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/benelser/docent/agents-mdgit clone --depth 1 https://github.com/benelser/docentWhat 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 | $0.02611 | $0.02611 |
| Opus 5 | $0.01306 | $0.01306 |
| Sonnet 5 | $0.00522 | $0.00522 |
| Haiku 4.5 | $0.00261 | $0.00261 |
Grade A, and why
docent 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 — 193 lines — stays where its author put it; the contents beside it link to each section on GitHub.
docent — Agent Instructions
You turn what can be learned about a codebase into a narrated, animated explainer — a short film that shows and explains software, the way a museum docent walks a visitor through an exhibit.
docent is generic machinery. Point it at any repository, give it a prompt, and it runs the whole pipeline. It works in two modes:
- Architecture review — the whole system, or a subsystem, in depth: the components, how control and data flow, the idioms that make it what it is.
- PR review — a pull request, reviewed the way a principled engineer would: why the change exists, whether the design is sound, the core before → after, what ripples, what could break, and a verdict.
Either way the job is the same: survey it, write a film spec, run the build. A film is a flowing narrative, never a checklist. This file is your brief.
The core principle
Content is data; presentation is the engine. You never write animation code for a particular codebase. You survey the subject and author a single declarative JSON file — a film spec. The engine renders it. The engine knows nothing about any specific repository.
A second principle, learned the hard way: narration is decoupled from animation. Speech is its own pipeline stage with its own cache. No renderer ever blocks on TTS.
The stack
All local — no API keys, no cloud.
- Remotion — the render engine. Scenes are typed React/TSX components, driven entirely by the film spec. Embarrassingly parallel: it shards frames across every core.
- Manim — optional embedded clips, for continuous physics-driven motion where it beats CSS. Rendered as transparent video, composited by Remotion.
- Kokoro — local neural TTS (voice
af_heart). - bun — runs the pipeline and the Remotion CLI. ffmpeg — encode/mux.
One-time setup: bun install and uv sync.
The pipeline — a cascade, parallel at every stage
survey → films/<id>.json the spec — you author this
tts → public/audio/<id>/* Kokoro, beats in parallel + manifest.json
clips → public/clips/<id>/* optional Manim inserts, parallel
render → out/<id>.mp4 Remotion, frames in parallel
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 · 193 lines · 2,611 tokens per session scan A c486586f1162
docent AGENTS.md is an instructions file published in the GitHub repository benelser/docent (6 stars, last pushed 2mo ago), licensed MIT. It adds 2,611 tokens to every session, about $0.0131 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
remocn CLAUDE.md
Instructions for Remocn/remocn, covering remocn, что это, целевая аудитория, архитектура and два уровня компонентов.
claude-youtube-editor CLAUDE.md
Instructions for hassancs91/claude-youtube-editor, covering claude.md — claude-youtube-editor, layout, the example and conventions (hard rules).
marketing-studio CLAUDE.md
Instructions for ucsandman/marketing-studio: Remotion (studio/) renders all final video for ALL products/brands; assets are copied out to the product's repo at the end. brands/ .json holds per-product tokens (zod-validated via studio/src/lib/brand.ts; mark components in studio/src/brands/marks.ts); templates resolve…
remotion-agent-catalog AGENTS.md
Instructions for victorsodre/remotion-agent-catalog, covering agents.md, as seis armadilhas, 1. usevideoconfig() reporta a composição, não o container, 2. sequence renderiza um absolutefill and 3. defaultprops passa por json.
motion-superpowers CLAUDE.md
Claude Code instructions for ardha27/motion-superpowers, covering agent instructions: motion superpowers, 1. grounded research first, 2. deterministic frame timelines, 3. style rulebook enforcement and 4. visual test-driven development (tdd).
remotion-marketing-video-skill CLAUDE.md
Instructions for xsourabhsharma/remotion-marketing-video-skill, covering shared brain, project, structure and rules.