version: "1.0.0" name: render-video description: Produce a video deliverable — video abstract, grant-pitch video, results explainer, or teaser — by invoking a separately-installed OpenMontage as an external subprocess. Use when the user says "make a video", "video abstract", "render explainer", "grant video", "video pitch", "results explainer", or "make a teaser". Owned by the Reporting council; consumed by Outreach. Requires OpenMontage installed separately (see PROVISION.md). No video renders without human approval at G5/G6.

render-video — evidence-bound video production via OpenMontage (subprocess)

This skill lets Cambium produce a video deliverable by shelling out to a separately-installed, AGPLv3-licensed OpenMontage process. Cambium itself remains MIT-licensed; the license boundary is the OS process boundary.

1. Ownership and governance

2. License boundary — non-negotiable

OpenMontage (https://github.com/calesthio/OpenMontage) is AGPLv3. Cambium is MIT.

The boundary is the process boundary:

• Cambium ONLY calls OpenMontage as an external subprocess (CLI) via OPENMONTAGE_HOME.
• No OpenMontage source code, skill text, prompts, or vendored files may be copied,

pasted, forked, or transcribed into this repository.

• Hosting OpenMontage as a network-accessible service triggers AGPL §13 on the

operator — this must be disclosed to any team running it as a server. Running it locally on the same machine that invoked it does not.

• References to OpenMontage in this skill use only its public command names and

file conventions from its published README.

3. Inputs are evidence-bound

The video script and on-screen data may show only facts and numbers that carry a Proved or Code-verified tier in the findings ledger. Pass the ledger IDs (source_ledger_ids) in the VideoDeliverableRequest.

• Asserted or Open findings must NOT appear as on-screen claims.
• No fabricated statistics, illustrative numbers, or paraphrased results that change

the meaning.

• All source data is cross-referenced in governance/provenance.json before render.

4. Outputs fold back into Cambium governance

After every render, capture into Cambium:

1. `governance/provenance.json` — append an entry with:

• providers/models used by OpenMontage (from its decision log)
• actual cost vs budget cap
• decision_log_ref (path or URL to OpenMontage's own log file)

1. AI Use Statement (AI_USE_STATEMENT.md) — disclose:

• which AI providers and which models were used for narration, imagery, music
• what each was used for (voice synthesis, B-roll selection, colour grading, etc.)
• that a human verified the output before G5/G6 approval

1. Asset disclosures (ai_assets_disclosed) — every AI-generated or

AI-selected asset listed explicitly (no silent AI imagery).

1. Stock / archive licenses — verify that every third-party clip, image, or

audio track used from Pexels, Archive.org, NASA, or Wikimedia carries a license compatible with the intended distribution platform; record the license for each.

1. Budget cap — cost_actual_usd must not exceed budget_cap_usd; if it

would, STOP and bring back to the Director before rendering.

5. Consent and safety guardrails

• No non-consensual synthetic likeness or voice. If a real person's voice or face is

synthesised, obtain and document explicit written consent before rendering.

• Honor each AI provider's Terms of Service; flag any ToS incompatibility to the

Director before use.

• Do not use real_footage_only=false for grant submissions to funders that prohibit

AI-generated imagery without explicit disclosure; check funder rules first.

6. Flow

Reporting drafts VideoDeliverableRequest
  ↓  (validate source_ledger_ids — must be Proved/Code-verified)
  ↓
Human approval — Gate G5 (internal release) or G6 (external publish)
  ↓  APPROVE → proceed  |  REVISE → rework  |  REJECT → stop
  ↓
Skill shells out:
  $OPENMONTAGE_HOME/pipeline_defs/ <profile> render \
      --request <request.json> \
      --output  <output_path>
  ↓
OpenMontage runs in its own process (AGPLv3 boundary preserved)
  ↓
Skill captures VideoResult → provenance.json + AI Use Statement
  ↓
Human reviews rendered output — G5 (release) or G6 (publish)

7. Schema reference

Input/output contracts: skills/render-video/video_contract.schema.json Provisioning: skills/render-video/PROVISION.md

8. Self-review checklist (before returning VideoResult)

• [ ] All source_ledger_ids resolve to Proved or Code-verified findings.
• [ ] cost_actual_usd <= budget_cap_usd.
• [ ] Every AI-generated asset is listed in ai_assets_disclosed.
• [ ] Stock footage licenses verified and recorded.
• [ ] No non-consensual synthetic likeness or voice.
• [ ] decision_log_ref written to governance/provenance.json.
• [ ] AI Use Statement updated.
• [ ] self_review_passed: true only when ALL checks pass.