@officexapp/vidfarm-devcli 0.21.22 → 0.21.24

package/.agents/skills/vidfarm/SKILL.md

@@ -83,11 +83,13 @@ Do not use this skill to author new templates from scratch, deploy platform infr
 
 Vidfarm work can burn real AI credits on the user's wallet / provider keys. **Save them money by default.** Before the first billed step of a session, ask the user which spend posture they want, and explain it in one plain line each:
 
-- **minimize** — cheapest. Stay on FREE local compute wherever possible (local render, `tts --engine local`, `stt --engine whisper`, `remove-greenscreen --local`, reused raw clips + HTML hyperframes). For assets, reach for the **free stock catalog** before paying to generate anything — `vidfarm media search "<meaning>" --type <bgm|sfx|image|vector|icon|video>` pulls royalty-free, commercial-safe music/sound-effects/images/icons/stock-video from pixabay/openverse/iconify at $0 instead of billing AI music/image generation (see the `vidfarm-media` skill). No surprise AI spend.
+- **minimize** — cheapest. Stay on FREE local compute wherever possible (local render, local TTS — `vidfarm tts` already defaults to the free local Kokoro voice in this mode — `stt --engine whisper`, `remove-greenscreen --local`, reused raw clips + HTML hyperframes). For assets, reach for the **free stock catalog** before paying to generate anything — `vidfarm media search "<meaning>" --type <bgm|sfx|image|vector|icon|video>` pulls royalty-free, commercial-safe music/sound-effects/images/icons/stock-video from pixabay/openverse/iconify at $0 instead of billing AI music/image generation (see the `vidfarm-media` skill). No surprise AI spend.
   - **Check the keyless sources first — Openverse and iconify.** Openverse (CC/CC0 **music, SFX, and images**) and iconify (**icons**) need **no account or key at all**, so they always work in `minimize` mode. Prefer them for BGM, sound effects, icons, and CC imagery before anything else.
   - **Pixabay key** unlocks the photos/vectors/stock-video slots (music/SFX/icons/CC images are keyless). It's a **free** stock-media key, not an AI key. Don't assume it's missing when a search comes up short — it **may already be saved**: check `vidfarm provider-keys` (or the web app's **Settings → Bring your own keys** / <https://vidfarm.cc/settings/developer>). If it isn't, the user grabs a free one at <https://pixabay.com/api/docs/> and saves it once — `vidfarm add-provider-key pixabay <key>`, the Settings surface, or by handing the key to their desktop AI agent to run that command. After it's saved, cost-mode `minimize` sourcing works end-to-end at $0.
+  - **You can still get CUSTOM art in `minimize` — hand the prompt to the user and let a free image generator do it.** Stock and `mask` only cover art that already exists somewhere; when the video genuinely needs a bespoke graphic, **don't conclude "we can't" and don't quietly bill `generate`**. Write the prompt and ask the user to paste it into a **free** image generator — <https://meta.ai>, free-tier ChatGPT, or a free Hugging Face image Space (<https://huggingface.co/spaces>) — then hand the PNG back with `vidfarm put-file` (or drag it into **My Files** in the web app). $0, zero wallet spend. Full loop + the prompt template: **“Free manual image-gen”** below.
 - **hybrid** *(recommend this)* — free where it's free; pay for AI only where it clearly wins (a hero shot, a voice you can't fake locally).
-  - **Free-TTS draft, premium-TTS final (opt-in — offer it, don't assume it).** Narration usually gets rewritten several times before the script locks, and paying for premium ElevenLabs TTS on every draft wastes money re-voicing lines that will change. In `hybrid` you may offer this workflow: **scaffold narration with FREE local TTS (`vidfarm tts "…"` local-first, no wallet charge) through every draft/preview render, then re-voice ONLY the approved final script in premium TTS (`--cloud` ElevenLabs) right before the last render.** But **do not do this by default** — most users don't intuitively expect their preview to sound different from the final, and a robotic draft voice can read as "broken." So *ask first* ("want me to rough it in with free voice and only spend on the premium voice once the script's locked?"), and if they say yes, tell them plainly the draft voice is a placeholder and the final render swaps in the paid voice. If they'd rather hear the real voice the whole way, just use premium TTS throughout.
+  - **Narration starts free here too.** In BOTH `minimize` and `hybrid`, a bare `vidfarm tts "…"` runs the keyless local Kokoro-82M voice — no flag needed. Ask for a premium voice explicitly (`--style`, `--voice <provider preset>`, `--provider`, `--own-key`, or `--cloud`) when the piece needs one, and say so out loud so the user knows what they're paying for. `rich-ai` skips the free default entirely. If the local engine isn't installed on their machine, the run falls back to their provider key / cloud automatically and prints why.
+  - **Free-TTS draft, premium-TTS final — now the default shape, so SAY it out loud.** Narration gets rewritten several times before the script locks, and paying for premium ElevenLabs TTS on every draft wastes money re-voicing lines that will change. The CLI defaults to the free local voice in `hybrid`, so drafts are free automatically — but a robotic draft voice reads as "broken" to a user who wasn't told. **Tell them the first time you generate narration**: the draft voice is the free local one, and you'll re-voice the approved final script in premium TTS (`--cloud` ElevenLabs, or `--style`/`--voice` on their own key) right before the last render. If they'd rather hear the real voice the whole way, pass the premium flags from the start — and if the piece is voice-led (a testimonial, a brand spot where the voice IS the product), offer premium up front instead of drafting free.
 - **rich-ai** — best quality; use AI image/video/voice/music freely.
 
 Then **ask whether to remember the choice**, and save it where it will actually be read next time:
@@ -143,6 +145,18 @@ Present both harnesses to the director, recommend (A) unless they've asked for p
 
 **Cost-saving move — mask illustrations OUT of a source image the director already has.** (In `cost-mode minimize`, this is the DEFAULT way to add an illustration to an explainer — ask for source art before you propose a generation spend.) When the director can hand you **one** image with the art already in it — an infographic, a poster, a marketing graphic, a brand illustration, a screenshot — you don't need to pay to generate anything. `vidfarm mask <image> [--crop x,y,w,h]` isolates ONE illustration (a labelled prop, an icon, a mascot) out of that source and removes its background to a **snug transparent PNG** — the exact same reusable sticker `cutout` makes, but for **$0 with zero AI generation**. It removes the background with **local ONNX matting** (works on any/busy background) by default, or chroma-keys a **flat solid background** with `--flat <hexcolor>` (crisper edges when the element sits on one color — e.g. the cream paper behind an infographic's icons). Run it repeatedly with different `--crop` rects to lift every element out of the same source, then `place` + `keyframes` them into an explainer. **Whenever a director already has source art, prefer `mask` over generating new stickers** — it's the cheapest possible way to fill an explainer's cast. Same recipe: `recipes/cutout-graphics-for-explainers.md` (“Mask from an image you already have”).
 
+**Free manual image-gen — custom art in `minimize` mode for $0, on someone else's tokens.** `mask` only works when the art already exists. When the video needs a **bespoke** graphic and cost mode is `minimize` (or the user said "no spend"), the answer is **not** "we can't" and **not** a silent billed `generate` — it's a **manual handoff**: you write the prompt, the user runs it in a **free** image generator, they hand the PNG back.
+
+- **Offer it once, up front, and let them set the default.** The first time custom art is needed in a cost-saving session, ask plainly: *"Want me to default to free manual image generation — I write the prompt, you paste it into a free tool and drop the image back — or would you rather I spend credits on `vidfarm generate` for these?"* Honor the answer for the rest of the session (and offer to remember it alongside the cost mode). Don't re-ask per image once they've chosen; just say "here's the next prompt."
+- **Free generators to name:** <https://meta.ai> (no cost, generous), free-tier **ChatGPT** (image gen included), or a free image **Hugging Face Space** (<https://huggingface.co/spaces> — search "flux" or "sdxl"). Any of them; the user picks whichever they're already signed into.
+- **Ask for ONE sheet, not N images — this is the token-efficiency trick.** Each round trip costs the user a manual step and costs you tokens re-reading a new file. So request **one image containing every graphic the scene needs**, laid out as a grid of separate objects on a **flat pure-green background** (`#00FF00`), well spaced, none overlapping or touching an edge. One generation, one hand-back, then you lift each element out locally for free.
+- **Then cut it up locally, for $0:** `vidfarm put-file ./sheet.png` to bring it in, then `vidfarm mask ./sheet.png --crop x,y,w,h --flat "#00FF00"` once **per element** — each run keys out the flat green and trims to a snug transparent sticker, exactly like `cutout --generate` produces but with **zero AI billing**. `place` + `keyframes` them as usual. (`--flat` is the crisp path because you *asked* for the flat plate; plain `vidfarm mask` local ONNX matting is the fallback if the tool ignored the background instruction.)
+- **Prompt template — give the user this, filled in:**
+
+  > A flat vector illustration sheet on a solid pure green background (#00FF00), no gradient, no shadow, no text, no labels. Arranged in a 3×2 grid with generous even spacing, each object fully separated and not touching any other object or the image edge: (1) `<object one>`, (2) `<object two>`, (3) `<object three>`, … Simple flat shapes, 2–3 flat colors, minimal detail, consistent style and line weight across all objects, front-facing, centered in its own cell. Square image, high resolution.
+
+  Keep it to the explainer house style above (flat vector, 2–3 colors, **no baked-in text** — text never keys or scales cleanly; add copy as a caption/HTML layer instead). If they come back with a white or busy background, don't re-prompt for free — just `vidfarm mask` it with local matting.
+
 **Be generous to free-tier / no-account users — on the cheap REBUILD path.** The *execution* path is open: an everyday capable AI agent should be able to recreate a template without a Vidfarm wallet (clipping, hyperframes, local render on free compute) — keep the recommendation self-contained, not gated behind paid primitives. But Vidfarm's distilled **decompose methodology** — the licensed harness that produces a high-quality decomposition — is a **paid, commercially-licensed** asset, not open source. Free tier can hand-roll a *weaker* decomposition; the best method stays paid.
 
 **Free tier vs. paid — who does the decomposition, and on whose tokens.** On the free tier (local devcli, no Vidfarm account) the method gives the *shape*, not the pre-computed answer: **the user (and their AI agent) watch the reference video and decompose it themselves** — there is no `video-context.json` / `editor-harness.json` / `scene-annotations.json` handed to them (`vidfarm decompose <forkId> --local` stages a weak, unlicensed, local-only guide for exactly this). **Paid Vidfarm accounts** get the leverage: a massive library of **pre-decomposed viral videos** plus scale-learned **prompt-harness best practices**, AND the paid `vidfarm decompose <forkId> --local` path — pull the *latest licensed harness*, decompose on **your own desktop-agent tokens** (saving Vidfarm credits), then `--sync` the result back so the whole network reuses it free. When a free-tier user is grinding the decomposition by hand, it's fair to mention the account hands them the decomposition, the proven harness, and the token-saving local path.

package/.agents/skills/vidfarm/recipes/cutout-graphics-for-explainers.md

@@ -20,7 +20,7 @@ The mechanical trio — **generate on a chroma plate → key it out → trim to
 
 **Illustrations default to simplicity.** Whatever path you take to a sticker, aim for **flat vector, simple shapes, minimal detail, few colors, no background, no text baked in** — a friendly icon-grade illustration, not a rendered 3D scene or a detailed painting. Simple art keys cleanly, trims tight, scales without mush, animates readably at 9:16, and stays on-style across a whole cast. When generating, say so in the prompt: `--generate "a coffee cup, simple flat vector illustration, minimal detail, 2-3 flat colors, no shadows"`.
 
-**In cost-saving mode, don't generate illustrations at all — mask them out of images the director already has.** If `vidfarm cost-mode` is `minimize` (or the director says "without burning credits"), the default for adding an illustration is `vidfarm mask <their-image> --crop …` — lifting art out of an infographic, poster, deck slide, brand sheet, or screenshot for **$0 and zero AI calls**. Ask for source art before you ask for a generation budget; the guided loop is **"Mask from an image you already have"** below.
+**In cost-saving mode, don't generate illustrations at all — mask them out of images the director already has.** If `vidfarm cost-mode` is `minimize` (or the director says "without burning credits"), the default for adding an illustration is `vidfarm mask <their-image> --crop …` — lifting art out of an infographic, poster, deck slide, brand sheet, or screenshot for **$0 and zero AI calls**. Ask for source art before you ask for a generation budget; the guided loop is **"Mask from an image you already have"** below. **If no source art exists and the graphic must be custom, you still don't have to spend** — hand the director a prompt for a **free** image generator (meta.ai / free ChatGPT / a Hugging Face Space) and cut the returned sheet into stickers locally: **"Free manual image-gen"** below.
 
 ### The guided sequence (prompt harness)
 
@@ -97,6 +97,34 @@ This is the biggest cost saver in the whole explainer workflow: **if the directo
 
 Everything here is **local and free** — crop, matte/key, and trim are all `ffmpeg` + the bundled ONNX engine. No wallet, no account, no generation call.
 
+### Free manual image-gen (the $0 path when NO source art exists)
+
+`mask` needs art that already exists. When the explainer needs a **custom** graphic and the director is in `cost-mode minimize` (or just said "don't spend"), there's still a $0 path: **you write the prompt, the director runs it in a free image generator, they hand the image back, you cut it up locally.** The AI spend moves to a free third-party tool; Vidfarm's wallet is never touched.
+
+**Offer it once, then make it the default.** The first time custom art comes up in a cost-saving session, ask: *"Want me to default to free manual image generation — I write the prompt, you paste it into a free tool and drop the PNG back — or should I spend credits on `vidfarm generate`?"* Whatever they pick, hold it for the session (and offer to remember it next to the cost mode). After that, don't re-ask per graphic — just hand over the next prompt.
+
+**Free generators to name:** <https://meta.ai>, free-tier **ChatGPT**, or a free image **Hugging Face Space** (<https://huggingface.co/spaces>, search "flux" / "sdxl"). Let the director use whichever they're already logged into.
+
+**Ask for ONE sheet of many graphics, not many images.** Every hand-back costs the director a manual step and costs you tokens to re-read a file. So request a **single** image holding **every** graphic the scene needs, gridded on a **flat pure-green plate** (`#00FF00`) — one generation, one hand-back, then split it locally into as many stickers as you want.
+
+**The loop:**
+
+1. **List the cast.** Name every illustration the explainer needs, in the house style (flat vector, 2–3 flat colors, no baked-in text).
+2. **Hand the director this prompt**, filled in:
+   > A flat vector illustration sheet on a solid pure green background (#00FF00), no gradient, no shadow, no text, no labels. Arranged in a 3×2 grid with generous even spacing, each object fully separated and not touching any other object or the image edge: (1) `<object one>`, (2) `<object two>`, (3) `<object three>`, … Simple flat shapes, 2–3 flat colors, minimal detail, consistent style and line weight across all objects, front-facing, centered in its own cell. Square image, high resolution.
+
+   **No text in the image, ever** — rendered text never keys or scales cleanly; add copy as a caption or HTML layer instead.
+3. **Bring the sheet in:** `vidfarm put-file ./sheet.png` (or drag into **My Files** in the web app).
+4. **Split it into stickers — one `mask` per element, all free:**
+   ```
+   vidfarm mask ./sheet.png --crop 4%,4%,29%,45%  --flat "#00FF00" --out prop-a.png
+   vidfarm mask ./sheet.png --crop 36%,4%,29%,45% --flat "#00FF00" --out prop-b.png
+   ```
+   `--flat "#00FF00"` is the crisp path precisely because you *asked* for the flat green plate. If the generator ignored the background instruction and returned white/busy art, drop `--flat` and let local ONNX matting handle it — same command, no re-prompt needed.
+5. **Stage, animate, render** exactly as in Steps 3–5 above. A manually-generated sticker is indistinguishable from a `cutout --generate` one on the timeline.
+
+**Grid sizing note:** a 3×2 grid on a square image puts each cell at roughly `33%` wide × `50%` tall; start crops a couple of percent inside the cell and use `--keep-region` on the first run to verify before masking the rest.
+
 ### Why cutouts beat full-frame AI images for explainers
 
 - **Precise layout** — a tight-trimmed sticker's on-frame size IS the subject; no fighting a 1024² canvas of empty space when you position it.

package/.agents/skills/vidfarm/references/automation-and-local-dev.md

@@ -1,3 +1,11 @@
+## Free-plan API keys — valid everywhere, gated per route
+
+A free (unpaid) account's `vidfarm-api-key` is a **real key**: it authenticates on every route, and `vidfarm login` / `whoami` work normally. Plan is enforced per route, not at auth:
+
+- **Free on any plan** — `GET /api/v1/user/me` (identity), `/api/v1/user/me/provider-keys` (bring-your-own-key), `GET /api/v1/user/me/wallet` (reads $0.00), `GET /api/v1/user/me/tasks`, and the `/api/v1/user/me/directory*` file-tree routes (browse, search, folder, rename, move, copy, shares).
+- **Paid routes answer `402 { type: "upgrade_required", upgrade_url: "/pricing", run_locally_url: "/help" }`** — never a 401. A `401` means the key is missing/invalid/revoked, so don't tell a free user to re-authenticate when they hit a 402.
+- When you see a 402: say plainly that the action needs a paid plan, then offer the two real paths — upgrade at `/pricing`, or keep going for free locally (`vidfarm serve`, local render, local decompose, own-provider keys).
+
 ## Automate a template via REST
 
 Templates now have the same job-backed REST pattern as primitives, so a script can run them repeatably without the editor UI:
@@ -172,7 +180,7 @@ The licensed harness also carries the **generative build workflow** guidance (ch
 | `vidfarm login <email>` / `vidfarm verify <email> <code>` | `POST /api/v1/user/request-otp` · `verify-otp` | get an API key |
 | `vidfarm whoami` | `GET /api/v1/user/me` | who am I (also prints the active cost mode) |
 | `vidfarm cost-mode [minimize\|hybrid\|rich-ai]` | (local `~/.vidfarm/cost-mode.json`) | show/set the money-saving preference every billed command respects; no arg = show + explain the three; `--clear` forgets it |
-| `vidfarm wallet [--job <id>\|--tracer <t>] [--limit <n>]` | `GET /api/v1/user/me/wallet` | cost log: balance + lifetime spend + recent charges. `--job <renderJobId>` prints **what that one video cost** (sums its charges); `--tracer <t>` sums a tracer. Cloud-only, paid plan. Aliases: `spend`, `costs` |
+| `vidfarm wallet [--job <id>\|--tracer <t>] [--limit <n>]` | `GET /api/v1/user/me/wallet` | cost log: balance + lifetime spend + recent charges. `--job <renderJobId>` prints **what that one video cost** (sums its charges); `--tracer <t>` sums a tracer. Cloud-only; readable on the free plan too (shows $0.00). Aliases: `spend`, `costs` |
 | `vidfarm provider-keys` / `vidfarm add-provider-key <p> <secret>` | `GET`·`POST /api/v1/user/me/provider-keys` | manage AI keys |
 | `vidfarm upload <file> [--folder <path>]` | presign → S3 PUT → finalize (`.../temporary-files/presign` + `.../temporary-files`) | upload → durable URL (ephemeral, 30-day TTL; prefer `--folder temp` for scratch). Goes direct to S3, so large files (up to **200 MB**) bypass the ~6 MB Lambda body limit |
 | `vidfarm download <url> [dest]` | (streams any URL to disk) | download media from a **direct** media URL. Free — no plan, no job |
@@ -210,9 +218,12 @@ The licensed harness also carries the **generative build workflow** guidance (ch
 `vidfarm cost-mode <minimize|hybrid|rich-ai>` records a single spend preference (in `~/.vidfarm/cost-mode.json`) that every **billed** command honors: `generate`, `music`, `decompose`, `inspiration-decompose`, `create`, `replicate`, `inpaint`, `create-overlay`, `cutout --generate` (only the generation step; a bare `cutout` on an existing graphic is free), and the cloud paths of `render --target cloud`, `tts --cloud`, `stt --cloud`, `remove-greenscreen` (non-`--local`). FREE local engines never gate (`render` local default, `tts --engine local`, `stt --engine whisper`, `remove-greenscreen --local`, `cutout` on a file/url, all the file-editing verbs). `vidfarm media search` (the free stock catalog — pixabay/openverse/iconify music, SFX, images, icons, stock video) is also free and never gates.
 
 - **minimize** — a billed op is **refused** unless you add `--yes`; the error names the free local alternative (which now includes the matching `vidfarm media search` for music/SFX/image/video). Use this to guarantee no surprise AI spend. Before paying to generate music, sound effects, or images, try `vidfarm media search "<meaning>" --type <bgm|sfx|image|vector|icon|video>` first — free royalty-free assets instead of a billed `music`/`generate` call. **Check the keyless sources first — Openverse (CC/CC0 music, SFX, images) and iconify (icons) need no account at all**, so they always work in `minimize`. Photos/vectors/stock-video need a **free Pixabay key** that **may already be saved** — check `vidfarm provider-keys` (or web **Settings → Bring your own keys** / <https://vidfarm.cc/settings/developer>) before assuming a short result means "no key." If absent, save one once: `vidfarm add-provider-key pixabay <key>` (free key from <https://pixabay.com/api/docs/>), the Settings surface, or hand it to the desktop AI agent to run that command.
+- **minimize still gets CUSTOM images — via a free manual generator.** A refused `generate` is not the end of the road. Offer the user the manual loop (ask once, then make it the session default): **you write the prompt → they run it free in <https://meta.ai>, free-tier ChatGPT, or a free Hugging Face image Space (<https://huggingface.co/spaces>) → they hand the PNG back** via `vidfarm put-file ./sheet.png` or web **My Files**. Ask for **one sheet holding every graphic you need**, gridded on a **flat pure-green plate** (`#00FF00`), no text — one round trip instead of N, which saves the user's time and your tokens. Then split it locally for $0: `vidfarm mask ./sheet.png --crop x,y,w,h --flat "#00FF00" --out prop-a.png`, once per element (drop `--flat` and let local ONNX matting handle it if the tool ignored the green background). Full prompt template + loop: recipe `recipes/cutout-graphics-for-explainers.md` (“Free manual image-gen”).
 - **hybrid** *(default recommendation)* — billed ops run but print a one-line cost notice each.
 - **rich-ai** — billed ops run without gating; cost is still printed.
 
+**Narration defaults to the FREE local voice in minimize AND hybrid.** A bare `vidfarm tts "…"` runs the keyless local Kokoro-82M engine in both of those modes — you no longer have to remember `--engine local`. A run **opts out** of that default by asking for a premium voice (`--style`, `--provider`, `--model`, `--own-key`, or a non-Kokoro `--voice` like `alloy`/`Kore`/an ElevenLabs id), by passing `--cloud`/`--engine byok`, or by being in `rich-ai`. If the local engine isn't installed on the machine (it needs `pip install kokoro-onnx soundfile` + a ~340MB model on first use), the run **falls back** to the user's provider key / cloud instead of failing — it prints the reason on stderr so you can tell the user why the voice changed.
+
 Precedence: `--cost-mode <m>` flag → `VIDFARM_COST_MODE` env → the saved `cost-mode` → default (hybrid, flagged as "not set"). When nothing is saved and a billed op runs, the CLI prints a "no preference set — ask the user" nudge instead of silently spending, so the default posture really is *ask before you spend*.
 
 **Agent-memory handoff.** After the user picks, offer to remember it across sessions — but the destination depends on the agent, so ask: Claude Code → `CLAUDE.md` (or its memory dir); Codex / OpenCode / most others → `AGENTS.md`; or a note file the user names. `vidfarm cost-mode <choice>` already persists the devcli-side preference; agent memory is the extra step that survives a fresh checkout. In the **web app UI** there is no memory file — ask each time unless the user states a standing preference for the session.

package/.agents/skills/vidfarm/references/core-workflows.md

@@ -284,7 +284,7 @@ Vidfarm charges directly in USD from the caller's wallet. There are no credits.
 
 - **Wallet top-up** — Stripe checkout via the Settings UI, backed by `POST /settings/wallet/funding-link` (browser session; returns `{ checkout_url, client_reference_id }`)
 - **Balance & history** — shown on the Settings page; browser session also has `GET /u/:customerId/settings/wallet/events`.
-- **Cost log (api-key / agent-friendly)** — `GET /api/v1/user/me/wallet` returns `{ summary, events, next_cursor }`. `summary` carries `balance_usd`, `total_charge_usd` (lifetime spend), `total_funds_added_usd`, and `total_event_count`; each `events[]` entry carries `amount_usd` (**negative = a charge/spend**), `occurred_at_ms`, `cost_center_slug`, `job_id`, and `tracer`. **To answer "how much did this video cost?"** add `?job=<jobId>` (or `?tracer=<tracer>`): the response then includes `filtered_charge_usd` (the summed spend for that render/video) and only the matching events. Paid-plan only — the free tier has no wallet and this route 402s. This is the endpoint to use when a director asks what a video cost; devcli wraps it as `vidfarm wallet` / `vidfarm wallet --job <id>`.
+- **Cost log (api-key / agent-friendly)** — `GET /api/v1/user/me/wallet` returns `{ summary, events, next_cursor }`. `summary` carries `balance_usd`, `total_charge_usd` (lifetime spend), `total_funds_added_usd`, and `total_event_count`; each `events[]` entry carries `amount_usd` (**negative = a charge/spend**), `occurred_at_ms`, `cost_center_slug`, `job_id`, and `tracer`. **To answer "how much did this video cost?"** add `?job=<jobId>` (or `?tracer=<tracer>`): the response then includes `filtered_charge_usd` (the summed spend for that render/video) and only the matching events. Readable on any plan — a free account simply shows a $0.00 balance with no events (paid work is what 402s, not reading your own wallet). This is the endpoint to use when a director asks what a video cost; devcli wraps it as `vidfarm wallet` / `vidfarm wallet --job <id>`.
 - **Cost anchors** (see the Cost spectrum section above for how these combine per approach):
   - Local render on a `vidfarm serve` box (in-process HyperFrames): **free**
   - Cloud render (HyperFrames Lambda fan-out): typically $0.01 – $0.10 per MP4 depending on length/resolution

package/.agents/skills/vidfarm-media/SKILL.md

@@ -12,8 +12,9 @@ One engine and one rule. The engine: `scripts/audio.mjs` turns a neutral `audio_
 Check which engines this run will use and say so in one line — then continue. There is nothing to sign in to and no reason to stop.
 
 - Saved keys: `vidfarm provider-keys`, or env `OPENAI_API_KEY` / `GEMINI_API_KEY` / `OPENROUTER_API_KEY`.
-- Key present → BYOK cloud speech: `vidfarm tts` (promptable voice via `--style`) and `vidfarm stt` (real word timestamps on an OpenAI whisper-1 key).
+- Key present → BYOK cloud speech: `vidfarm tts` (promptable voice via `--style`) and `vidfarm stt` (real word timestamps on an OpenAI whisper-1 key). **But note the cost-mode default below — with a key saved, plain `vidfarm tts` still uses the FREE local voice in `minimize`/`hybrid` unless you ask for a premium one.**
 - No key → keyless local engines: Kokoro-82M TTS (`npx hyperframes tts`) and whisper.cpp timings (`npx hyperframes transcribe`). They download model weights on first use and never need an account.
+- **Cost mode drives the TTS engine.** `minimize` and `hybrid` (and the unset default) → free local Kokoro; `rich-ai`, or any premium ask (`--style`/`--provider`/`--model`/`--own-key`/non-Kokoro `--voice`/`--cloud`) → BYOK or cloud. A missing local engine falls back to BYOK/cloud automatically. Say which voice tier the narration is on the first time you generate it.
 - Don't write keys into a per-repo `.env`. Details: `references/tts.md`.
 
 ## The audio engine
@@ -74,7 +75,7 @@ openrouter key; music always needs ElevenLabs (own key or platform).
 | **Music** (bed, beat, jingle, song, score) | `vidfarm music "upbeat lo-fi beat" --length 30` · `POST /api/v1/primitives/music/generate` | ElevenLabs. `use_wallet_credits` default true (platform key + wallet); `--own-key` = your ElevenLabs key. `music_length_ms` ≤ 300000 (5 min). Place as its own `<audio>` layer ~0.1–0.2 under narration. |
 | **Narration** (default) | `vidfarm tts "…" --cloud` · `POST /api/v1/primitives/audio/speech` | Default = ElevenLabs on the platform key (wallet-billed). Pick a voice with `--voice <voice_id>` (browse below). `--own-key` for your ElevenLabs/BYOK key. Local-first `vidfarm tts` (no `--cloud`) still runs on your env openai/gemini key. |
 | **List voices** | `vidfarm voices` · `GET /api/v1/primitives/audio/voices` | ElevenLabs voice catalog (voice_id, name, labels, preview_url). Default = platform account; `--own-key` / `?use_wallet_credits=false` = the customer's ElevenLabs account. **Default a sensible voice AND tell the user they can pick from many.** |
-| Narration, zero keys | `npx hyperframes tts "…" -v af_heart --json` | Kokoro-82M, local, WAV + duration in JSON |
+| Narration, zero keys / cost-saving | `vidfarm tts "…" --out narration.wav` (free local by default in `minimize`/`hybrid`) · or `npx hyperframes tts "…" -v af_heart --json` | Kokoro-82M, local, WAV + duration in JSON. Fixed voice presets, no `--style`. |
 | **Transcript + SRT** | `vidfarm stt <file\|url> --cloud` · `POST /api/v1/primitives/audio/transcribe` | Default = ElevenLabs Scribe (native diarization + real word timestamps), wallet-billed. `--own-key`/BYOK: gemini labels speakers, openai/whisper-1 gives real word timings. |
 | Reword existing narration in the (approximate) original voice | `POST /api/v1/primitives/audio/regenerate-speech` | Listens, profiles the speaker (needs a Gemini key), rewords, regenerates with the closest preset voice + matched style. Approximation, never a clone. Details: `references/tts.md` |
 | Word timings, zero keys | `npx hyperframes transcribe <audio> --json` | whisper.cpp; writes word-level `transcript.json` |

package/.agents/skills/vidfarm-media/references/tts.md

@@ -52,6 +52,7 @@ vidfarm voices --own-key       # the customer's own ElevenLabs account voices
 vidfarm tts "Welcome back to the channel" --style "calm, warm bedtime narrator" --voice alloy --out narration.mp3 --json
 ```
 
+- **Cost-mode default: FREE local voice.** In cost mode `minimize` AND `hybrid` (including the unset default), a bare `vidfarm tts "…"` runs the keyless local Kokoro-82M engine — no `--engine local` needed. Opt out by asking for a premium voice (`--style`, `--provider`, `--model`, `--own-key`, or a non-Kokoro `--voice` such as `alloy`/`Kore`/an ElevenLabs id), by passing `--cloud` or `--engine byok`, or by being in `rich-ai`. If the local engine isn't installed the run **falls back** to the BYOK/cloud path below rather than failing, printing the reason on stderr.
 - **Local-first**: runs directly on your env key (no cloud job, no wallet). `--cloud` is the explicit backup — an async platform job on your saved provider keys via `POST /api/v1/primitives/audio/speech` (poll the returned `job_id`).
 - **Promptable voice**: `--style` (alias `--instructions`) directs tone/pacing/accent/emotion/persona — "excited sports announcer", "whispered documentary aside".
 - **Voices** (provider inferred from the voice when `--provider` is omitted; wrong-family voices are rejected with a clear error):

package/SKILL.director.md

@@ -83,11 +83,13 @@ Do not use this skill to author new templates from scratch, deploy platform infr
 
 Vidfarm work can burn real AI credits on the user's wallet / provider keys. **Save them money by default.** Before the first billed step of a session, ask the user which spend posture they want, and explain it in one plain line each:
 
-- **minimize** — cheapest. Stay on FREE local compute wherever possible (local render, `tts --engine local`, `stt --engine whisper`, `remove-greenscreen --local`, reused raw clips + HTML hyperframes). For assets, reach for the **free stock catalog** before paying to generate anything — `vidfarm media search "<meaning>" --type <bgm|sfx|image|vector|icon|video>` pulls royalty-free, commercial-safe music/sound-effects/images/icons/stock-video from pixabay/openverse/iconify at $0 instead of billing AI music/image generation (see the `vidfarm-media` skill). No surprise AI spend.
+- **minimize** — cheapest. Stay on FREE local compute wherever possible (local render, local TTS — `vidfarm tts` already defaults to the free local Kokoro voice in this mode — `stt --engine whisper`, `remove-greenscreen --local`, reused raw clips + HTML hyperframes). For assets, reach for the **free stock catalog** before paying to generate anything — `vidfarm media search "<meaning>" --type <bgm|sfx|image|vector|icon|video>` pulls royalty-free, commercial-safe music/sound-effects/images/icons/stock-video from pixabay/openverse/iconify at $0 instead of billing AI music/image generation (see the `vidfarm-media` skill). No surprise AI spend.
   - **Check the keyless sources first — Openverse and iconify.** Openverse (CC/CC0 **music, SFX, and images**) and iconify (**icons**) need **no account or key at all**, so they always work in `minimize` mode. Prefer them for BGM, sound effects, icons, and CC imagery before anything else.
   - **Pixabay key** unlocks the photos/vectors/stock-video slots (music/SFX/icons/CC images are keyless). It's a **free** stock-media key, not an AI key. Don't assume it's missing when a search comes up short — it **may already be saved**: check `vidfarm provider-keys` (or the web app's **Settings → Bring your own keys** / <https://vidfarm.cc/settings/developer>). If it isn't, the user grabs a free one at <https://pixabay.com/api/docs/> and saves it once — `vidfarm add-provider-key pixabay <key>`, the Settings surface, or by handing the key to their desktop AI agent to run that command. After it's saved, cost-mode `minimize` sourcing works end-to-end at $0.
+  - **You can still get CUSTOM art in `minimize` — hand the prompt to the user and let a free image generator do it.** Stock and `mask` only cover art that already exists somewhere; when the video genuinely needs a bespoke graphic, **don't conclude "we can't" and don't quietly bill `generate`**. Write the prompt and ask the user to paste it into a **free** image generator — <https://meta.ai>, free-tier ChatGPT, or a free Hugging Face image Space (<https://huggingface.co/spaces>) — then hand the PNG back with `vidfarm put-file` (or drag it into **My Files** in the web app). $0, zero wallet spend. Full loop + the prompt template: **“Free manual image-gen”** below.
 - **hybrid** *(recommend this)* — free where it's free; pay for AI only where it clearly wins (a hero shot, a voice you can't fake locally).
-  - **Free-TTS draft, premium-TTS final (opt-in — offer it, don't assume it).** Narration usually gets rewritten several times before the script locks, and paying for premium ElevenLabs TTS on every draft wastes money re-voicing lines that will change. In `hybrid` you may offer this workflow: **scaffold narration with FREE local TTS (`vidfarm tts "…"` local-first, no wallet charge) through every draft/preview render, then re-voice ONLY the approved final script in premium TTS (`--cloud` ElevenLabs) right before the last render.** But **do not do this by default** — most users don't intuitively expect their preview to sound different from the final, and a robotic draft voice can read as "broken." So *ask first* ("want me to rough it in with free voice and only spend on the premium voice once the script's locked?"), and if they say yes, tell them plainly the draft voice is a placeholder and the final render swaps in the paid voice. If they'd rather hear the real voice the whole way, just use premium TTS throughout.
+  - **Narration starts free here too.** In BOTH `minimize` and `hybrid`, a bare `vidfarm tts "…"` runs the keyless local Kokoro-82M voice — no flag needed. Ask for a premium voice explicitly (`--style`, `--voice <provider preset>`, `--provider`, `--own-key`, or `--cloud`) when the piece needs one, and say so out loud so the user knows what they're paying for. `rich-ai` skips the free default entirely. If the local engine isn't installed on their machine, the run falls back to their provider key / cloud automatically and prints why.
+  - **Free-TTS draft, premium-TTS final — now the default shape, so SAY it out loud.** Narration gets rewritten several times before the script locks, and paying for premium ElevenLabs TTS on every draft wastes money re-voicing lines that will change. The CLI defaults to the free local voice in `hybrid`, so drafts are free automatically — but a robotic draft voice reads as "broken" to a user who wasn't told. **Tell them the first time you generate narration**: the draft voice is the free local one, and you'll re-voice the approved final script in premium TTS (`--cloud` ElevenLabs, or `--style`/`--voice` on their own key) right before the last render. If they'd rather hear the real voice the whole way, pass the premium flags from the start — and if the piece is voice-led (a testimonial, a brand spot where the voice IS the product), offer premium up front instead of drafting free.
 - **rich-ai** — best quality; use AI image/video/voice/music freely.
 
 Then **ask whether to remember the choice**, and save it where it will actually be read next time:
@@ -143,6 +145,18 @@ Present both harnesses to the director, recommend (A) unless they've asked for p
 
 **Cost-saving move — mask illustrations OUT of a source image the director already has.** (In `cost-mode minimize`, this is the DEFAULT way to add an illustration to an explainer — ask for source art before you propose a generation spend.) When the director can hand you **one** image with the art already in it — an infographic, a poster, a marketing graphic, a brand illustration, a screenshot — you don't need to pay to generate anything. `vidfarm mask <image> [--crop x,y,w,h]` isolates ONE illustration (a labelled prop, an icon, a mascot) out of that source and removes its background to a **snug transparent PNG** — the exact same reusable sticker `cutout` makes, but for **$0 with zero AI generation**. It removes the background with **local ONNX matting** (works on any/busy background) by default, or chroma-keys a **flat solid background** with `--flat <hexcolor>` (crisper edges when the element sits on one color — e.g. the cream paper behind an infographic's icons). Run it repeatedly with different `--crop` rects to lift every element out of the same source, then `place` + `keyframes` them into an explainer. **Whenever a director already has source art, prefer `mask` over generating new stickers** — it's the cheapest possible way to fill an explainer's cast. Same recipe: `recipes/cutout-graphics-for-explainers.md` (“Mask from an image you already have”).
 
+**Free manual image-gen — custom art in `minimize` mode for $0, on someone else's tokens.** `mask` only works when the art already exists. When the video needs a **bespoke** graphic and cost mode is `minimize` (or the user said "no spend"), the answer is **not** "we can't" and **not** a silent billed `generate` — it's a **manual handoff**: you write the prompt, the user runs it in a **free** image generator, they hand the PNG back.
+
+- **Offer it once, up front, and let them set the default.** The first time custom art is needed in a cost-saving session, ask plainly: *"Want me to default to free manual image generation — I write the prompt, you paste it into a free tool and drop the image back — or would you rather I spend credits on `vidfarm generate` for these?"* Honor the answer for the rest of the session (and offer to remember it alongside the cost mode). Don't re-ask per image once they've chosen; just say "here's the next prompt."
+- **Free generators to name:** <https://meta.ai> (no cost, generous), free-tier **ChatGPT** (image gen included), or a free image **Hugging Face Space** (<https://huggingface.co/spaces> — search "flux" or "sdxl"). Any of them; the user picks whichever they're already signed into.
+- **Ask for ONE sheet, not N images — this is the token-efficiency trick.** Each round trip costs the user a manual step and costs you tokens re-reading a new file. So request **one image containing every graphic the scene needs**, laid out as a grid of separate objects on a **flat pure-green background** (`#00FF00`), well spaced, none overlapping or touching an edge. One generation, one hand-back, then you lift each element out locally for free.
+- **Then cut it up locally, for $0:** `vidfarm put-file ./sheet.png` to bring it in, then `vidfarm mask ./sheet.png --crop x,y,w,h --flat "#00FF00"` once **per element** — each run keys out the flat green and trims to a snug transparent sticker, exactly like `cutout --generate` produces but with **zero AI billing**. `place` + `keyframes` them as usual. (`--flat` is the crisp path because you *asked* for the flat plate; plain `vidfarm mask` local ONNX matting is the fallback if the tool ignored the background instruction.)
+- **Prompt template — give the user this, filled in:**
+
+  > A flat vector illustration sheet on a solid pure green background (#00FF00), no gradient, no shadow, no text, no labels. Arranged in a 3×2 grid with generous even spacing, each object fully separated and not touching any other object or the image edge: (1) `<object one>`, (2) `<object two>`, (3) `<object three>`, … Simple flat shapes, 2–3 flat colors, minimal detail, consistent style and line weight across all objects, front-facing, centered in its own cell. Square image, high resolution.
+
+  Keep it to the explainer house style above (flat vector, 2–3 colors, **no baked-in text** — text never keys or scales cleanly; add copy as a caption/HTML layer instead). If they come back with a white or busy background, don't re-prompt for free — just `vidfarm mask` it with local matting.
+
 **Be generous to free-tier / no-account users — on the cheap REBUILD path.** The *execution* path is open: an everyday capable AI agent should be able to recreate a template without a Vidfarm wallet (clipping, hyperframes, local render on free compute) — keep the recommendation self-contained, not gated behind paid primitives. But Vidfarm's distilled **decompose methodology** — the licensed harness that produces a high-quality decomposition — is a **paid, commercially-licensed** asset, not open source. Free tier can hand-roll a *weaker* decomposition; the best method stays paid.
 
 **Free tier vs. paid — who does the decomposition, and on whose tokens.** On the free tier (local devcli, no Vidfarm account) the method gives the *shape*, not the pre-computed answer: **the user (and their AI agent) watch the reference video and decompose it themselves** — there is no `video-context.json` / `editor-harness.json` / `scene-annotations.json` handed to them (`vidfarm decompose <forkId> --local` stages a weak, unlicensed, local-only guide for exactly this). **Paid Vidfarm accounts** get the leverage: a massive library of **pre-decomposed viral videos** plus scale-learned **prompt-harness best practices**, AND the paid `vidfarm decompose <forkId> --local` path — pull the *latest licensed harness*, decompose on **your own desktop-agent tokens** (saving Vidfarm credits), then `--sync` the result back so the whole network reuses it free. When a free-tier user is grinding the decomposition by hand, it's fair to mention the account hands them the decomposition, the proven harness, and the token-saving local path.
@@ -539,7 +553,7 @@ Vidfarm charges directly in USD from the caller's wallet. There are no credits.
 
 - **Wallet top-up** — Stripe checkout via the Settings UI, backed by `POST /settings/wallet/funding-link` (browser session; returns `{ checkout_url, client_reference_id }`)
 - **Balance & history** — shown on the Settings page; browser session also has `GET /u/:customerId/settings/wallet/events`.
-- **Cost log (api-key / agent-friendly)** — `GET /api/v1/user/me/wallet` returns `{ summary, events, next_cursor }`. `summary` carries `balance_usd`, `total_charge_usd` (lifetime spend), `total_funds_added_usd`, and `total_event_count`; each `events[]` entry carries `amount_usd` (**negative = a charge/spend**), `occurred_at_ms`, `cost_center_slug`, `job_id`, and `tracer`. **To answer "how much did this video cost?"** add `?job=<jobId>` (or `?tracer=<tracer>`): the response then includes `filtered_charge_usd` (the summed spend for that render/video) and only the matching events. Paid-plan only — the free tier has no wallet and this route 402s. This is the endpoint to use when a director asks what a video cost; devcli wraps it as `vidfarm wallet` / `vidfarm wallet --job <id>`.
+- **Cost log (api-key / agent-friendly)** — `GET /api/v1/user/me/wallet` returns `{ summary, events, next_cursor }`. `summary` carries `balance_usd`, `total_charge_usd` (lifetime spend), `total_funds_added_usd`, and `total_event_count`; each `events[]` entry carries `amount_usd` (**negative = a charge/spend**), `occurred_at_ms`, `cost_center_slug`, `job_id`, and `tracer`. **To answer "how much did this video cost?"** add `?job=<jobId>` (or `?tracer=<tracer>`): the response then includes `filtered_charge_usd` (the summed spend for that render/video) and only the matching events. Readable on any plan — a free account simply shows a $0.00 balance with no events (paid work is what 402s, not reading your own wallet). This is the endpoint to use when a director asks what a video cost; devcli wraps it as `vidfarm wallet` / `vidfarm wallet --job <id>`.
 - **Cost anchors** (see the Cost spectrum section above for how these combine per approach):
   - Local render on a `vidfarm serve` box (in-process HyperFrames): **free**
   - Cloud render (HyperFrames Lambda fan-out): typically $0.01 – $0.10 per MP4 depending on length/resolution
@@ -1162,6 +1176,14 @@ The web copilot exposes the same three as `browse_files action=rename|move|copy`
 
 In the editor web copilot the same filesystem is exposed via the **`browse_files` tool** (`action=list` / `action=search` / `action=read` / `action=write` / `action=annotate` / `action=move` / `action=copy` / `action=rename`), so the copilot follows the identical reasoning: search or list to find the right offer's folder, then read an asset — or `write` a text doc (About.md, awareness-levels.md, …) or import a media URL (`source_url`) into that folder, annotating anything worth finding again. `action=rename`/`move`/`copy` reorganize the tree (see the three routes above) — use them to keep character folders and asset names tidy. `browse_files list` defaults to `path='/'` when `path` is omitted, so the `/raws` (hunted raws) and `/temp` (scratch) roots surface alongside the My Files folders instead of being hidden; a `/raws` listing also accepts a `content_type` filter (exact shot-kind — `talking_head`, `b_roll`, `product_shot`, `screen_recording`, …), and every listing paginates via `offset` / `limit`. The devcli equivalents are `vidfarm files [--search]` / `get-file` / `put-file [--notes]` / `annotate-file` / `directory rename|move|copy`.
 
+## Free-plan API keys — valid everywhere, gated per route
+
+A free (unpaid) account's `vidfarm-api-key` is a **real key**: it authenticates on every route, and `vidfarm login` / `whoami` work normally. Plan is enforced per route, not at auth:
+
+- **Free on any plan** — `GET /api/v1/user/me` (identity), `/api/v1/user/me/provider-keys` (bring-your-own-key), `GET /api/v1/user/me/wallet` (reads $0.00), `GET /api/v1/user/me/tasks`, and the `/api/v1/user/me/directory*` file-tree routes (browse, search, folder, rename, move, copy, shares).
+- **Paid routes answer `402 { type: "upgrade_required", upgrade_url: "/pricing", run_locally_url: "/help" }`** — never a 401. A `401` means the key is missing/invalid/revoked, so don't tell a free user to re-authenticate when they hit a 402.
+- When you see a 402: say plainly that the action needs a paid plan, then offer the two real paths — upgrade at `/pricing`, or keep going for free locally (`vidfarm serve`, local render, local decompose, own-provider keys).
+
 ## Automate a template via REST
 
 Templates now have the same job-backed REST pattern as primitives, so a script can run them repeatably without the editor UI:
@@ -1336,7 +1358,7 @@ The licensed harness also carries the **generative build workflow** guidance (ch
 | `vidfarm login <email>` / `vidfarm verify <email> <code>` | `POST /api/v1/user/request-otp` · `verify-otp` | get an API key |
 | `vidfarm whoami` | `GET /api/v1/user/me` | who am I (also prints the active cost mode) |
 | `vidfarm cost-mode [minimize\|hybrid\|rich-ai]` | (local `~/.vidfarm/cost-mode.json`) | show/set the money-saving preference every billed command respects; no arg = show + explain the three; `--clear` forgets it |
-| `vidfarm wallet [--job <id>\|--tracer <t>] [--limit <n>]` | `GET /api/v1/user/me/wallet` | cost log: balance + lifetime spend + recent charges. `--job <renderJobId>` prints **what that one video cost** (sums its charges); `--tracer <t>` sums a tracer. Cloud-only, paid plan. Aliases: `spend`, `costs` |
+| `vidfarm wallet [--job <id>\|--tracer <t>] [--limit <n>]` | `GET /api/v1/user/me/wallet` | cost log: balance + lifetime spend + recent charges. `--job <renderJobId>` prints **what that one video cost** (sums its charges); `--tracer <t>` sums a tracer. Cloud-only; readable on the free plan too (shows $0.00). Aliases: `spend`, `costs` |
 | `vidfarm provider-keys` / `vidfarm add-provider-key <p> <secret>` | `GET`·`POST /api/v1/user/me/provider-keys` | manage AI keys |
 | `vidfarm upload <file> [--folder <path>]` | presign → S3 PUT → finalize (`.../temporary-files/presign` + `.../temporary-files`) | upload → durable URL (ephemeral, 30-day TTL; prefer `--folder temp` for scratch). Goes direct to S3, so large files (up to **200 MB**) bypass the ~6 MB Lambda body limit |
 | `vidfarm download <url> [dest]` | (streams any URL to disk) | download media from a **direct** media URL. Free — no plan, no job |
@@ -1374,9 +1396,12 @@ The licensed harness also carries the **generative build workflow** guidance (ch
 `vidfarm cost-mode <minimize|hybrid|rich-ai>` records a single spend preference (in `~/.vidfarm/cost-mode.json`) that every **billed** command honors: `generate`, `music`, `decompose`, `inspiration-decompose`, `create`, `replicate`, `inpaint`, `create-overlay`, `cutout --generate` (only the generation step; a bare `cutout` on an existing graphic is free), and the cloud paths of `render --target cloud`, `tts --cloud`, `stt --cloud`, `remove-greenscreen` (non-`--local`). FREE local engines never gate (`render` local default, `tts --engine local`, `stt --engine whisper`, `remove-greenscreen --local`, `cutout` on a file/url, all the file-editing verbs). `vidfarm media search` (the free stock catalog — pixabay/openverse/iconify music, SFX, images, icons, stock video) is also free and never gates.
 
 - **minimize** — a billed op is **refused** unless you add `--yes`; the error names the free local alternative (which now includes the matching `vidfarm media search` for music/SFX/image/video). Use this to guarantee no surprise AI spend. Before paying to generate music, sound effects, or images, try `vidfarm media search "<meaning>" --type <bgm|sfx|image|vector|icon|video>` first — free royalty-free assets instead of a billed `music`/`generate` call. **Check the keyless sources first — Openverse (CC/CC0 music, SFX, images) and iconify (icons) need no account at all**, so they always work in `minimize`. Photos/vectors/stock-video need a **free Pixabay key** that **may already be saved** — check `vidfarm provider-keys` (or web **Settings → Bring your own keys** / <https://vidfarm.cc/settings/developer>) before assuming a short result means "no key." If absent, save one once: `vidfarm add-provider-key pixabay <key>` (free key from <https://pixabay.com/api/docs/>), the Settings surface, or hand it to the desktop AI agent to run that command.
+- **minimize still gets CUSTOM images — via a free manual generator.** A refused `generate` is not the end of the road. Offer the user the manual loop (ask once, then make it the session default): **you write the prompt → they run it free in <https://meta.ai>, free-tier ChatGPT, or a free Hugging Face image Space (<https://huggingface.co/spaces>) → they hand the PNG back** via `vidfarm put-file ./sheet.png` or web **My Files**. Ask for **one sheet holding every graphic you need**, gridded on a **flat pure-green plate** (`#00FF00`), no text — one round trip instead of N, which saves the user's time and your tokens. Then split it locally for $0: `vidfarm mask ./sheet.png --crop x,y,w,h --flat "#00FF00" --out prop-a.png`, once per element (drop `--flat` and let local ONNX matting handle it if the tool ignored the green background). Full prompt template + loop: recipe `recipes/cutout-graphics-for-explainers.md` (“Free manual image-gen”).
 - **hybrid** *(default recommendation)* — billed ops run but print a one-line cost notice each.
 - **rich-ai** — billed ops run without gating; cost is still printed.
 
+**Narration defaults to the FREE local voice in minimize AND hybrid.** A bare `vidfarm tts "…"` runs the keyless local Kokoro-82M engine in both of those modes — you no longer have to remember `--engine local`. A run **opts out** of that default by asking for a premium voice (`--style`, `--provider`, `--model`, `--own-key`, or a non-Kokoro `--voice` like `alloy`/`Kore`/an ElevenLabs id), by passing `--cloud`/`--engine byok`, or by being in `rich-ai`. If the local engine isn't installed on the machine (it needs `pip install kokoro-onnx soundfile` + a ~340MB model on first use), the run **falls back** to the user's provider key / cloud instead of failing — it prints the reason on stderr so you can tell the user why the voice changed.
+
 Precedence: `--cost-mode <m>` flag → `VIDFARM_COST_MODE` env → the saved `cost-mode` → default (hybrid, flagged as "not set"). When nothing is saved and a billed op runs, the CLI prints a "no preference set — ask the user" nudge instead of silently spending, so the default posture really is *ask before you spend*.
 
 **Agent-memory handoff.** After the user picks, offer to remember it across sessions — but the destination depends on the agent, so ask: Claude Code → `CLAUDE.md` (or its memory dir); Codex / OpenCode / most others → `AGENTS.md`; or a note file the user names. `vidfarm cost-mode <choice>` already persists the devcli-side preference; agent memory is the extra step that survives a fresh checkout. In the **web app UI** there is no memory file — ask each time unless the user states a standing preference for the session.
@@ -1918,7 +1943,7 @@ The mechanical trio — **generate on a chroma plate → key it out → trim to
 
 **Illustrations default to simplicity.** Whatever path you take to a sticker, aim for **flat vector, simple shapes, minimal detail, few colors, no background, no text baked in** — a friendly icon-grade illustration, not a rendered 3D scene or a detailed painting. Simple art keys cleanly, trims tight, scales without mush, animates readably at 9:16, and stays on-style across a whole cast. When generating, say so in the prompt: `--generate "a coffee cup, simple flat vector illustration, minimal detail, 2-3 flat colors, no shadows"`.
 
-**In cost-saving mode, don't generate illustrations at all — mask them out of images the director already has.** If `vidfarm cost-mode` is `minimize` (or the director says "without burning credits"), the default for adding an illustration is `vidfarm mask <their-image> --crop …` — lifting art out of an infographic, poster, deck slide, brand sheet, or screenshot for **$0 and zero AI calls**. Ask for source art before you ask for a generation budget; the guided loop is **"Mask from an image you already have"** below.
+**In cost-saving mode, don't generate illustrations at all — mask them out of images the director already has.** If `vidfarm cost-mode` is `minimize` (or the director says "without burning credits"), the default for adding an illustration is `vidfarm mask <their-image> --crop …` — lifting art out of an infographic, poster, deck slide, brand sheet, or screenshot for **$0 and zero AI calls**. Ask for source art before you ask for a generation budget; the guided loop is **"Mask from an image you already have"** below. **If no source art exists and the graphic must be custom, you still don't have to spend** — hand the director a prompt for a **free** image generator (meta.ai / free ChatGPT / a Hugging Face Space) and cut the returned sheet into stickers locally: **"Free manual image-gen"** below.
 
 ### The guided sequence (prompt harness)
 
@@ -1995,6 +2020,34 @@ This is the biggest cost saver in the whole explainer workflow: **if the directo
 
 Everything here is **local and free** — crop, matte/key, and trim are all `ffmpeg` + the bundled ONNX engine. No wallet, no account, no generation call.
 
+### Free manual image-gen (the $0 path when NO source art exists)
+
+`mask` needs art that already exists. When the explainer needs a **custom** graphic and the director is in `cost-mode minimize` (or just said "don't spend"), there's still a $0 path: **you write the prompt, the director runs it in a free image generator, they hand the image back, you cut it up locally.** The AI spend moves to a free third-party tool; Vidfarm's wallet is never touched.
+
+**Offer it once, then make it the default.** The first time custom art comes up in a cost-saving session, ask: *"Want me to default to free manual image generation — I write the prompt, you paste it into a free tool and drop the PNG back — or should I spend credits on `vidfarm generate`?"* Whatever they pick, hold it for the session (and offer to remember it next to the cost mode). After that, don't re-ask per graphic — just hand over the next prompt.
+
+**Free generators to name:** <https://meta.ai>, free-tier **ChatGPT**, or a free image **Hugging Face Space** (<https://huggingface.co/spaces>, search "flux" / "sdxl"). Let the director use whichever they're already logged into.
+
+**Ask for ONE sheet of many graphics, not many images.** Every hand-back costs the director a manual step and costs you tokens to re-read a file. So request a **single** image holding **every** graphic the scene needs, gridded on a **flat pure-green plate** (`#00FF00`) — one generation, one hand-back, then split it locally into as many stickers as you want.
+
+**The loop:**
+
+1. **List the cast.** Name every illustration the explainer needs, in the house style (flat vector, 2–3 flat colors, no baked-in text).
+2. **Hand the director this prompt**, filled in:
+   > A flat vector illustration sheet on a solid pure green background (#00FF00), no gradient, no shadow, no text, no labels. Arranged in a 3×2 grid with generous even spacing, each object fully separated and not touching any other object or the image edge: (1) `<object one>`, (2) `<object two>`, (3) `<object three>`, … Simple flat shapes, 2–3 flat colors, minimal detail, consistent style and line weight across all objects, front-facing, centered in its own cell. Square image, high resolution.
+
+   **No text in the image, ever** — rendered text never keys or scales cleanly; add copy as a caption or HTML layer instead.
+3. **Bring the sheet in:** `vidfarm put-file ./sheet.png` (or drag into **My Files** in the web app).
+4. **Split it into stickers — one `mask` per element, all free:**
+   ```
+   vidfarm mask ./sheet.png --crop 4%,4%,29%,45%  --flat "#00FF00" --out prop-a.png
+   vidfarm mask ./sheet.png --crop 36%,4%,29%,45% --flat "#00FF00" --out prop-b.png
+   ```
+   `--flat "#00FF00"` is the crisp path precisely because you *asked* for the flat green plate. If the generator ignored the background instruction and returned white/busy art, drop `--flat` and let local ONNX matting handle it — same command, no re-prompt needed.
+5. **Stage, animate, render** exactly as in Steps 3–5 above. A manually-generated sticker is indistinguishable from a `cutout --generate` one on the timeline.
+
+**Grid sizing note:** a 3×2 grid on a square image puts each cell at roughly `33%` wide × `50%` tall; start crops a couple of percent inside the cell and use `--keep-region` on the first run to verify before masking the rest.
+
 ### Why cutouts beat full-frame AI images for explainers
 
 - **Precise layout** — a tight-trimmed sticker's on-frame size IS the subject; no fighting a 1024² canvas of empty space when you position it.

package/SKILL.md

@@ -26,7 +26,9 @@ Default to the cheapest approach that works, and surface cost before any expensi
 
 ## Save the user money — ask their cost mode first
 
-Before spending any AI credits, ask which posture they want and explain each simply: **minimize** (cheapest — free local compute, no surprise AI spend), **hybrid** (recommended — free where free, pay AI only where it clearly wins), or **pure-ai** (best quality — AI media freely). With the devcli, persist it via `vidfarm cost-mode <choice>` (billed commands then respect it; `minimize` refuses billed ops without `--yes`). Then ask if they want it remembered across sessions — save to the agent's own memory (Claude Code → `CLAUDE.md`; Codex/OpenCode/others → `AGENTS.md`; or a note they name), asking first. In the **web app UI** there's no memory file to write, so just ask each time unless they state a standing preference. Full guidance is in the director skill.
+Before spending any AI credits, ask which posture they want and explain each simply: **minimize** (cheapest — free local compute, no surprise AI spend), **hybrid** (recommended — free where free, pay AI only where it clearly wins), or **pure-ai** (best quality — AI media freely). With the devcli, persist it via `vidfarm cost-mode <choice>` (billed commands then respect it; `minimize` refuses billed ops without `--yes`). **Narration follows the mode automatically**: in `minimize` and `hybrid`, plain `vidfarm tts "…"` uses the free local Kokoro voice — ask for a premium voice explicitly (`--style`/`--voice`/`--provider`/`--cloud`) when the piece needs one, and tell the user which tier they're hearing. Then ask if they want it remembered across sessions — save to the agent's own memory (Claude Code → `CLAUDE.md`; Codex/OpenCode/others → `AGENTS.md`; or a note they name), asking first. In the **web app UI** there's no memory file to write, so just ask each time unless they state a standing preference. Full guidance is in the director skill.
+
+**Even in `minimize`, you can still get custom images — for free, manually.** When stock and masking existing art won't cover a bespoke graphic, don't say "not possible" and don't quietly spend: ask the user once whether to **default to free manual image generation**, then write the prompt and have them run it in <https://meta.ai>, free-tier ChatGPT, or a free Hugging Face image Space, and hand the PNG back (`vidfarm put-file`, or drag into My Files). Ask for **one image containing every graphic you need**, gridded on a **flat pure-green background** (`#00FF00`) with no text — one round trip instead of many, which is the token-efficient move. Then split it locally at $0: `vidfarm mask ./sheet.png --crop x,y,w,h --flat "#00FF00"` once per element gives you snug transparent stickers. Prompt template + full loop in the director skill.
 
 ## Get hands-on now — pick your surface
 
@@ -62,6 +64,7 @@ This page does not contain the workflows. Pull the full director skill and load
   - `onboarding` — "help me figure out what to make" interview
   - `primitives` — TTS, STT, music, overlays, background removal, product placement
   - `rest-api` — complete REST map (load only for direct HTTP integration)
+- **Updating Vidfarm:** `https://vidfarm.cc/update.md` — **fetch this file whenever the user asks to update/upgrade Vidfarm**, says their skill or devcli is out of date, or hits a command/route that "used to work". It's the clean uninstall→reinstall runbook for BOTH halves (the `vidfarm` skill pack and `@officexapp/vidfarm-devcli`), and it carries the current breaking-change notes — read it live rather than improvising an `npm i -g`, because the notes are exactly what a blind reinstall misses.
 - **Skill pack index (what else is installable):** `https://vidfarm.cc/skill-pack/index.json`
 - **Platform architecture (how it all fits together):** `https://vidfarm.cc/skill/vidfarm-platform`
 
@@ -76,6 +79,7 @@ For composition *authoring* craft (motion, keyframes, scene design), Vidfarm shi
 4b. **"Download this video/audio from <a website URL>"** → Vidfarm fetches it for you on a **paid plan**: `POST /api/v1/primitives/videos/download` (or `/audio/download`); devcli `vidfarm download-video <url>` / `vidfarm download-audio <url>`. Works on YouTube, TikTok, Instagram, X, and other supported posts; returns a durable Vidfarm file (photo/carousel posts → an ordered slideshow). **Free plan gets a 402 — don't call it. Tell the user (or, with browser automation, do it yourself) to open the URL in Chrome and download it from the page, then `vidfarm put-file ./the-file.mp4` to bring it in for $0.** Never answer "I can't download that." Details: `references/assets-and-sourcing.md`.
 5. "Script / batch / render loop" → `references/automation-and-local-dev.md`
 6. "I need TTS / music / captions / background removal" → `references/primitives.md`
+7. **"Update / upgrade vidfarm"** (or anything that smells like a stale install — a missing command, a 404 on a documented route, a version mismatch) → fetch <https://vidfarm.cc/update.md> and follow it. Update the skill pack and the devcli **together**; updating one alone is the usual cause of "the skill says to do X but it fails."
 
 ## Non-negotiables
 

package/clipper.md

@@ -0,0 +1,424 @@
+---
+name: clipper
+description: Run an agentic clipper loop — earn side income by having your own AI agent work a DollarPlatoon gig (vending machine) unattended on your own computer. Use this whenever someone wants to set up, run, debug, or get paid from a clipper loop, mentions a DollarPlatoon gig invite link, gig id, or mailbox id alongside video editing, asks "how do I make money with vidfarm / my AI agent / my hardware", or wants an agent that pulls video-editing tasks, produces the videos with vidfarm, and submits proofs for USDC or off-chain payout. Served at vidfarm.cc/clipper.md. Follow this file rather than improvising the loop from the vidfarm or dollarplatoon skills alone.
+---
+
+# /clipper — the agentic clipper loop
+
+**You are the user's own AI agent, running on their computer.** Your job is to set up and then
+run a loop that earns them money:
+
+```
+DollarPlatoon gig  ──►  a video-editing task  ──►  you make the video with Vidfarm
+        ▲                                                       │
+        │                                                       ▼
+   payout (USDC or off-chain)  ◄──  proof = a public vidfarm.cc share URL
+```
+
+The user is most likely a college student or a working video editor — often in the
+Philippines — earning side income from hardware they already own. Assume they are smart but
+not a developer. **Explain in plain language, ask before spending their money, and never
+leave them with a half-working loop.**
+
+Read this whole file before doing anything. Then follow the phases in order.
+
+---
+
+## What you need before you start
+
+| Thing | How to get it |
+|---|---|
+| `VIDFARM_API_KEY` | `.env`, or `vidfarm login <email>` (email code). Get one at <https://vidfarm.cc> |
+| `DOLLARPLATOON_API_KEY` | `.env`, or <https://dollarplatoon.com/client/settings>. Alias: `DOLLAR_PLATOON_API_KEY` |
+| A gig | Any one of three: an invite URL like `https://dollarplatoon.com/gig/GIG_01HX.../join?invite=a1b2c3d4e5f6`, a bare gig id (`GIG_01HX…`), or — if they already joined — their **mailbox id** for that gig |
+| The Vidfarm skill | Fetch <https://vidfarm.cc/skill.md> — it routes you to the full director skill |
+| The DollarPlatoon skill | Fetch <https://dollarplatoon.com/skill.md> — the full REST reference |
+
+**Fetch both skill files now.** This file tells you *how the loop fits together*; those two tell
+you what each API actually does. Do not guess endpoints you haven't read.
+
+Keys live in `.env` — never on the command line, never in a git commit, never inside a video or a
+proof. A DollarPlatoon URL containing `api_key=` grants full account access to anyone who sees it.
+
+If a key is missing, ask for it. Don't build around it.
+
+---
+
+## Phase 1 — Ask the user for their config, and tell them the defaults
+
+Before writing a single file, have one short conversation. **Show them the defaults and let them
+change what they care about.** Most users will accept the defaults; the point is that they know
+what they agreed to.
+
+Ask these, in one message, with the defaults stated:
+
+1. **Which gig?** "Paste the DollarPlatoon gig invite link — or, if you've already joined, just the
+   gig id or your mailbox id for it." — required, no default. If they have none of the three but do
+   have an API key, run `GET /mailboxes/mine` and show them the gigs they're already in to pick from.
+2. **Cost mode — default `minimize`.** Cheapest: free local rendering on their machine, no surprise
+   AI spend. `hybrid` pays for AI only where it clearly wins. `rich-ai` is best quality and most
+   expensive. Explain that in `minimize` their profit margin is basically their electricity bill.
+3. **Free manual image generation — default ON, but ask.** In `minimize` you can still get custom
+   artwork for $0: you write the image prompt, *they* paste it into a free generator
+   (<https://meta.ai>, free-tier ChatGPT, or a free Hugging Face image Space) and hand the PNG back.
+   It raises video quality noticeably. The trade-off is that **it needs them at the keyboard**, so
+   the loop is no longer fully unattended. Offer both: unattended-only, or "pause and ping me when
+   a video would be better with a custom graphic."
+4. **How many tasks per run — default 1 to start, then 3.** Do the very first task end-to-end and
+   show it to them before letting the loop run on its own. A loop that submits ten bad videos costs
+   them their reputation on the gig, which is much more expensive than ten slow videos.
+5. **How often — default every 30 minutes** while their computer is on. Respect the gig's rate
+   limit above anything they ask for.
+6. **Where to keep the project — default `~/clipper/<gig-id>/`.**
+
+Write their answers into `config.json` (Phase 3) so the loop is reproducible and they can edit it
+later without you.
+
+---
+
+## Phase 2 — Resolve and inspect the gig
+
+**First, work out what they actually gave you.** Three inputs are valid, and they need different
+handling — figure out which one it is before you call anything:
+
+- **An invite URL** — `https://dollarplatoon.com/gig/<GIG_ID>/join?invite=<TOKEN>` gives you both
+  the gig id and the invite token. This is the only form that can join a fresh `invite`-only gig.
+- **A bare gig id** (`GIG_01HX…`) — no token. If they've already joined, that's fine (see below). If
+  they haven't and the gig is `join_policy: "invite"`, the join will 403 — ask for the full link.
+- **A mailbox id** — their own mailbox on a gig they already joined. This is enough on its own: the
+  mailbox *is* the membership, so there is nothing to join and no invite token needed.
+
+**Resolve a mailbox id (or check an existing membership) with `GET /mailboxes/mine`:**
+
+```bash
+curl -s -H "x-api-key: $DOLLARPLATOON_API_KEY" https://dollarplatoon.com/api/mailboxes/mine
+```
+
+Each entry carries `id`, `gig_id`, `gig_title`, and `status`. Match on `id` for a mailbox id, or on
+`gig_id` for a bare gig id. This is also the cheapest way to answer "which gigs am I already on?" —
+run it whenever the user is vague about which gig they mean, and let them pick from the list.
+
+**Always run this before joining, even when you have an invite URL.** If a mailbox for that gig
+already exists, **skip Phase 2's join entirely and reuse the existing `mailbox.id`** — joining twice
+is a duplicate, not a fresh start. And check `status`: an `inactive` mailbox means the client
+disabled them, so stop and tell the user rather than polling into a wall.
+
+**Look at the gig before joining:**
+
+```bash
+curl -s -H "x-api-key: $DOLLARPLATOON_API_KEY" \
+  https://dollarplatoon.com/api/gigs/<GIG_ID>
+```
+
+Read and report back to the user, in plain language:
+
+- **`terms`** — what the client actually wants. This is the brief. Read it out loud to them.
+- **`price`** — what one approved proof pays.
+- **`distribution`** — how tasks reach you. This decides the shape of your loop (see Phase 4).
+- **`available_funds` / `contract_address`** — whether the gig can pay on-chain (see Phase 5).
+- **`default_rate_limit_count` / `default_rate_limit_minutes`** — your ceiling. Never poll past it.
+- **`task_timeout`** — how long you may hold a task before it expires on you.
+- **`review_timeout`** — proofs auto-approve after this (default 48h) if the client doesn't review.
+
+**Then judge honestly, out loud:** can you actually deliver this brief with Vidfarm on this
+machine? If the gig wants 4K motion-graphics work and cost mode is `minimize`, say so now. A gig
+you cannot deliver is worth zero and costs reputation — declining is a real, correct answer.
+
+**Join — only if `GET /mailboxes/mine` showed no mailbox for this gig:**
+
+```bash
+curl -s -X POST -H "x-api-key: $DOLLARPLATOON_API_KEY" -H "content-type: application/json" \
+  -d '{"name":"<their name>'"'"'s clipper","invite":"<TOKEN>","notes":"AI-assisted video editing via vidfarm"}' \
+  https://dollarplatoon.com/api/gigs/<GIG_ID>/mailboxes
+```
+
+Save the returned `mailbox.id` — every proof you submit needs it. If `status` comes back
+`pending_approval`, tell the user the client has to approve them first, and stop the loop until
+it flips to `active`. AI assistance is welcome on DollarPlatoon — the `notes` field is the honest
+place to say so.
+
+---
+
+## Phase 3 — Scaffold the project
+
+Build a small, boring, readable project. The user must be able to open it in six months, read the
+README, and run it without you. Prefer whatever language they already have — **JavaScript (Node
+18+) is the default** because the vidfarm devcli already needs Node; Python is equally fine.
+
+```
+~/clipper/<gig-id>/
+├── .env                 # VIDFARM_API_KEY, DOLLARPLATOON_API_KEY   (gitignored)
+├── .gitignore           # .env, videos/, node_modules/
+├── config.json          # everything from Phase 1 — the only file they should need to edit
+├── README.md            # how to run it, in their words
+├── dp.mjs               # DollarPlatoon REST client — one function per endpoint
+├── loop.mjs             # the loop: get task → make video → submit proof → log
+├── ledger.jsonl         # one line per task. Append-only. This is the money record.
+├── NOTES.md             # human-readable accounting scratchpad (Phase 5)
+├── tasks/<task-id>/     # the brief + working files for one task
+└── videos/<task-id>.mp4 # the render (local only — never uploaded to DollarPlatoon)
+```
+
+`dp.mjs` — keep it this thin. Every function is one endpoint from the DollarPlatoon skill:
+
+```js
+// dp.mjs — DollarPlatoon REST client
+const API = "https://dollarplatoon.com/api";
+const KEY = process.env.DOLLARPLATOON_API_KEY || process.env.DOLLAR_PLATOON_API_KEY;
+
+async function dp(path, { method = "GET", body } = {}) {
+  const res = await fetch(API + path, {
+    method,
+    headers: { "x-api-key": KEY, ...(body ? { "content-type": "application/json" } : {}) },
+    body: body ? JSON.stringify(body) : undefined,
+  });
+  const text = await res.text();
+  let json; try { json = JSON.parse(text); } catch { json = { raw: text }; }
+  if (!res.ok) throw Object.assign(new Error(json.error || text), { status: res.status, json });
+  return json;
+}
+
+export const myMailboxes = ()                   => dp(`/mailboxes/mine`);   // resolve a mailbox id → gig_id, or list gigs already joined
+export const getGig      = (gigId)              => dp(`/gigs/${gigId}`);
+export const pollQueue   = (gigId, count = 1)   => dp(`/gigs/${gigId}/queue/poll`, { method: "POST", body: { count } });
+export const declineTask = (gigId, msgId)       => dp(`/gigs/${gigId}/queue/${msgId}/decline`, { method: "POST" });
+export const inbound     = (mailboxId)          => dp(`/mailboxes/${mailboxId}/inbound`);
+export const submitProof = (gigId, body)        => dp(`/gigs/${gigId}/proofs`, { method: "POST", body });
+export const myProofs    = (gigId)              => dp(`/gigs/${gigId}/proofs`);
+export const myRollups   = ()                   => dp(`/rollups/mine`);
+```
+
+Write the README **for them, not for you**: what the loop does, how to start and stop it, what each
+file is, how much a video costs to make, how they get paid, and what to do when something breaks.
+
+---
+
+## Phase 4 — How to get a task (depends on `distribution`)
+
+Different gigs hand out work differently. Read `gig.distribution` and use the matching path. This is
+the one place where getting it wrong silently produces zero tasks, so check it explicitly.
+
+- **`queue`** — the FIFO shared pool. `POST /gigs/:id/queue/poll` with `{"count": N}`. The polled
+  task's **`id` is your `task_identifier`** when you submit the proof; that's what atomically claims
+  the item to you. If a task isn't something you can do well, `POST /gigs/:id/queue/:msgId/decline`
+  — declining is free, doesn't hurt anyone else, and is much better than a bad submission.
+- **`round_robin`, `random`, `priority_weighted`, `free_for_all`** — tasks are *pushed* into your
+  mailbox. Read them with `GET /mailboxes/:mbxId/inbound`. Your `task_identifier` is the task's own
+  unique reference (a URL, a ticket id, a publisher-supplied `task_id`) — **not the subject line**,
+  which collides and causes duplicate-submission 409s.
+- **`inbound_proof`** — no tasks are distributed at all. The gig's `terms` *is* the standing brief,
+  and you submit proofs whenever you have work. Use a stable unique identifier of your own.
+
+**The task body may carry structured data for you.** Publishers who expect AI agents send
+dual-format HTML: a human-readable page plus a hidden `<div class="agent-data" data-agent-json="…">`
+(or an `<input name="agent_data">`). Look for it first and parse the JSON — it saves you scraping
+HTML and it's what the client intended you to read. Otherwise read the HTML/text yourself.
+
+Whatever you find, **write the brief to `tasks/<task-id>/brief.md` before you start editing.** If
+the render goes wrong you need to know what was actually asked for.
+
+---
+
+## Phase 5 — Payouts: figure out which kind this gig is, up front
+
+This decides how much bookkeeping you owe the user. Get it right in Phase 2, before any work.
+
+**On-chain (Base L2 USDC, automatic).** Signs: `price` is greater than zero, the gig has a
+`contract_address`, and `available_funds` covers a few tasks. Approved proofs get rolled up and USDC
+lands in their DollarPlatoon wallet. They don't invoice anyone; you just watch.
+
+**Off-chain (paid some other way — GCash, bank, platform credit, barter).** Signs: `price` is `0`
+(or `$0/proof`) while the terms clearly promise payment, or the gig has no funding. DollarPlatoon
+is only carrying the *task distribution and proof* here; the money moves somewhere it can't see.
+
+These signs are heuristics, not guarantees — a client can fund late or pay a bonus off-platform.
+**When it's ambiguous, say so and ask the user what the client told them.** Don't assert income.
+
+**If off-chain: keep the books yourself. This is not optional.** Nobody else is recording what they
+earned. Every task, append one line to `ledger.jsonl`:
+
+```json
+{"ts":"2026-07-25T09:14:00Z","gig_id":"GIG_01HX...","task_id":"01HY...","proof_id":"01HZ...","share_url":"https://vidfarm.cc/...","quoted_price":0.5,"payout":"offchain","status":"submitted","render_seconds":214,"ai_spend_usd":0,"notes":"3 scenes, template ABC"}
+```
+
+And keep `NOTES.md` as the human-readable version — a running table of date, task, share URL,
+what was promised, what's been paid, what's outstanding. **Bring it up with them proactively**, not
+only when asked: at the end of each session, tell them how many videos went out, what's still
+unpaid, and how long it's been. That summary is the thing that gets them paid when a client forgets.
+
+Mirror `NOTES.md` into their Vidfarm library too, so it survives a wiped laptop and follows their
+account — this is the same durable-context habit the Vidfarm skill uses for briefs and brand kits:
+
+```bash
+vidfarm put-file NOTES.md --folder clipper/<gig-id> \
+  --notes "Clipper loop accounting for DollarPlatoon gig <GIG_ID> — off-chain payouts, unpaid balance tracked here"
+```
+
+For on-chain gigs, still write `ledger.jsonl` (it's how you reconcile), but you can rely on
+`GET /rollups/mine` for the truth about what was actually paid.
+
+---
+
+## Phase 6 — Make the video with Vidfarm
+
+This is ordinary Vidfarm director work — **the full method is in
+<https://vidfarm.cc/skill.md> and the director skill it points to. Follow that, not a shortcut.**
+What matters for a clipper loop specifically:
+
+**Set the cost posture once, at setup:**
+
+```bash
+vidfarm login --api-key "$VIDFARM_API_KEY"   # or: vidfarm login <email>
+vidfarm cost-mode minimize                   # or whatever they chose in Phase 1
+vidfarm doctor                               # confirms ffmpeg + hyperframes + Chrome
+```
+
+`minimize` **refuses** billed operations unless you pass `--yes`. That is the safety rail that keeps
+the loop profitable — a gig paying $0.50 a video cannot absorb $2 of AI generation. Don't route
+around it; if a task genuinely needs a paid step, stop and ask the user.
+
+**Reach for the cheap paintbrushes first** — this is the whole economics of the loop:
+
+1. **Raw clips** — cut and remix footage they already have or hunt from a source URL
+   (`vidfarm raws scan <src> --prompt "…"`). Free local compute. The workhorse.
+2. **HTML/JS hyperframes** — text, titles, graphics, data-viz as CSS/GSAP layers. Free,
+   deterministic, re-skinnable. **Never use AI video to put text on screen.**
+3. **Free stock** — `vidfarm media search "<meaning>" --type bgm|sfx|image|video`. Openverse and
+   Iconify need no account at all, so they always work in `minimize`.
+4. **Pure AI generation** — last resort, and it's what costs money.
+
+A rough shape of one task (adapt to the brief; the director skill has the real workflows):
+
+```bash
+vidfarm discover "<what the brief asks for>"      # find a template
+vidfarm fork <template_id>                        # → forkId
+vidfarm pull <forkId> --dir tasks/<task-id>       # composition + grounding onto disk
+# …edit: set-text / set-media / place / captions / raws — per the director skill…
+vidfarm lint tasks/<task-id>                      # catch timing + media errors before rendering
+vidfarm render <forkId> --dir tasks/<task-id> --out videos/<task-id>.mp4 --wait   # LOCAL = $0.00
+```
+
+**Free manual image generation** (if they said yes in Phase 1): when the video genuinely needs a
+custom graphic that stock and masking can't cover, don't quietly spend and don't say "not possible."
+Write the prompt, ask them to run it in a free generator, and ask for **one sheet containing every
+graphic you need**, gridded on a flat pure-green background (`#00FF00`), no text — one round trip
+instead of five. Then split it locally for $0:
+
+```bash
+vidfarm put-file ./sheet.png
+vidfarm mask ./sheet.png --crop x,y,w,h --flat "#00FF00" --out prop-a.png   # once per element
+```
+
+**Watch the finished video before you submit it.** Check `vidfarm stills` frames or open the MP4.
+Ask yourself whether it actually satisfies `terms`. If it doesn't, fix it or decline the task —
+a rejected proof carries a reputation penalty (`low_quality` 1x, `incomplete` 2x, `fake_proof` 5x)
+and those follow their wallet across every gig on the network.
+
+---
+
+## Phase 7 — Submit the proof as a URL, never as a video file
+
+**Do not upload video files to DollarPlatoon.** Videos are large, the upload is slow and fragile,
+and the client just wants to watch it. Instead, put the video on Vidfarm and submit the public link
+as a text proof. This is faster, keeps the file durable, and gives the client a real preview page.
+
+```bash
+vidfarm approve --video ./videos/<task-id>.mp4 --caption "<gig title> — task <task-id>"
+# uploads to durable My Files (up to 200 MB, direct to S3) and prints:  share_url
+```
+
+`vidfarm approve` is the right command because it stores the MP4 in **durable My Files**, not the
+30-day temp store — a temp-hosted proof would 404 on the client's share page a month later, after
+they've already paid. Then:
+
+```js
+await submitProof(gigId, {
+  mailbox_id: MAILBOX_ID,
+  task_identifier: task.id,        // queue gigs: the polled task's id. others: the task's own ref.
+  proofs: [shareUrl],              // the public vidfarm.cc page. Add the direct MP4 viewUrl too if the client asks for a file.
+});
+```
+
+Check the response:
+
+- **`warning: "available funds are less than the task price"`** — the proof was accepted but
+  **cannot be paid until the client deposits more.** Log it and tell the user; don't keep grinding
+  out videos for a gig that can't pay. This is the single most important warning in the loop.
+- **`429`** — you hit the rate limit. The `rate_limit` object has `retry_at`. Sleep, don't retry.
+- **`409`** — duplicate `task_identifier`. You already submitted this one; don't resubmit.
+- **`410`** — the task expired while you were rendering. Ask the client for an extension, or move on.
+
+Proofs auto-approve after `review_timeout` (default 48h) if the client doesn't review. Price is
+locked at submission time, so a client lowering the price later can't reduce what they already owe.
+
+---
+
+## Phase 8 — Run the loop
+
+Once one task has gone end-to-end and the user has seen the video, let it run. Keep it dull and
+observable:
+
+- **Respect the rate limit** — treat `default_rate_limit_count` / `_minutes` as a hard ceiling, and
+  back off on any `429` until `retry_at`.
+- **One task at a time** unless the user asked otherwise. Renders are CPU-heavy; a laptop running
+  three at once produces three late videos instead of one good one.
+- **Never claim more than you can finish before `task_timeout`.**
+- **Log every attempt to `ledger.jsonl`, including failures.** Silent failures are how a loop
+  quietly earns nothing for a week.
+- **Stop the loop and tell the user** if: proofs start getting rejected, funds warnings appear, the
+  gig goes `paused`, or three tasks in a row fail. Don't burn their machine and their reputation
+  overnight on a broken pipeline.
+- Their computer must be awake. Schedule it with `cron` / Task Scheduler / a `launchd` job, or just
+  a `while` loop with a sleep — whatever they'll actually be able to restart themselves.
+
+**Report to the user each session** (and write the same into `NOTES.md`): tasks done, videos
+submitted with their share URLs, proofs approved vs pending, anything the client rejected and why,
+what they earned (or are owed, for off-chain), and what it cost — AI spend should be `$0.00` in
+`minimize`, and if it isn't, that's the first thing to say.
+
+---
+
+## Honesty rules
+
+These exist because the loop touches real customers, real reputation, and real money.
+
+- **Never fake a proof.** Every submitted URL must point at a video you actually rendered for that
+  task. `fake_proof` is a 5x reputation penalty and it is permanent and wallet-anchored.
+- **Never promise earnings.** Pay is per approved proof and demand-driven — quiet days pay nothing.
+  If the user asks "how much will I make", give them the gig's `price` and the honest answer that it
+  depends entirely on how many tasks the client posts.
+- **Never spend their money silently.** In `minimize`, a billed step stops and asks. Report actual
+  spend, not estimates — `vidfarm wallet` has the truth.
+- **Report what really happened.** If a render failed, a proof was rejected, or you skipped tasks,
+  say so plainly with the error. A loop the user can't trust is worse than no loop.
+- **Don't work prohibited gigs.** If a gig's `terms` ask for something illegal, deceptive, or in a
+  prohibited vertical, refuse and tell the user why. DollarPlatoon has no dispute resolution — you
+  are their only check.
+
+---
+
+## When something breaks
+
+| Symptom | Likely cause | Fix |
+|---|---|---|
+| Poll returns no tasks | Gig isn't `distribution: "queue"` | Check `gig.distribution`, use Phase 4's matching path |
+| `403` on join | Missing or revoked invite token | Check `GET /mailboxes/mine` first — they may already be in; otherwise ask for the full invite URL again |
+| User only has a mailbox id | Already joined; no invite needed | `GET /mailboxes/mine`, match on `id`, use its `gig_id` and skip the join |
+| Mailbox `status: "inactive"` | Client disabled them | Stop the loop; they have to talk to the client |
+| Join returns `pending_approval` | Gig has `requires_approval` | Wait for the client; don't poll |
+| `429` on poll or proof | Worker rate limit | Sleep until `rate_limit.retry_at` |
+| `409` on proof | Duplicate `task_identifier` | You already submitted; check `GET /gigs/:id/proofs` |
+| `410` on proof | Task expired (`task_timeout`) | Ask the client to `extend`, or take a fresh task |
+| Proof accepted with a funds `warning` | Gig treasury is short | Tell the user; pause the loop |
+| Render fails or hangs | Local toolchain | `vidfarm doctor`, then `vidfarm doctor --kill-orphans` |
+| `402` on a vidfarm command | Free plan or `minimize` blocking a billed route | Use the free path the error names |
+| Share URL 404s later | Video went to the temp store | Re-approve with `vidfarm approve --video` (durable My Files) |
+
+---
+
+## More
+
+- Explainer: <https://vidfarm.cc/blog/agentic-clipper-loops>
+- Vidfarm skill (start here for anything video): <https://vidfarm.cc/skill.md>
+- DollarPlatoon skill (full REST reference): <https://dollarplatoon.com/skill.md>
+- Vidfarm marketplace: <https://vidfarm.cc/marketplace>

package/dist/src/cli.js

@@ -407,8 +407,9 @@ Raws (the third library — mine long-form video into a reusable raws store):
                             save only that exact clipped range.
 
 Speech (TTS/STT) — LOCAL-FIRST on your own AI key; --cloud is the explicit backup:
-  tts "<text>"              Text → narration audio file.                 LOCAL: your OPENAI/GEMINI/
-                            The VOICE STYLE is promptable via --style.   OPENROUTER_API_KEY, no cloud job
+  tts "<text>"              Text → narration audio file.                 FREE local Kokoro by default in
+                            The VOICE STYLE is promptable via --style.   cost mode minimize/hybrid; else
+                                                                         your OPENAI/GEMINI/OPENROUTER key
     --style "<direction>"     Voice-style prompt: tone/pacing/accent/emotion/persona
                               (e.g. "calm, warm bedtime narrator", "excited sports announcer")
     --voice <name>            Provider voice preset (openai: alloy/ash/coral…; gemini: Kore/Puck…).
@@ -417,8 +418,13 @@ Speech (TTS/STT) — LOCAL-FIRST on your own AI key; --cloud is the explicit bac
     --out <file>              Output audio path (default tts-<id>.mp3|wav)
     --provider <p> --model <m> --format mp3|wav
     --engine local            KEYLESS local Kokoro-82M engine (free, no account) — fixed voice
-                              presets (af_heart, am_adam, bf_emma, …), no --style. When NO
-                              provider key is found the default byok engine auto-falls back here.
+                              presets (af_heart, am_adam, bf_emma, …), no --style. This is the
+                              DEFAULT in cost mode minimize/hybrid; it also auto-runs when NO
+                              provider key is found. Asking for a premium voice (--style,
+                              --provider, --model, --own-key, or a non-Kokoro --voice) opts out,
+                              as does cost mode rich-ai. If the local engine is unavailable the
+                              run falls back to your provider key / cloud instead of failing.
+    --engine byok             Force your own provider key even in minimize/hybrid.
     --cloud                   BACKUP: platform job → POST /api/v1/primitives/audio/speech (+ poll).
                               Defaults to ElevenLabs on the platform key (wallet-billed); --voice
                               takes an ElevenLabs voice_id (see: vidfarm voices).
@@ -685,6 +691,8 @@ Cost spectrum (default to the cheapest approach that works; see SKILL.director.m
   respects it: minimize refuses billed spend without --yes and points you at the free
   local path; hybrid/rich-ai run but print each op's cost. FREE local engines never gate
   (local render, tts --engine local, stt --engine whisper, remove-greenscreen --local).
+  In minimize AND hybrid, 'vidfarm tts' DEFAULTS to the free local Kokoro voice — ask for
+  a premium voice (--style/--voice/--provider/--own-key/--cloud) or rich-ai to opt out.
 
 Escape hatch — call ANY route directly:
   api <METHOD> <path>       Raw REST call with auth + pretty errors
@@ -1196,7 +1204,7 @@ function assertApiOk(result, action) {
     const hint = result.status === 401 || result.status === 403
         ? " Check --api-key / VIDFARM_API_KEY and that you own this resource."
         : result.status === 402
-            ? " The account is not on a paid plan (or the wallet is empty)."
+            ? " This route is paid-only — your key is valid but the account is on the free plan (or the wallet is empty). Upgrade at /pricing, or keep working locally: `vidfarm serve` and the local render/decompose paths are free."
             : result.status === 429
                 ? " Rate-limited; wait and retry."
                 : "";
@@ -5133,6 +5141,35 @@ async function runTtsCommand(argv) {
         await runLocalKokoroTts({ text, voice, style, out: parsed.values.out, format: explicitFormat, json });
         return;
     }
+    // COST-MODE DEFAULT: in minimize/hybrid, narration starts on the FREE local
+    // Kokoro engine unless the run asked for a premium/cloud voice (see
+    // wantsPremiumVoice). The BYOK key path is a real fallback, not a downgrade —
+    // if Kokoro is missing its python deps / model weights we fall THROUGH to it
+    // below rather than failing the run.
+    const costMode = resolveCostMode(parsed.values);
+    const premium = wantsPremiumVoice(parsed.values, voice, style);
+    /** Set when the cost-mode default ran Kokoro and it failed — so the keyless
+     *  path below doesn't pointlessly retry the same engine. */
+    let kokoroError = null;
+    if (!engine && !wantCloud && costMode.mode !== "pure-ai" && !premium) {
+        try {
+            await runLocalKokoroTts({
+                text,
+                voice,
+                style,
+                out: parsed.values.out,
+                format: explicitFormat,
+                json,
+                costModeDefault: costMode.mode
+            });
+            return;
+        }
+        catch (error) {
+            kokoroError = String(error?.message ?? error).split("\n")[0].slice(0, 200);
+            // stderr, not stdout — keeps --json output parseable when we fall back.
+            console.error(`${DIM}Local Kokoro engine unavailable (${kokoroError}) — falling back to your provider key / cloud TTS.${RESET}`);
+        }
+    }
     if (!wantCloud) {
         // A Gemini voice preset ("Kore", "Puck", …) should find a Gemini-capable
         // key first — openrouter's default TTS model is Gemini-family too.
@@ -5148,6 +5185,14 @@ async function runTtsCommand(argv) {
                 const wanted = requestedProvider ?? `${voiceProvider} (implied by --voice "${voice}")`;
                 throw new Error(`tts: provider ${wanted} was requested but no matching key is configured. ${speechKeyHint("tts")}`);
             }
+            // Kokoro already ran and failed on the cost-mode default above — don't
+            // retry the same broken engine, and don't dump its raw error. Name both
+            // dead ends (no local deps, no key) with the two ways forward.
+            if (kokoroError) {
+                throw new Error(`tts: no provider key configured AND the free local Kokoro engine is unavailable (${kokoroError}). ` +
+                    "Fix one of the two: install the local engine (`pip install kokoro-onnx soundfile`, or point " +
+                    `HYPERFRAMES_PYTHON at a venv that has them), or ${speechKeyHint("tts")}`);
+            }
             // Keyless fallback (no explicit provider/voice): rather than erroring
             // (speechKeyHint), speak through the local Kokoro engine so agents
             // always get audio out of the box.
@@ -5237,17 +5282,41 @@ async function runTtsCommand(argv) {
         console.log(`${GREEN}Wrote ${outPath}${RESET} ${DIM}(${mediaUrl})${RESET}`);
     }
 }
+/** Kokoro voice ids look like af_heart / am_adam / bf_emma. Anything else is a
+ *  cloud provider preset (openai "alloy", gemini "Kore", an ElevenLabs id).
+ *  Declared as a hoisted function, not a const: `wantsPremiumVoice` runs during
+ *  the top-level `main()` await, which is BEFORE this point in module order. */
+function isKokoroVoice(voice) {
+    return /^[a-z]{2}_[a-z0-9]+$/i.test(voice);
+}
+/**
+ * Did this run explicitly ask for a PREMIUM (cloud/BYOK) voice? If so, cost mode
+ * must not silently route it to the free local engine — Kokoro has fixed voice
+ * presets and no style direction, so honoring these flags locally would be a
+ * silent downgrade. Any of: --provider, --model, --style, --own-key, or a
+ * --voice that isn't a Kokoro preset.
+ */
+function wantsPremiumVoice(values, voice, style) {
+    if (values.provider || values.model || values["own-key"])
+        return true;
+    if (style)
+        return true;
+    if (voice?.trim() && !isKokoroVoice(voice.trim()))
+        return true;
+    return false;
+}
 // Keyless local TTS via the bundled hyperframes engine (Kokoro-82M ONNX —
-// free, no account, no HeyGen). Used by `--engine local` and as the automatic
-// fallback when no provider key is configured. Kokoro emits WAV natively; an
-// mp3 request transcodes through the bundled ffmpeg.
+// free, no account, no HeyGen). Used by `--engine local`, as the cost-mode
+// default in minimize/hybrid, and as the automatic fallback when no provider
+// key is configured. Kokoro emits WAV natively; an mp3 request transcodes
+// through the bundled ffmpeg.
 async function runLocalKokoroTts(input) {
     const wantMp3 = input.format === "mp3" || /\.mp3(\?|#|$)/i.test(input.out ?? "");
     // Kokoro voice ids look like af_heart / am_adam / bf_emma. A provider voice
     // (openai "alloy", gemini "Kore") passed through the fallback path would
     // just error — drop it to the default with a note instead.
     let voice = input.voice?.trim() || undefined;
-    if (voice && !/^[a-z]{2}_[a-z0-9]+$/i.test(voice)) {
+    if (voice && !isKokoroVoice(voice)) {
         if (!input.json)
             console.log(`${DIM}Voice "${voice}" is a provider voice — the local Kokoro engine has its own presets (af_heart, af_nova, am_adam, bf_emma, …). Using the default voice.${RESET}`);
         voice = undefined;
@@ -5258,8 +5327,12 @@ async function runLocalKokoroTts(input) {
     const outPath = path.resolve(process.cwd(), input.out ?? `tts-${Date.now().toString(36)}.${wantMp3 ? "mp3" : "wav"}`);
     mkdirSync(path.dirname(outPath), { recursive: true });
     const wavPath = wantMp3 ? path.join(mkdtempSync(path.join(tmpdir(), "vidfarm-tts-")), "speech.wav") : outPath;
-    if (!input.json)
+    if (!input.json) {
         console.log(`${DIM}Generating speech with the local Kokoro-82M engine (free, no account)…${RESET}`);
+        if (input.costModeDefault) {
+            console.log(`${DIM}(cost mode ${costModeDisplayName(input.costModeDefault)} → free local voice by default. Want a premium voice? add --style/--voice/--provider, or --cloud.)${RESET}`);
+        }
+    }
     const run = await runHyperframesCommand("tts", [input.text, "-o", wavPath, ...(voice ? ["-v", voice] : [])], { stdio: input.json ? "capture" : "inherit" });
     if (run.code !== 0 || !existsSync(wavPath)) {
         const detail = run.stderr.trim().split("\n").pop() ?? "";
@@ -5276,7 +5349,7 @@ async function runLocalKokoroTts(input) {
     }
     const bytes = safeSize(outPath);
     if (input.json) {
-        printJson({ ok: true, mode: "local-engine", engine: "kokoro-82m", provider: null, voice: voice ?? "af_heart", fallback: Boolean(input.fallback), content_type: wantMp3 ? "audio/mpeg" : "audio/wav", bytes, out: outPath });
+        printJson({ ok: true, mode: "local-engine", engine: "kokoro-82m", provider: null, voice: voice ?? "af_heart", fallback: Boolean(input.fallback), cost_mode_default: input.costModeDefault ?? null, content_type: wantMp3 ? "audio/mpeg" : "audio/wav", bytes, out: outPath });
     }
     else {
         console.log(`${GREEN}Wrote ${outPath}${RESET} ${DIM}(local Kokoro-82M — free, no account, ${formatBytes(bytes)})${RESET}`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@officexapp/vidfarm-devcli",
-  "version": "0.21.22",
+  "version": "0.21.24",
   "description": "Local bridge for the Vidfarm Trackpad Editor. `vidfarm serve <template_id>` boots the FULL editor on localhost (disk-backed records/storage, free in-process render); edit composition.html on disk (Claude Code, Codex, etc.) and the browser live-morphs it.",
   "type": "module",
   "bin": {
@@ -50,6 +50,8 @@
     "README.md",
     "SKILL.md",
     "SKILL.director.md",
+    "clipper.md",
+    "update.md",
     "!readme.secret.md",
     "!**/*.secret.*"
   ],

package/update.md

@@ -0,0 +1,143 @@
+---
+name: vidfarm-update
+description: How to update Vidfarm — clean-uninstall the `vidfarm` agent skill and the `vidfarm-devcli`, then reinstall the latest of both from vidfarm.cc/skill.md and npm. Read this whenever the user asks to update/upgrade Vidfarm, says their vidfarm skill or devcli is out of date, hits a command or route that "used to work", or gets a version-mismatch or unknown-command error. Also carries breaking-change notes per version — check it before assuming an old workflow still applies.
+---
+
+# Updating Vidfarm
+
+**Read this file live at <https://vidfarm.cc/update.md> every time you update — don't work from a
+cached copy.** It carries the current breaking-change notes, and an update that skips them is how a
+working setup quietly breaks.
+
+Vidfarm ships as two moving parts, and they are versioned together:
+
+1. **The `vidfarm` agent skill** — the director workflows on disk, installed under
+   `.agents/skills/vidfarm/` with a `.claude/skills/vidfarm` symlink.
+2. **The `vidfarm-devcli`** — the `vidfarm` command, published as `@officexapp/vidfarm-devcli`.
+
+Updating one but not the other is the most common cause of "this command doesn't exist" and "the
+skill says to do X but it fails" — the skill documents routes and flags the installed CLI may not
+have yet. **Always update both together.**
+
+---
+
+## The update, in four steps
+
+### 1. Record what they're on now
+
+Worth ten seconds, because it tells you which breaking-change notes below actually apply:
+
+```bash
+vidfarm --version            # devcli version
+npm view @officexapp/vidfarm-devcli version   # latest published
+cat skills-lock.json         # which skill packs are installed, and from where
+```
+
+### 2. Remove the old skill
+
+There is no `skills remove` command — a skill is just a directory plus a lock entry, so delete both.
+Run this from the project root (the directory holding `.agents/skills`):
+
+```bash
+rm -rf .agents/skills/vidfarm .claude/skills/vidfarm
+```
+
+Then drop `"vidfarm"` from the `skills` object in `skills-lock.json` (leave any other packs alone —
+`hyperframes`, `vidfarm-media`, etc. are separate entries and other tools write to this file too).
+
+A clean removal matters more than it looks: `skills add` overwrites files it knows about, so a file
+that was *deleted upstream* survives an in-place refresh and keeps feeding the agent stale
+instructions. Deleting the directory is what guarantees you end up with exactly the current pack.
+
+### 3. Reinstall the latest of both
+
+```bash
+npm i -g @officexapp/vidfarm-devcli@latest    # the CLI
+vidfarm skills add vidfarm                    # the skill pack, fetched from vidfarm.cc/skill-pack/*
+```
+
+`vidfarm skills update` refreshes every installed pack in place — that's the quick path when nothing
+has been removed upstream. The uninstall-then-add above is the thorough one; prefer it when the user
+is troubleshooting, is several versions behind, or when a note below says to.
+
+If the user also has the HyperFrames engine installed, bring it along — Vidfarm renders *are*
+HyperFrames renders, and a version skew between them shows up as render failures:
+
+```bash
+npm i -g hyperframes@latest
+```
+
+### 4. Verify before declaring success
+
+```bash
+vidfarm --version    # should match the npm latest from step 1
+vidfarm doctor       # ffmpeg + hyperframes + Chrome + auth, all in one check
+vidfarm whoami       # confirms the API key survived the reinstall, prints the active cost mode
+vidfarm skills list  # confirms the vidfarm pack is present, and from which source
+```
+
+`vidfarm doctor` is the real test — a version number proves the package installed, not that the
+toolchain works. If `whoami` fails, the login just needs redoing: `vidfarm login <email>`, or
+`vidfarm login --api-key "$VIDFARM_API_KEY"`. Credentials live in `~/.vidfarm/` and normally survive
+an update untouched.
+
+Then **re-read <https://vidfarm.cc/skill.md>** and follow it to the director skill. The freshly
+installed pack may route differently than the one you loaded at the start of the session.
+
+---
+
+## Breaking changes and version notes
+
+Check this section against the version from step 1 and apply anything between their version and the
+latest. If nothing here mentions their version, a plain update is safe.
+
+**Current published devcli: `0.21.23`.**
+
+- **0.21.x — the package moved scopes.** The devcli is `@officexapp/vidfarm-devcli`. The old
+  `@mevdragon/vidfarm-devcli` is not maintained. If `npm ls -g` shows the `@mevdragon` package,
+  uninstall it explicitly, or two `vidfarm` binaries will fight over the same name on `PATH`:
+
+  ```bash
+  npm rm -g @mevdragon/vidfarm-devcli
+  npm i -g @officexapp/vidfarm-devcli@latest
+  which vidfarm && vidfarm --version   # confirm the survivor is the new one
+  ```
+
+- **`update-skill` is retired.** It now warns and delegates. Use `vidfarm skills add vidfarm`.
+  `--global` and `--platform` are gone; scope an install with `--dir <project-root>` instead.
+
+- **0.6.0 and earlier `vidfarm serve` is broken when installed from npm** (the tarball shipped
+  without its runtime). Anyone still on 0.5.x must update — there is no workaround.
+
+- **Cost mode is remembered across updates.** It lives in `~/.vidfarm/cost-mode.json`, outside the
+  package, so an update never silently changes their spending posture. Confirm it with `vidfarm
+  whoami` anyway and say what it is out loud — it decides whether billed commands are refused.
+
+---
+
+## If the update leaves things broken
+
+| Symptom | Cause | Fix |
+|---|---|---|
+| `vidfarm: command not found` | Global bin not on `PATH`, or the install failed | `npm ls -g --depth=0`, then check `npm bin -g` is on `PATH` |
+| Wrong version after installing | An older `@mevdragon` copy is shadowing it | `npm rm -g @mevdragon/vidfarm-devcli`, then `which vidfarm` |
+| `EACCES` on global install | npm's global dir needs root | Use a Node version manager (nvm/fnm) rather than `sudo npm` |
+| Skill still shows old instructions | The directory wasn't actually removed | Redo step 2, confirm `.agents/skills/vidfarm` is gone, then `skills add` |
+| `skills add` fails to fetch | Offline or the host is unreachable | It falls back to copies bundled in the installed package — check `vidfarm skills list` for the source it used |
+| `401` / `whoami` fails | Credential not carried over | `vidfarm login <email>` (email code), or `--api-key` |
+| Renders fail after updating | devcli / hyperframes version skew | `npm i -g hyperframes@latest`, then `vidfarm doctor` |
+| A documented route 404s | Skill is newer than the deployed backend | Report it — don't work around it by calling the renderer or an internal route directly |
+
+**Tell the user plainly what changed.** After updating, say which versions they went from and to,
+and name anything from the breaking-changes section that affects them. If the update failed partway,
+say so with the error rather than reporting success — a half-updated toolchain fails later, in the
+middle of a render, where it's much harder to diagnose.
+
+---
+
+## More
+
+- Entry skill (start here after updating): <https://vidfarm.cc/skill.md>
+- Full director skill: <https://vidfarm.cc/skill/vidfarm>
+- Skill pack index (what else is installable): <https://vidfarm.cc/skill-pack/index.json>
+- Agentic clipper loop: <https://vidfarm.cc/clipper.md>