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/craftled/openklip/agents-mdgit clone --depth 1 https://github.com/craftled/openklipWhat 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 | $0.21283 | $0.21283 |
| Opus 5 | $0.10641 | $0.10641 |
| Sonnet 5 | $0.04257 | $0.04257 |
| Haiku 4.5 | $0.02128 | $0.02128 |
Grade B, and why
openklip AGENTS.md scanned grade B with 1 finding 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.
Unrestricted tool accessmediumExcessive agency
A wildcard tool grant or "run any command" leaves no least-privilege boundary at all.
**Deferred MCP surface (default):** at connect the server enables a **core** edit-loop set (~28 tools: status, transcript, cut, export, brief, tasks, revert, …) plus three meta tools: `tools_catalog` (name/summary search How it starts
The opening of the file, as written. The whole thing — 686 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
Single source of truth for AI agents working in this repo: OpenKlip editing workflow, guardrails, copy rules, and code standards.
OpenKlip agent skill
OpenKlip is a local-first, agent-native video toolchain: external agents run the edit loop via CLI; the browser reviews the same project.json.
The file model (read this first)
Each project lives as plain files under projects/<slug>/ in a layered layout. The parent directory is resolved by projectsRoot():
OPENKLIP_PROJECTS_ROOTif set- else
.openklip/projects-rootin the app cwd (GUI folder picker) - else
~/Movies/OpenKlip
projects/<slug>/
project.json the EDL - the edit itself (the only file you edit)
brief.md optional project brief: audience, goal, tone,
must-use assets, avoid list, target length, formats
assets/ user originals (flat): drop b-roll, music, stills here
graphics/ optional project-local graphic templates (override bundled ids)
working/ derived media + scratch: proxy.mp4, transcript.json,
audio16k.f32, frames/, asset proxies, chats.json,
actions.jsonl, tasks.json…
output/out.mp4 the rendered export
Edit templates (repo root, not per project):
templates/<id>/skill.md agent playbook (cuts, overlays, export loop)
Optional template field on project.json points at a template id (e.g. talking-head).
project.json IS the edit. It holds every transcribed word with a deleted flag, b-roll overlays, still (Ken Burns) overlays, push-in zooms, title cards, graphics (HTML/CSS template overlays), music placements, captions settings, look flags, optional sceneLog (visual spans + speaker focus), optional highlights (LLM clip candidates), and export (aspect + reframe crop). Everything under working/ and output/ is regenerated from it. The GUI editor and these CLI commands both read and write this same file; they are equivalent (parity). Edit it through the CLI; the browser editor will show the same result, and vice-versa.
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 · 686 lines · 21,283 tokens per session scan B 9715856702c2
openklip AGENTS.md is an instructions file published in the GitHub repository craftled/openklip (2 stars, last pushed 22d ago), licensed MIT. It adds 21,283 tokens to every session, about $0.1064 per session on Opus 5. A static security scan graded it B with 1 finding (unrestricted tool access). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
heygen-cli AGENTS.md
Instructions for heygen-com/heygen-cli, covering heygen cli, build & test, key rules, gen/ is generated — never hand-edit and command.spec is immutable.
screenpipe CLAUDE.md
Claude Code instructions for screenpipe/screenpipe: Conventions live in one file so they cannot drift. Read AGENTS.md.
ditherer AGENTS.md
Instructions for gyng/ditherer, covering ditherer — agent guidelines, project overview, architecture, component hierarchy (atomic design) and state management.
capcut-ai-editor CLAUDE.md
Instructions for mrbuslov/capcut-ai-editor, covering smartcut mcp server — project context, what is this?, project structure, key files and config.py.
augent CLAUDE.md
Instructions for AugentDevs/augent, covering augent — audio + agents, audio-downloader (built by augent), download audio from youtube (or any video url), download to custom folder and download multiple urls.
contexture webrtc.instructions.md
Instructions for AcKeskin/contexture, covering webrtc rules, data channels and reliability, media and tracks and peer connection lifecycle.