvotes. ThreadisResolved/resolvedByare also returned. Review comments, review summaries and conversation comments all carry their GitHub permalink asurl, so an agent can link an earlier discussion (a thread's permalink is its first comment'surl). The field is namedurleverywhere even though it comes from GraphQLurlfor review comments and RESThtml_urlfor the REST-sourced ones, since the output shape should be more consistent than the GitHub API is. A pending review comment already carries its final permalink, which starts resolving when the review is submitted, so an agent can cross-link its own findings while drafting. TheReviewComment.urldoc comment and the pending tool descriptions say so, and the integration suite asserts that a comment's permalink survives submission unchanged. Safety boundary: comment/thread mutations exclude theeventfield, a post-creation tripwire verifiesPENDINGstate (its error message tells the agent to stop and alert the human, since many MCP clients do not surface tool errors), and write operations are limited to the authenticated user's review. Submission is gated:submitReview(the only placesubmitPullRequestReviewis called) refuses any action not in theallowSubmitset passed to the constructor, and always prefixes the review body with the fixedsubmitBody(an optional calleradditionalBodyis appended below it, never replacing it). Thread resolution is gated too:resolveReviewThreadruns only whenallowResolveis set AND the thread's first comment is authored by the authenticated user (it refuses others' threads), so a bot can tidy up its own now-fixed findings but not close anyone else's conversations. PR scoping is enforced here too: when the client is constructed with ascope(owner/repo/PR), every public method callsassertInScopeand refuses input targeting a different PR/repo, andresolveReviewThreadverifies the thread's PR matches the scope. This is the authoritative boundary. The server-side schema change is convenience on top of it.src/server.ts:createMcpServer(client)factory. Registersget_pr_review_context,list_pending_review,add_review_comments,modify_review_comment, anddelete_pending_reviewwith Zod schemas. Additionally registerssubmitonly whenclient.allowedSubmitActionsis non-empty (itsactionenum is restricted to that set), andresolve_review_threadonly whenclient.resolveEnabled(started with--allow-resolve). Whenclient.scopedPullRequestis set, the PR tools omit theirowner/repo/pull_numberarguments and act on the scoped PR implicitly. Shared by both transports.src/stdio.ts: Stdio transport entry point. Connects the MCP server to stdin/stdout for IDE-managed lifetime (Theia, VS Code).src/http.ts: HTTP transport. A plainnode:httpserver exposing stateless Streamable HTTP at/mcp(POST only, GET/DELETE return 405, other paths 404). No web framework: the MCP transport parses the request body and enforces the Host header itself. Receives{ port, host }from the entry point and binds that address (default127.0.0.1). Validates the Host header of incoming requests (DNS rebinding protection): loopback aliases only for a loopback bind, plus the bind address and the container-runtime host names (host.docker.internal,host.containers.internal) for a non-loopback bind.src/index.ts: Thin dispatcher. Parses the CLI viaparseCliOptions, resolves the GitHub token, applies the default submit disclaimer, creates the client, then delegates to--stdioor HTTP mode.src/args.ts:parseCliOptions, the single place that knows every CLI flag. Accepts--flag valueand--flag=value. ThrowsCliUsageErroron unknown flags, missing values, or duplicates (the entry point prints it and exits non-zero), so a typo'd hardening flag (e.g. the--repo/--prscope) can never be silently ignored.test/: vitest suite.args.test.tscovers CLI validation.github.test.tscovers the safety gates (submit refusal, fixed-body prefixing, scope assertion, own-thread resolve, PENDING tripwire) with mockedgql/octokitinternals.server.test.tscovers tool registration and dispatch over an in-memory MCP transport.server-json.test.tscovers the MCP Registry entry (see Key conventions). Changes to the safety boundary must keep this coverage.test/integration/review-guard.itest.tsis a manual end-to-end suite (npm run test:integration, own config invitest.integration.config.ts) that drives every tool against a real GitHub repository with two accounts and verifies the results through direct API reads. It is excluded fromnpm testand CI, setup is documented in CONTRIBUTING.md.
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.
- 2d ago First seen · 35 lines · 1,854 tokens per session scan A d4822e364a9e
review-guard AGENTS.md is an instructions file published in the GitHub repository eclipsesource/review-guard (4 stars, last pushed 6d ago), licensed MIT. It adds 1,854 tokens to every session, about $0.0093 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.
Other instructions, from other repositories
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.
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.
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.
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).
next.js AGENTS.md
Instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.