secondsky/claude-skills / plugins/clefcut-cloudflare-video/skills/clefcut-cloudflare-video
clefcut-cloudflare-video
VerifiedCut long videos into clips for YouTube, YouTube Shorts, TikTok, Instagram Reels, X (Twitter), and similar creator platforms. Free local scene, silence, and OCR analysis; Cloudflare CLEF rating; verified ffmpeg cuts. Use when repurposing long video.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 20.5 KB |
| references/ai-gateway-and-mcp.md | markdown | 5.6 KB |
| references/clairic.md | markdown | 7.9 KB |
| references/clef.md | markdown | 11.2 KB |
| references/clip-selection.md | markdown | 6.5 KB |
| references/platform-specs.md | markdown | 12.6 KB |
| references/workflow.md | markdown | 15.9 KB |
| scripts/clipcut.mjsexecutable | script | 205.3 KB |
| templates/env.example | other | 588 B |
| templates/questions.example.json | config | 2.5 KB |
references/ai-gateway-and-mcp.md
View on GitHub# AI Gateway and the Cloudflare MCP server
Two optional paths on top of plain REST calls: routing through an AI Gateway,
and calling the Cloudflare API through the Cloudflare MCP server. Facts
verified against Cloudflare docs on 2026-10-20; inferred patterns are labeled.
## AI Gateway
### What it adds
Routing model calls through an AI Gateway provides unified logging and
analytics, response caching, rate limiting, retries, and security controls,
plus dynamic routing (fallbacks, A/B tests). Without a gateway you get none of
that — the script's own retry loop is all you have. For a clip run over a
dozen sheets, the useful pieces are caching (identical re-reads of unchanged
sheets cost nothing) and logs (a per-request spend trail).
### The unified endpoint (current, documented form)
Same URL as plain Workers AI, one mandatory extra header:
```
POST https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/@cf/cloudflare/clef
Authorization: Bearer {CLOUDFLARE_API_TOKEN}
cf-aig-gateway-id: {GATEWAY_ID}
Content-Type: application/json
```
- The `cf-aig-gateway-id` header is required when routing through a gateway
(for example `cf-aig-gateway-id: default`). Without a gateway configured,
the script calls the API directly and sends no such header.
- Auth stays the Cloudflare API token. An AI-Gateway-only token returns 401
(error code 10000); the token needs Workers AI - Read (account scope).
- A `default` gateway is auto-created; it uses Standard billing. To spend
prepaid credits, the gateway must be on Unified billing.
In this skill, set `CLEFCUT_GATEWAY_ID` in the environment and the script
routes through the gateway.
### Per-request cache headers
Caching is DISABLED by default. Sending `cf-aig-cache-key` alone opts the
request in.
| Header | Effect |
|---|---|
| `cf-aig-cache-key` | Custom cache key; its presence opts the request into caching |
| `cf-aig-cache-ttl` | TTL in seconds; min 60 s, max 1 month. Default TTL when caching is on with no TTL set: 5 minutes |
| `cf-aig-cache-status` (response) | `HIT` or `MISS` |
| `cf-aig-skip-cache: true` | Bypass the cache for this request |
The default cache key is a SHA-256 over provider, endpoint, model, provider
auth header, and the FULL request body — exact-match caching only. Re-rating
the same unchanged sheet bytes with the same questions is a HIT; any change
to the question text is a MISS. Only text and image responses are cacheable,
and the cache is volatile.
Other per-request headers (they override gateway-level settings):
`cf-aig-max-attempts` (retry attempts), `cf-aig-retry-delay`,
`cf-aig-backoff`, `cf-aig-request-timeout`, `cf-aig-metadata` (tag requests
with ids), `cf-aig-collect-log`, `cf-aig-custom-cost`. Using
`cf-aig-max-attempts` moves retries from the client loop into the gateway —
handy for multi-segment sweeps.
### The legacy provider form — do not use
A provider-specific form `https://gateway.ai.cloudflare.com/v1/{account_id}/{gateway_id}/{provider}`
exists generically and Workers AI is among its providers, but the live
Workers-AI provider documentation no longer shows this URL, and the exact
workers-ai path segment is unverified. Use the unified `api.cloudflare.com`
form above only. Do not put the `gateway.ai.cloudflare.com` form into scripts
or docs without testing it first.
## Cloudflare API MCP server
### What it is
- URL: `https://mcp.cloudflare.com/mcp` (Streamable HTTP; old `/sse` URLs are
aliases). Docs server: `https://docs.mcp.cloudflare.com/mcp`.
- Client config: `{"mcpServers": {"cloudflare-api": {"url": "https://mcp.cloudflare.com/mcp"}}}`.
- Auth: interactive OAuth (redirect to Cloudflare, pick permissions) or, for
CI/CD, a Cloudflare API token passed as a bearer token in the
`Authorization` header (user and account tokens both work).
### Search-and-execute (Code Mode)
The server exposes two tools, `search` and `execute`, each taking a single
required `code` property — a JavaScript async arrow function. This keeps the
tool schemas tiny (~1,000 tokens instead of ~1,170,000 for full schemas).
- `search`: code runs against `spec` — the full OpenAPI spec with `$refs`
pre-resolved. Example from the docs:
`spec.paths['/zones/{zone_id}/rulesets']?.get`.
- `execute`: code gets an authenticated client and runs in an isolated
Dynamic Worker sandbox. Verbatim docs example:
`await cloudflare.request({ method: "GET", path: \`/zones/${zoneId}/rulesets\` })`.
### Pattern for a clef call through execute — INFERRED
The rulesets example above is documented verbatim; the `/ai/run` adaptation
below is a pattern inferred from it, not a docs example. Read the request
schema with `search` first, then:
```js
// search: read the endpoint's request schema
async () => spec.paths['/accounts/{account_id}/ai/run/{model_name}']?.post
// execute: run the clef call (path-param spelling and body-passing inferred)
async () => await cloudflare.request({
method: "POST",
path: `/accounts/${accountId}/ai/run/@cf/cloudflare/clef`,
body: { model: "clef", state, questions },
})
```
If the path parameters or body handling differ, `search` will show the real
shape — trust it over this pattern.
### When MCP is worth it
- Use the MCP path when the user already has the Cloudflare MCP server
connected, or wants to explore the API shape interactively (search the spec,
inspect schemas) before committing to a script change.
- Use plain curl or Node fetch (the default in this skill) for every routine
rating or read call: fewer moving parts, no sandbox, exact same endpoint.
- Never route sheets or frames through MCP to "save" setup: the data path is
the same, and the sandbox adds a failure mode the script does not control.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |