tiktok-video · diff
git:20260908.ee795e9 to git:20260908.dbbaaf6
29 added, 15 removed. Audit A to A.
---
name: tiktok-video
description: Create an original 1–120 second portrait video through brief-specific research, accepted direction and audiovisual review.
disable-model-invocation: true
---
Require: subject, audience, intent, duration in 1–120 seconds, brand and assets
(explicit none allowed), git workspace, provider and license constraints,
authoring-owner pointer and per-call bound. Carry the brief in Goal and all
constraints and the owner pointer in governed downstream Context.
tickets.py frame-open <run> --goal-file <video-goal> --workflow tiktok-video
--context-file <video-context>
Read [admission](references/admission.md) for required input preparation,
verifiers and partial-result handling; [inventory](references/inventory.md)
identifies the package boundaries and qualified resources.
- **Research first.** Invoke `super-research` for one bounded brief-specific
- question: which market/reference observations inform this subject, audience and
- intent, and does current Remotion-versus-HyperFrames evidence change the renderer
- choice? Supply named live primary sources, window where applicable, frozen as_of
- at or after reads and a hard per-step cap. Ask for actual available playback
- inspection, citations, dates, exposed metrics with sampling limits, contradictions
- and gaps. Reuse applicable prepared evidence explicitly with provenance; refresh
- only decision-sensitive gaps. This public call resolves its own standards.
+ **Research first.** Own two ordinary `orch-do` lanes under `orch-research`,
+ using the questions, source policy, bounds and evidence carriers in
+ [admission](references/admission.md). Run market/reference and renderer research
+ in parallel when useful. Assess prepared evidence for relevance and gaps;
+ refresh only decision-sensitive gaps.
- tickets.py frame-open <run> --parent <frame> --goal-file <research-question> --workflow super-research
+ tickets.py do <run> --parent <frame> --standard orch-research
+ --goal-file <market-question> --context-file <market-context> --bound <per-call-bound>
- Consume the actual dossier identity and coverage findings, never a planned path.
- Missing required verification returns partial research and prevents an accepted
- handoff. Pin one main renderer and dependency identity before original artifact
- production, carrying the evidence, license branch and decision into both helpers.
+ tickets.py do <run> --parent <frame> --standard orch-research
+ --goal-file <renderer-question> --context-file <renderer-context> --bound <per-call-bound>
+ The market packet returns cited dated observations and available viewing evidence;
+ the renderer packet returns the current comparison and constraints/pins.
+ Compare actual packets after both exist, preserving provenance and gaps without
+ requiring format conversion.
+
+ tickets.py judge <run> --parent <frame> --standard orch-research
+ --artifacts evidence:<market-id> --artifacts evidence:<renderer-id>
+ --goal-file <coverage-goal> --context-file <research-context> --bound <per-call-bound>
+
+ Use the makers' exact research standard digest for independent coverage review.
+ It checks the original question, both lanes and their relevance/gap assessment.
+ Only blocker-free coverage of the required decisions permits an accepted
+ handoff; missing required evidence preserves partial packets and findings.
+ "Where the judge blocks, one repair `do` is handed the
+ `findings:` line verbatim, then one re-judge; two rounds is the bound."
+ A repair remains a single research lane. Pin one main renderer and dependency
+ identity before original artifact production, carrying the accepted packet
+ identities, coverage findings, license branch and decision into both helpers.
+
**Direction.** Invoke `video-direction` with the complete brief, actual research
identity and gaps, renderer constraints/pins, rights/provider limits, target
document directory, voice contract, numeric document budget, citation policy,
- per-call bound and outside document verifier. At 1s reduce to one readable idea,
- explicitly deciding voice/CTA omission; at 120s budget explanation and proof.
+ per-call bound and outside document verifier.
tickets.py frame-open <run> --parent <frame> --goal-file <direction-goal> --workflow video-direction
--context-file <direction-context>
Drive each invoked body in its opened frame, without opening its entry twice.
Only its independently accepted document and actual findings feed
`video-production`, with the brief, renderer decision, assets/rights/provider
constraints, git workspace, owner pointer and per-call bound.
tickets.py frame-open <run> --parent <frame> --goal-file <production-goal> --workflow video-production
--context-file <production-context>
Preserve partial returns and unresolved listening/license findings; a technical
probe does not establish audiovisual acceptance.
Never: invent accepted identities, rank efficacy from exposed counts, force
variants or cut quotas, copy unlicensed expression, authorize paid providers or
publication implicitly, or claim admission from static checking.
Return: `tickets.py frame-close <run> <frame> --done <outside-output-probe>`;
research and direction identities, fixed production artifact and independent
findings, renderer/package/standard pins, preview/capture/audition evidence,
outside probe exits and gaps (`[]` when none), preserving partial results.