agentleFS
danicatdanicat/speedgrapher/rules/AGENTS.md

speedgrapher / rules

AGENTS.md51 starsChanged 17 days agoApache-2.0View on GitHub

Speedgrapher pairs an Antigravity side-panel editor with a Model Context Protocol server and five writing skills. You use it to brainstorm articles, draft clear technical prose, cut corporate jargon and synthetic phrasing, lint grammar, and audit search engine metadata. Whenever a user creates, edits, or reviews a *.md file, surface a pill in chat to open Speedgrapher in the auxiliary pane while keeping standard filename.md links for the default Markdown viewer: You can switch between four workspace modes in the…

We split this AGENTS.md into the rules it gives an agent

7 rules. About 0 are common, 0 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use file.md to open a specific file (using the file's basename as the link text).

    line 9
  2. Very specific:

    Use Speedgrapher when no specific file sits in scope.

    line 10
  3. Very specific:

    Use speedgrapher to surface the Speedgrapher pill and open the side-panel Markdown editor.

    line 49
  4. Very specific:

    Use ideate to interview engineers one question at a time (Open -> Focused -> Closed) and outline blog posts, README.md files, or SKILL.md …

    line 50
  5. Very specific:

    Use draft to write technical articles and docs that match the author's voice with a 150-word outcome lead, *why-before-how* code examples, …

    line 51
  6. Very specific:

    Use readability to replace status-signaling corporate jargon and undefined acronyms with plain English using the Harvard Business Review …

    line 52
  7. Very specific:

    Use publish to run fog, slop, vale, and seo, resolve or inject inline <!-- review: ...

    line 53

≈ “Common” is an estimate: at least 1 in 5 of the 36,368 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Speedgrapher agent rules

Speedgrapher pairs an Antigravity side-panel editor with a Model Context Protocol server and five writing skills. You use it to brainstorm articles, draft clear technical prose, cut corporate jargon and synthetic phrasing, lint grammar, and audit search engine metadata.

Side-panel workspace and modes

Whenever a user creates, edits, or reviews a *.md file, surface a pill in chat to open Speedgrapher in the auxiliary pane while keeping standard [filename.md](file:///...) links for the default Markdown viewer:

  • Use [file.md](sidecar://speedgrapher/panel/?file=/absolute/path/to/file.md) to open a specific file (using the file's basename as the link text).
  • Use [Speedgrapher](sidecar://speedgrapher/panel/) when no specific file sits in scope.

You can switch between four workspace modes in the side panel:

  1. In Reader mode (Reader), you read a clean, read-only view of the rendered Markdown without body highlights, hover bars, or inline comment threads.
  2. In Editor mode (Editor), you click any block to edit its Markdown in place with ⌘Z and ⌘Y (⇧⌘Z) undo and redo, toggle four Editorial Lenses (Heatmap, Slop, Vale, and Fog), and inspect Before and after fix previews. Each active paragraph shows a docked header bar with its score pill, Comment, Send to chat, AI Rewrite, and 3-Angle Variant Tournament actions. You can also review staged ✓ Accept Change and ✕ Reject diff cards, manage AI provider and project or global Vale rules in ⚙️ Settings, and open the SEO & GEO popover.
  3. In Reviewer mode (Reviewer), direct text editing locks while all <!-- review: ... --> threads render inline below their paragraphs alongside the diagnostic lenses (Heatmap, Slop, Vale, and Fog). You can add or resolve review comments or dispatch them to the panel AI or main chat.
  4. In Raw mode (Raw), you edit the full Markdown source in a single plain-text buffer with ⌘Z and ⌘Y (⇧⌘Z) undo and redo.

Inline review comments and panel AI vs. main chat

You and your user exchange paragraph feedback through standard HTML comments placed right below a paragraph on disk. Comments use either plain <!-- review: ... --> tags or UI tags with id, quote, and ts fields (<!-- review(id="...", quote="...", ts="..."): ... -->):

Our collector buffers spans in memory before flushing them to storage.
<!-- review: Mention the 5-second timeout and Zstandard compression here. -->
<!-- review(id="c1", quote="buffers spans", ts="2026-10-16T00:00:00Z"): Clarify the ring buffer size limit. -->

In-panel AI actions (Send Comments, ✨ Rewrite with AI, 3-Angle Variant Tournament, and Fix SEO) run through Speedgrapher's built-in Gemini or Vertex AI client (set in ⚙️ Settings to fast models like gemini-3.8-flash). They return proposals in memory and stage ✓ Accept Change and ✕ Reject diff cards without writing to disk until the user clicks ✓ Accept Change.

The user clicks 💬 Send to Main Chat when they want the main chat session to read the text or run chat workflows. When your user asks you in chat to resolve review comments in a file, find every <!-- review: ... --> and <!-- review(...): ... --> block, rewrite the preceding paragraph to answer the note, and delete the tag. When your user asks you to mark up a file in place instead of rewriting it right away, insert concise <!-- review: [Dimension] Suggestion... --> tags below the paragraphs that need work.

Analysis tools

The speedgrapher server provides four tools. Always pass absolute file paths in the path field when you audit workspace files.

The fog tool computes file and per-paragraph Gunning Fog readability scores from sentence length and words with three or more syllables. Aim for a score between 11 and 15. Split sentences over 30 words and replace multi-syllable noun stacks with active verbs when a paragraph exceeds 16.

The slop tool scores text from 0 to 100 across five dimensions: structural clichés, lexical tells, filler words, sentence rhythm variance, and active voice. Keep scores under 20, and never over 25.

The vale tool runs static style and grammar checks with the Google, proselint, and write-good rule sets through Vale. Fix all error and warning alerts before release.

The seo tool audits titles, meta tags (summary versus description), heading order, image alt text, links, and word count. Pass path, html, or url along with doc_type (page or doc) and a target keyword. When you audit a Hugo site with hugo installed, the tool builds draft posts before checking the output HTML.

Writing skills

Activate the matching skill whenever your user's task fits one of these five stages:

  1. Use speedgrapher to surface the [Speedgrapher](sidecar://speedgrapher/panel/) pill and open the side-panel Markdown editor.
  2. Use ideate to interview engineers one question at a time (Open -> Focused -> Closed) and outline blog posts, README.md files, or SKILL.md files with the inverted pyramid.
  3. Use draft to write technical articles and docs that match the author's voice with a 150-word outcome lead, why-before-how code examples, and sentence-case headings.
  4. Use readability to replace status-signaling corporate jargon and undefined acronyms with plain English using the Harvard Business Review jargon framework, purge AI tropes (deslopify), and fix paragraph fog and vale trouble spots.
  5. Use publish to run fog, slop, vale, and seo, resolve or inject inline <!-- review: ... --> notes, build a line-indexed scorecard, and check front matter metadata and translations before release.

Download now (ZIP)

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 28,693 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository6 other files it gives its agents.

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.