agentleFS
shandleyshandley/stenograph/CLAUDE.md

stenograph

CLAUDE.md1 starsChanged 11 months agoMITView on GitHub

A stenographic grammar for efficient human-AI coding collaboration. "Structure in, intelligence where needed." Court stenographers capture complex proceedings at 225+ WPM using compressed, structured input. Steno-graph brings this efficiency to AI coding — fewer keystrokes, maximum precision, direct execution. The goal is flow state, not interrupt-driven batch processing. Steno-graph IS: - A terse grammar for expressing intent to Claude - A workflow tracking system with branching and bookmarks - Domain-agnostic (works for any coding task) - A skill that Claude…

  • In this repositoryOne other file it gives its agents.

The file

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

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    Search: steno:search pattern (by verb, modifier, file)

    line 95
  2. Very specific:

    Search: Search commands by text, jump to matches (press /)

    line 114
  3. Very specific:

    Fix: Use plain text in tables, not code formatting.

    line 149
  4. Distinctive:

    After skill changes, test in fresh Claude Code session:

    line 204

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.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
CLAUDE.md

Steno-Graph

A stenographic grammar for efficient human-AI coding collaboration.

Vision

"Structure in, intelligence where needed."

Court stenographers capture complex proceedings at 225+ WPM using compressed, structured input. Steno-graph brings this efficiency to AI coding — fewer keystrokes, maximum precision, direct execution.

The goal is flow state, not interrupt-driven batch processing.

What Steno-Graph Is (and Isn't)

Steno-graph IS:

  • A terse grammar for expressing intent to Claude
  • A workflow tracking system with branching and bookmarks
  • Domain-agnostic (works for any coding task)
  • A skill that Claude learns and interprets

Steno-graph IS NOT:

  • An execution engine or computational backend
  • A replacement for actual tools (R, Python, etc.)
  • A guarantor of computational correctness

Claude interprets steno commands using its available tools. The quality of execution depends on what Claude can actually do, not on steno-graph itself.

Architecture: Skill-First

Key insight: Claude can learn the grammar via a skill and parse/execute directly. No preprocessing, no daemon, no separate parser needed.

You type:     dx:@app.ts
     ↓
Claude recognizes the pattern (via skill)
     ↓
Executes using available tools, returns results

Source of Truth

.claude/skills/steno/SKILL.md — The complete grammar, session tracking, and branching behavior.

Project Structure

steno-graph/
├── .claude/
│   ├── skills/steno/       # THE SKILL - source of truth
│   ├── hooks/              # Claude Code hooks for steno-flow
│   │   ├── flow-pre.sh     # Pre-tool event hook
│   │   ├── flow-post.sh    # Post-tool event hook
│   │   └── flow-stop.sh    # Command complete hook
│   └── settings.json       # Hook configuration
├── .steno/
│   ├── current-session.json  # Active session state
│   ├── flow-server.cjs       # WebSocket relay server
│   ├── flow.html             # Real-time visualization dashboard
│   └── start-flow.sh         # Server startup script
├── examples/               # Usage vignettes
├── design/                 # Active design docs
│   ├── steno-flow.md       # Real-time visualization design
│   └── biostack-integration.md
├── CHEATSHEET.md           # Quick reference
└── README.md               # User-facing docs

Features

Core Grammar

  • 9 verbs: dx, mk, ch, rm, fnd, viz, stat, ts, doc
  • Modifiers: @file, ^, +add, -exclude, .flag
  • Precision: ~ (flexible), ! (exact), ? (ask), ~deep
  • Modes: ?plan, ?sketch, ?challenge, ?explore

Session Tracking

  • Command history with .steno/current-session.json
  • Stale detection and refresh
  • Bookmarks for reference points
  • ASCII graph visualization
  • Real-time updates via steno-flow

Branching

  • fork/switch/compare/merge/abandon
  • Explore alternatives without losing work
  • Lightweight (metadata-only, not file state)

Advanced Features

  • Cross-branch references: @branch:^, @branch:n_001
  • Undo/redo: steno:undo, steno:redo
  • Export/import: steno:export, steno:import

Power Commands (v1.1+)

  • Aliases: steno:alias name "cmd" (with {param} support)
  • Search: steno:search pattern (by verb, modifier, file)
  • Replay: steno:replay n_001..n_005
  • Templates: steno:template react-component Button
  • Diff: steno:diff @baseline ^
  • Transcripts: steno:transcript +generate / +all (integrates claude-code-transcripts)
  • Rich export: steno:export .html +transcript (interactive HTML with search)
  • Theming: steno:transcript +theme (shadcn-inspired, light/dark mode)

Steno Flow (Real-time Visualization)

Inspired by Strudel REPL's live coding visualizations, steno-flow provides real-time feedback during Claude Code sessions. Instead of waiting passively, you can "surf" your coding session.

Components:

  • Command Roll: Horizontal timeline of steno commands with verb-colored nodes
  • File Activity: Shows files being read/written as animated chips with elapsed time
  • Diff Scope: Oscilloscope-style waveform showing tool activity intensity
  • Session Graph: ASCII tree of all nodes and branches
  • Session Stats: Operations count, success rate, top files accessed (toggle with S)
  • Command Filter: Filter nodes by verb type with colored buttons
  • Search: Search commands by text, jump to matches (press /)
  • Error Log: Collapsible panel showing errors with timestamps (toggle with E)
  • Audio Feedback: Soft ambient tones for tool start/complete/error (toggle with A)

Starting Steno Flow:

# Terminal 1: Start the flow server
.steno/start-flow.sh

# Terminal 2: Open dashboard
open http://localhost:3847

# Terminal 3: Run Claude Code with hooks enabled
claude --dangerously-skip-permissions

Architecture:

Claude Code → Hooks (pre/post/stop) → HTTP POST → Flow Server → WebSocket → Browser

The hooks in .claude/hooks/ fire on every tool use, sending events to the flow server on port 3847. The browser connects via WebSocket for real-time updates.

Key Files:

  • .claude/hooks/flow-*.sh — Event capture hooks
  • .claude/settings.json — Hook configuration
  • .steno/flow-server.cjs — Node.js WebSocket relay
  • .steno/flow.html — Dashboard UI

Development Lessons

1. Skill Description Parsing

Problem: Backticks in skill description tables cause bash parsing errors.

Fix: Use plain text in tables, not code formatting.

# Bad - causes "Bash command failed"
| `!` | Exact |

# Good
| ! | Exact |

2. Branching Commands Need Explicit Triggers

Problem: compare:main experiment was interpreted as git command.

Fix: Explicitly list branching verbs in skill description and trigger patterns.

3. Track-Only Branching

Branches track commands and metadata, not file states. Files stay as-is when switching branches. Use git for file state management if needed.

4. Execution Reality

Steno-graph is a grammar, not an execution engine. If Claude doesn't have access to a tool (e.g., a statistical package), it will interpret the command using its training knowledge, not execute the actual algorithm. This is a Claude limitation, not a steno-graph limitation.

Grammar Quick Reference

[mode][verb]:[target] [@refs] [+add] [-exclude] [.flag] [precision]

Core verbs: dx, mk, ch, rm, fnd, viz, stat, ts, doc Branching: fork, switch, compare, merge, abandon Session: steno:history, steno:graph, steno:bookmark, steno:undo, steno:redo, steno:export, steno:import Power: steno:alias, steno:search, steno:replay, steno:template, steno:diff, steno:transcript

See CHEATSHEET.md for complete reference.

Future Considerations

Backend Integration (design/biostack-integration.md)

For domain-specific validated computation, steno-graph can integrate with execution backends like biostack. The design supports:

  • Backend detection and routing
  • Command translation
  • Provenance linking
  • Two-tier branching (lightweight vs. computational)

ggterm Integration (design/future-analytics.md)

The session graph could power advanced visualizations:

  • steno:timeline — Command timeline
  • steno:stats — Workflow analytics
  • steno:flow — Dependency network

Testing Workflow

After skill changes, test in fresh Claude Code session:

claude --dangerously-skip-permissions

Test core commands:

dx:@examples/grammar-basics/README.md
steno:history
steno:graph
fork:test
switch:main
compare:main test
steno:undo
steno:export

Test power commands:

steno:alias greet "dx:@{1}"
greet README.md
steno:search dx
steno:replay n_001
steno:template
steno:diff n_001 n_002

Key Files

FilePurpose
.claude/skills/steno/SKILL.mdGrammar + behavior (edit here)
.claude/hooks/flow-*.shClaude Code hooks for steno-flow
.claude/settings.jsonHook configuration
.steno/current-session.jsonActive session state
.steno/flow-server.cjsWebSocket relay server
.steno/flow.htmlReal-time visualization dashboard
CHEATSHEET.mdUser quick reference
README.mdUser-facing documentation
design/steno-flow.mdFlow visualization design doc

Download now (ZIP)

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.