anabologyco-maker/claudia-bot/clips

Claude Code and agent skill by anabologyco-maker/claudia-bot

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Use when making video clips - drafting them from the board, writing the prompt book, hosting audio windows and start images, submitting to OpenRouter or through the Higgsfield connector, finalizing a Higgsfield draft, polling, ingesting takes, lip-sync checks, covers, and declined or failed takes.

Install

npx vskill@latest install anabologyco-maker/claudia-bot/clips

SKILL.md

Clips - one prompt book, the owner reads it, then spend, sync, covers

Read docs/60-clips.md §1 and §9 (the order of work) first. The prompt is the product: write every prompt yourself, in one sitting.

Draft and write (free)

  1. slop --json -p NAME board audit must pass. Run each of its clip_cmds with --json added (slop --json -p NAME clip new SHOT --kind sung). Photoreal sung close-ups: add --provider higgsfield (§6.5). Quick tier motion: --model bytedance/seedance-2.0-mini --res 480p (§6.4).
  2. Timing for each clip: slop --json -p NAME clip status CLIP (its draft_notes). Write the whole book (§4.2). Screenplay for acting and motion, short for copy. Sung clips: the short sung form (§4.0, template in §5): the head quoting every word, one line of framing and light, at most 2 timed beats, then One continuous shot, no cuts, no head turns. Only she sings. No text, letters or logos., about 90 words; a long prompt makes the model re-sing instead of copy, so the drama goes in the entry's story line. Motion clips: the screenplay (§4.1), timestamps to 0.01 s on real word and kick times, blocking, action and reaction. A sung clip's plate shows the whole mouth, nothing in front of it (no microphone, hand or hair); else re-pick or re-roll it first (skill plates).
  3. Each clip: save its prompt to clips/prompts/CLIP.txt, slop --json -p NAME clip set CLIP --prompt-file clips/prompts/CLIP.txt, then slop --json -p NAME clip lock CLIP (fix what it names).
  4. G5: send the book with the total from slop --json -p NAME clip list, budgeted at 2x (§8). An explicit yes only, then slop --json -p NAME gate prompts --approve --note '<their words>'.

The tunnel (HTTPS for audio windows and start images)

  1. slop --json -p NAME gen tunnel start once. It needs 10-30 s: in a LATER call slop --json -p NAME gen tunnel status --verify must say reachable. Paid submits and media_import_url wait for that; never reachable: the media_upload fallback (step 8).

OpenRouter (§6.1, §6.3)

  1. slop --json -p NAME clip submit CLIP --dry-run; nothing blocks: slop --json -p NAME clip submit CLIP --yes (it hosts the window on the tunnel). Never resubmit a request that got a job id.
  2. slop --json -p NAME job start -- clip poll --wait 300, then wait by the profile's turn rule (70-130 s a clip).

Higgsfield connector (§6.2; an MCP call dies after 120 s idle, so submit, then poll)

  1. slop --json -p NAME clip submit CLIP --provider higgsfield --yes writes clips/CLIP/higgsfield-request.json. Do its how_to_run steps in order:
  • uploads, preferred: each uploads item's https_url (added while the tunnel runs; else slop --json -p NAME gen host CLIP --role start --tunnel for the start image and slop --json -p NAME gen host CLIP --tunnel for the window) goes to media_import_url {url, type}. Fallback, no tunnel: media_upload {filename}, curl -sS -o /dev/null -w '%{http_code}' -X PUT -H 'If-None-Match: *' --data-binary @FILE 'UPLOAD_URL' (must print 200), media_confirm (images and audio in separate calls). Record each media id: slop --json -p NAME gen host CLIP --role start --media-id ID (and --role audio);
  • preflight generate_video with get_cost; credits differ from the file: stop and tell the owner;
  • generate_video_batch with the request verbatim, then at once slop --json -p NAME clip ingest CLIP --job-id ID --pending.
  1. Poll with jobs_wait (timeout_seconds 15) every poll_after_seconds; each poll is a wait of the profile's turn rule (renders take 4-11 min). Completed: slop --json -p NAME clip ingest CLIP --url 'URL' --job-id ID --credits CREDITS (CREDITS = what transactions charged for that job). Failed: slop --json -p NAME clip ingest CLIP --failed '<the reason, verbatim>' (its credits come off the log), then the next step it prints. Rejected, no job id (how_to_run's REJECTED step): slop --json -p NAME clip ingest CLIP --failed 'rejected: <the reply, verbatim>', tell the owner, stop. Auth error: ask the owner to reconnect Higgsfield. A transport timeout: check transactions before anything else.
  2. Finalize a kept 480p draft (within 7 days, about 12 credits a second; the owner's yes on the get_cost price): slop --json -p NAME clip submit CLIP --provider higgsfield --finalize-from TAKE --yes, the same steps 8-9, then slop --json -p NAME clip sync CLIP --take N (the new take) before any edit or render: the edit uses it at once, unmeasured. A failed or refused finalize counts 0 and the draft stays; a retry, or a second finalize of one draft (--again), only on the owner's yes.

After the takes

  1. Declined by a face filter: slop --json -p NAME clip ingest CLIP --declined '<the error>'. Final for that image on that provider: never alter it. Allowed: the same image on the other provider, another candidate, a reworded prompt, or keep the 2.5D plate (docs/62-moderation.md §3).
  2. Sung clips (docs/61-lipsync.md §3-§8): slop --json -p NAME clip sync CLIP, then slop --json -p NAME clip coverage, then slop --json -p NAME clip cover CLIP --make (covers need their own lock and yes) or cut away.
  3. Routing for a sung window (docs/61 §8): takes from a long prompt that did not copy: rewrite it short and retake on the same provider first (steps 2-4). Then two OpenRouter takes of the short form that do not copy the lips, then a Higgsfield 480p draft (finalize it if kept, step 10), else cut away. Never a third short OpenRouter try.
  4. A clip made after the board plays once its shots link it (audit: plays on no shot): slop --json -p NAME board set SHOT --clip CLIP (a board change: the owner re-approves), then slop --json -p NAME edit edl. Real spend: slop --json -p NAME spend; Higgsfield: only what transactions charged (the project log is an estimate).

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/anabologyco-maker/claudia-bot/clips/badge)](https://verified-skill.com/skills/anabologyco-maker/claudia-bot/clips)

Something wrong with this skill or its page? Tell us.