podcast

name	podcast
description	Use when the user asks for podcast or a task matching the examples below. Two-host podcast video for any URL or free-form topic — 1 minute, 4 acts × ~15s, native multi-shot dialogue, optional voice cloning for Host A. Use when the user asks to "make a podcast", "podcast about [thing]", "podcast review of [url]", "two-host explainer", "interview-style clip", "two people talking on camera", "I/me and X talk about Y", or "interview with [persona] about [topic]". Native audio is the deliverable; captions are skipped by default because podcast dialogue mistranscribes domain terms.
argument-hint	<url-or-topic> [bg_img=] [host_a_img=] [host_b_img=] [voice_a=] [voice_b=] [aspect_ratio=16:9]

/pika:podcast

4 acts × 15s each = 60s. Host A always LEFT, Host B always RIGHT. Accepts a URL or a free-form topic / brief.

Parameters

Param	Default	Notes
input	required	URL to review or free-form topic / brief (e.g. "I and Elon Musk talk about Mars")
bg_img	auto-generated	Podcast studio background
host_a_img	auto-generated	Host A portrait — see Real-person handling below
host_b_img	auto-generated	Host B portrait — see Real-person handling below
voice_a	876341503281471517	Kling preset or cloned voice ID for Host A
voice_b	829837252279803904	Kling preset or cloned voice ID for Host B
aspect_ratio	16:9	Output aspect ratio

Cost transparency gate

Before any paid MCP call, call identity_balance({verbose: true}) once. Surface the current balance, recent burn rate, and remaining runway, then gate the run with this exact message:

Estimated cost: about 6,000-9,000 credits (~$60-$90) for four Kling v3-omni pro 15s acts, optional missing-asset image generation, one act corrective retry, concat, and post-flight analyze_media QA. This exceeds $5, so Reply proceed to continue or cancel to stop.

Do not call any paid MCP tool until the user replies proceed. If the user replies cancel, stop without generating. This is the only yes/no gate; after proceed, render the four acts and return the URL.

Defaults — fire fast, no mid-flow confirmation

• Use the param-table defaults silently for voices. voice_a defaults to the Kling preset 876341503281471517 and voice_b to 829837252279803904. Do not ask "which voice?" before firing — only honor explicit overrides (voice_a=, voice_b=).
• Auto-generate any missing host portraits silently (Step 1's archetype prompts). Do not ask "should I generate a host image?" — just generate.
• Only the cost transparency gate asks for proceed. After proceed, submit → render the 4 acts → return URL. Account credit balance + provider failover are the canonical guardrails. The --yes flag is accepted as a no-op for backward compatibility.
• Topic-mode personas (Step 3) — when the user names a real public figure, follow Step 4 (Real-person handling) silently: archetype portrait by default, no auto-generated photographic likeness, no question to the user about likeness rights.

Pre-generation wall-clock guard

Start a timer at skill start once the podcast input is resolved and the cost gate has passed. The first paid generation call is generate_image for missing background/host assets. If all assets are already provided, the first paid generation call is generate_reference_video. The first paid generation call must be invoked within 5 minutes of skill start. If you have not invoked the first paid generation call within 5 minutes of skill start, stop before any paid generation call and report failed_pre_generation_timeout with what you have so far: input mode, asset status, voice status, URL capture/WebFetch status, script draft status, and the exact blocker. Do not keep refining host archetypes, factual grounding, jokes, or act wording.

Print a single-line progress checkpoint after each prep stage and right before the paid generation call:

• Stage 1/4 done — input resolved and cost gate passed, locking missing-asset image prompts.
• Stage 2/4 done — missing asset prompts ready, calling image generation now. Use this right before the first paid generate_image call when any background or host image is missing. If all images are already provided, emit Stage 2/4 done — assets and voices ready, collecting URL/topic facts. instead and continue without image generation.
• Stage 3/4 done — script draft locked, preparing first act.
• Stage 4/4 done — first act prompt ready, calling Kling now.

Missing-asset image prompt iteration is maximum 2 passes before the first generate_image call. After the max 2 passes, ship what you have to generate_image; do not continue polishing host archetypes, studio background details, or persona styling. Script and act-prompt iteration is maximum 2 passes before Kling. After the max 2 passes, ship what you have to generate_reference_video; do not continue polishing jokes, interruptions, persona framing, or camera wording.

Local images on Claude Desktop

Claude Desktop can't pass inline-pasted images to MCP tools yet (Anthropic-side limitation). If the user pastes a photo inline, or mentions a local file they want as host_a_img / host_b_img, pause Step 1 and kindly send them this — something like:

Heads up — pasted images don't reach MCP tools on Claude Desktop yet (Anthropic limitation). Two easy options for your photo:

• Paste a URL if it's already hosted (Imgur, S3, your site) — fastest
• Attach the image file so I can upload it before generation.

When a local file arrives, convert it to a public URL with upload_asset and use the returned public_url as the parameter before Step 1. Already-hosted https://... URLs work as-is and skip this entirely.

If the user names a real public figure without attaching anything, do NOT auto-generate their likeness — Step 4 (Real-person handling) uses an archetype portrait instead.

Long-running task_status polling

When any long-running generation or edit call returns a task_id with or without an initial status, including {task_id}, {task_id, status: "queued"}, or an initial queued, running, or processing status, record the task id and start time immediately.

• Call task_status({task_id}) in a tight loop until terminal (completed | failed | cancelled). No manual sleep and no Bash polling; the worker holds each status call open.
• Emit ONE visible progress line every 60s while status is queued, running, or processing: Seedance i2v queued for {N}m {S}s... still processing. Replace the provider/stage label when polling Kling, image generation, clone voice, or concat tasks.
• On completed, unwrap the returned result URL and continue.
• On failed or cancelled, surface failure to the user with task_id, status, and the last status message.
• After 15 min total from the original submit, call task_cancel({task_id}) if the task is still non-terminal, then surface failure to the user. If cancel reports the task is already terminal, call status once more and report that terminal result.
• Do not submit a duplicate request while the original task is still queued, running, or processing.

Steps

0. Resolve input (empty-args menu)

Strip flags (--yes, --no-captions, etc.) and key=value parameters from $ARGUMENTS. If what remains is empty or whitespace-only, print this menu verbatim as your full response, then stop and wait for the user's next message — do NOT call any tool, do NOT proceed to Step 1, do NOT invent a topic or URL. If the stripped input is non-empty (a URL or any prose), skip this step silently and proceed to Step 1.

What would you like a podcast about? I can take any of:

• A website URL (product page, docs site, launch page) — e.g. https://pika.art
• A GitHub repo — e.g. https://github.com/anthropics/claude-code
• A blog post / article URL — e.g. a recent piece you'd like discussed
• A free-form topic or brief — e.g. "I and Elon Musk talk about Mars" or "two scientists debate AGI"

Reply with your choice and I'll generate a 1-minute two-host podcast video (4 acts × ~15s).

Tip: you don't need to type /pika:podcast — just say things like "make a podcast about ", "podcast review of ", or "I and talk about " and I'll fire this skill automatically.

When the user replies, treat their reply as the resolved input (URL or topic) and proceed to Step 1. Do not re-prompt.

1. Generate missing assets (parallel)

Generate only what's not provided. Default archetype prompts:

• bg_img — modern podcast studio, two chairs, warm lighting, no people, 16:9
• host_a_img — enthusiastic host, studio portrait, left-side framing, 1:1
• host_b_img — pragmatic skeptic host, studio portrait, right-side framing, 1:1

If the input mentions specific personas (Step 3), tune the archetype to match the persona vibe — see Real-person handling below.

2. Resolve voice IDs

Use voice_a / voice_b as given; otherwise fall back to the default Kling presets (876341503281471517 / 829837252279803904). If the user supplied a cloned voice ID via voice_a= / voice_b=, use it directly.

3. Parse input mode — URL vs topic

Strip flags (--yes, --no-captions, etc.) and key=value parameters from $ARGUMENTS. Inspect what remains.

URL mode — input contains a https?:// URL:

• Call capture_website on the URL for visual grounding only: product/page name, visible layout, screenshots, b-roll cues, and one jokeable visual detail.
• Call WebFetch on the same URL and use its full-page text as the script source of truth. Do not rely on the screenshot alone for facts.
• Extract from the combined inputs: product name, value prop, 2–3 specific features or facts, pricing, one jokeable detail, and any concrete quotes or claims.
• For long-form pages (Wikipedia, docs, blog posts, or articles), scan the WebFetch full-page text for deep sections beyond the lede. When at least 3 substantive deep sections are present, the script must mention at least 3 distinct sections beyond the lede (for example architecture, training, hallucinations, limitations, deployment, or safety), not just the opening summary.
• Use WebFetch text for factual anchors and capture_website for visuals. If they conflict, prefer WebFetch for factual claims and treat the screenshot as a visual reference.

Topic mode — input is free-form prose (no URL):

• Treat the whole input as the brief. Parse for:
  • Subject — what the conversation is about
  • Hosts — explicit if mentioned ("I and Elon Musk", "two scientists", "Joe and Sarah"); otherwise use defaults (enthusiastic host + skeptic host)
  • Angle — debate / interview / explainer / casual
  • Concrete facts — any specific claims, numbers, dates, quotes the user gave
• If no concrete facts are given, use 2–3 clearly framed observations or hypotheses to anchor jokes and the "wait, actually..." pivot. Do not present invented claims as facts; if factual accuracy matters for the topic, ask for a source or URL.
• If the user says "I and X" or "me and X", Host A = the user (represented by host_a_img= if supplied, else a generated host portrait) and Host B = X.

4. Real-person handling (topic mode only)

If the parsed input names a specific real public figure as a host (e.g. "Elon Musk", "Taylor Swift", "Joe Rogan"):

• Default behavior: do NOT auto-generate that person's photographic likeness. Generate an archetype portrait matching the persona vibe — e.g. "tech-billionaire-energy CEO at a podcast desk" for an Elon-style host, "pop-star aesthetic" for a Taylor-style host. Clearly inspired-by, not impersonation.
• Override: if the user explicitly provides host_a_img=<url> or host_b_img=<url>, use the provided image as-is. The user takes responsibility for likeness rights.
• Voices: same logic — default to a generic Kling preset; only use a cloned voice when the user provides one (voice_a= / voice_b=).
• Script tone: the dialogue can riff on the named persona's known public positions or vibe (e.g. Mars enthusiasm for Elon-style) — public-record opinions are fair game. Do NOT put specific defamatory, off-character, or fabricated-private-life statements in their mouth.

This guardrail keeps the skill creative ("I want a podcast where I argue with a tech CEO about Mars") without auto-generating deepfakes of named real people.

CJK / non-Latin handling

Use this section when the URL, topic, product name, host name, quote, or required script detail includes Chinese, Japanese, Korean, Arabic, Devanagari, Cyrillic, accented Latin, emoji, or any other non-Latin text.

• Script preservation: preserve user-supplied non-Latin characters exactly in the script state and in any factual recap. Do not romanize, transliterate, or translate product names, person names, slogans, or quoted phrases unless the user explicitly asks. If a pronunciation aid is useful, put it in nearby English prose, not as a replacement for the original characters.
• Kling voice IDs: voice_a= and voice_b= must still be valid Kling voice IDs. For Chinese / Japanese / Korean dialogue, prefer a Kling voice ID or cloned voice sample known to speak that language; if no language-matched Kling voice is available, keep the original characters in the dialogue and surface that pronunciation may be accented instead of silently swapping to an unrelated TTS provider.
• Mixed-language dialogue: keep each host's line short and unambiguous. Avoid asking Kling to pronounce long mixed-language paragraphs; split complex names or quotes across acts and keep surrounding English simple.
• Captions / font fallback: this skill skips add_captions by default because podcast dialogue mistranscribes jargon. If the user explicitly asks for captions anyway, pass manual subtitles[] from the authored script rather than auto-transcribing, set font: "noto-cjk" for Chinese / Japanese / Korean, and preserve the original non-Latin text exactly.
• QA: before delivery, include a one-line note in the final verdict when non-Latin text was present: whether original characters were preserved in the script and whether the selected voices were language-matched or best-effort.

Jargon-heavy native audio handling

Use this section when the URL, topic, product name, quote, or required script detail is domain-heavy or jargon-heavy, especially finance, law, medicine, AI/crypto, acronyms, coined terms, unusual product names, long numbers, or multi-word compounds. Kling-omni native TTS can garble coined terms in the spoken native audio, not just in auto-captions, so reduce the risk before the paid Kling calls.

• Risk scan: mark the run jargon-heavy when a line would contain several domain-specific terms, acronyms, long numbers, or nested / multi-clause economist-style phrasing. Treat examples like stablecoins, unbundle, and reassure as load-bearing terms that must remain intelligible.
• Script rewrite: preserve the canonical term in the script state and final recap, but shorten the spoken line around it. Use short lines, split long clauses, keep at most one or two high-risk terms per host line, and avoid nested clauses that make Kling infer similar-sounding filler words.
• Pronunciation aid: create a small per-act pronunciation aid for high-risk terms, for example stablecoins = STAY-bul coins, unbundle = un-BUN-dul, reassure = ree-uh-SHOOR. The pronunciation aid is prompt metadata only; do not replace the canonical term in the dialogue, captions, facts, or final recap.
• Known-hard-term correction: if a term is known to drift, or post-flight localizes a near-neighbor mishearing, do not retry the same standalone hard word. Keep the canonical term in act metadata (canonical_terms: ["unbundle"]), then make the corrected voice-token line a short context phrase with a speakable cue / micro-pause, for example we un-bundle payments. Add the observed wrong form as a negative pronunciation guard in metadata, for example unbundle = un-BUN-dul, not un-bumble.
• Known-hard publishable fallback: before the paid Kling calls, mark observed hard terms such as unbundle as known-hard when prior E2E or the current post-flight findings show drift like unbundle -> unbubble, un-bumble, or unbundable. Record canonical_terms and observed_wrong_forms. Do not render known-hard spoken terms with Kling native audio. Instead, generate the affected visual act or final video with no spoken known-hard term in the native voice-token lines, then generate controlled narration with generate_speech(text=<spoken-only script>, provider: "minimax-tts", minimax_model: "speech-2.8-hd", language: "en") using the canonical spelling (unbundle, not the hyphen cue). The spoken-only script must preserve the authored host-turn order but contain no HOST_A/HOST_B labels, no metadata, no pronunciation aids, and no observed wrong-form spellings. Run pre-replacement QA on the TTS audio with both transcribe_audio(audio=<tts_audio_url>) and analyze_media(media=<tts_audio_url>); every required canonical_terms entry must appear in the transcribe_audio text and must appear in the analyze_media observed transcript. Reject the TTS if a canonical term is missing, replaced by any observed wrong form, or replaced by a new near-neighbor. Then call edit_audio_replace(video_url=<final_mp4>, audio_url=<tts_audio_url>, duration_policy: "video") to discard the native audio and replace it with the controlled TTS track. Run post-replacement QA on the replaced MP4 with both transcribe_audio(audio=<replaced_mp4_url>) and analyze_media(media=<replaced_mp4_url>); every required canonical_terms entry must appear in both final checks, and analyze_media must also verify host-turn order and acceptable audio/video sync. If any final check hears a wrong form, misses a canonical term, or says the fallback collapses the podcast turn structure, return not publish-ready; if both checks pass, the fallback output is publish-ready even though it uses controlled TTS instead of Kling native speech.
• Numbers and acronyms: write spoken-friendly forms when they matter: $175B becomes one hundred seventy-five billion dollars; API becomes A P I unless the brand normally says it as a word.
• Act prompt: include the per-act pronunciation aid near the voice-token lines with an instruction that it is not dialogue and should not be spoken verbatim. Ask for clear pronunciation of the canonical terms and no improvising similar words.

5. Write script

Write 4 acts × 2 lines (HOST_A / HOST_B). Each line ~10–12s of spoken dialogue.

Required (apply to both URL and topic modes):

• One specific joke tied to a concrete detail (scraped fact in URL mode; topic-derived claim in topic mode)
• One "wait, actually..." skeptic-flip moment
• At least one mid-sentence interruption
• Natural filler: "okay so", "wait", "right?", "i mean", "honestly"
• Real reactions, not generic praise
• Reference at least one actual feature name, price, claim, or quote
• Natural ending — no forced "bye!"
• Apply Jargon-heavy native audio handling before finalizing the act lines when the topic is domain-heavy, jargon-heavy, or uses coined terms.

Acts: Hook → Feature deep-dive → The Turn → Verdict (In topic mode the analogue: Hook → Substance → The Pivot → Verdict.)

6. Generate video acts (subagent, sequential)

Delegate to a subagent with all resolved assets and the script. The subagent runs acts 1→2→3→4 sequentially — do NOT parallelize. The subagent must follow the Long-running task_status polling contract above and relay every 60s progress line back to the parent while it is waiting on Kling, image generation, voice clone, or concat tasks.

Each normal act: one generate_reference_video call (kling-v3-omni, duration=15, sound=true, quality_mode: "pro"). Pass reference_images=[bg_img, host_a_img, host_b_img], voice_ids=[voice_a, voice_b], and quality_mode: "pro" on every normal act; this must pass quality_mode: "pro" because the cost gate quotes the pro-tier 15s act cost. Optional knob: kling_model to pin a specific kling family member if you need reproducibility across runs. Three shots:

• Known-hard fallback exception: for acts containing known-hard spoken terms, render the affected visual act with sound=false and no voice tokens for the affected lines. Do not include canonical known-hard terms in any <<<voice_*>>> native voice-token line; the controlled TTS replacement supplies those words later. For jargon-heavy but not known-hard acts, include the per-act pronunciation aid from Step 5 in the act prompt as non-dialogue metadata and preserve canonical terms in the quoted <<<voice_*>>> lines.
• Wide 5s: both hosts, no voice token
• MCU-A 5s: <<<voice_1>>> '<HOST_A line>'
• MCU-B 5s: <<<voice_2>>> '<HOST_B line>'

Emotional beats per act:

• Act 1: A excited, B skeptical
• Act 2: A gesturing/explaining, B questioning
• Act 3: A firm, B surprised and reconsidering
• Act 4: A satisfied, B conceding

After act 4, subagent calls edit_concat([act1, act2, act3, act4]), relays any async polling progress to the parent, and returns the final video URL. Keep the four act URLs in state so the post-flight quality gate can spend at most one targeted act correction without regenerating clean acts.