orchestrate-content-run

name	orchestrate-content-run
description	Spawn and monitor a full content production run (research → fact-check → psychology → story) for one topic opportunity, managing the iterative quality-score loop until it clears threshold. Use when the Chief Editor agent must turn an approved topic opportunity into a complete asset bundle, or when asked to "produce", "run", or "orchestrate" content for a topic.
metadata	{"agent":"chief-editor","source":"Project Atlas Agent Skills Manifest §1","layer":"executive"}

orchestrate-content-run

Direct the research → writing → design → analytics loop for a single topic, ensuring quality scores exceed threshold before human review.

When to use

• A topic opportunity has been approved (from Trend Intelligence / DS-Star backlog) and needs production.
• You must coordinate the Research, Fact-Check, Psychology, and Story agents on one run.

Workflow

1. Call the temporal-bridge MCP tool start_workflow("qualityLoopWorkflow", "quality-loop", [{"topic": topic, "subjectEntity": subject_entity, "durationMinutes": duration_minutes}]).
2. Call get_workflow_result(workflow_id) -- blocks until the workflow finishes its (max 3) durable revision loop: gather-citations + generate-psych-profile -> draft-video-script -> verify-claims, scored deterministically (100 - 30*high - 10*medium - 5*low per flagged contradiction, not by asking a model to grade itself), feeding contradictions back into the next draft as revision notes.
3. Result is {contentAssetId, finalScore, passed}. On passed: true, hand off to submit-editorial-review with contentAssetId. On passed: false (3 iterations exhausted), the content_assets row is left NEEDS_REVISION for a human to look at directly rather than looping forever.

Function signature (manifest contract)

def orchestrate_content_run(topic_opportunity: dict) -> dict:
    """
    Returns qualityLoopWorkflow's result, exactly as the Workflow section above
    describes: {"contentAssetId": str, "finalScore": float, "passed": bool}.
    On passed=True, hand contentAssetId to submit-editorial-review.
    """

Inputs / Outputs

• Input: topic_opportunity (dict from the Trend Intelligence / backlog table).
• Output: asset bundle + quality report (see signature).

Backend dependency

• content_assets table (Railway Postgres) is live; qualityLoopWorkflow creates the row at the start of the loop (not just at editorial-review time), so iteration state and accumulated revision feedback are durable -- a worker restart resumes exactly where it left off instead of losing all progress, the same way every other Temporal workflow in this codebase is durable.
• agent_runs table is not used; content_assets.status (DRAFT/NEEDS_REVISION/UNDER_REVIEW/ APPROVED/REJECTED) tracks the run's state instead.

Model

Runs on the Chief Editor's reasoning model (gemini-direct/gemini-2.5-flash) — judgment/coordination.