version: "1.0.0" name: vidxp-find-video-evidence description: Use VidXP to search indexed videos and surface inspectable evidence boards, keyframes, and clips before analysis. Trigger for requests such as "find where X appears," "when does Y happen," "what is said," "what happens," or "show me the matching clip," even when the user does not name VidXP. Favor one-pass evidence delivery and only add brief accuracy feedback; do not trigger for ingesting new media or ordinary video editing.

Find video evidence with VidXP

Workflow

1. Resolve the vidxp MCP tools, then call get_workspace. If the requested

video is not indexed, explain that it must be indexed first.

1. Submit one retrieval job. Use search_moments to locate moments; use

query_video only when the user asks for a synthesized answer. Use command.query with search_moments and command.question with query_video. Set command.media_id when the user means one video.

1. In that initial job, put exactly this inside command:

"evidence_delivery": {"mode": "keyframes_and_clips", "max_items": 3}. This prepares the ranked board, standalone keyframes, and clips without a second retrieval pass. Never send command.materialize.

1. Call wait_job for bounded waits. Pass its observation_token as

after_observation_token on the next wait. When terminal, call get_job_evidence once. It returns the concise evidence index and visual content without the full structured job dump. Search and query may take time; update the user when the stage changes or about once per minute, never after every wait and never with an invented ETA.

1. Surface the returned board, keyframes, and clips immediately. Do not call

get_job, repeat the search, materialize more evidence, create another board, or perform a self-directed verification loop before showing the initial evidence.

1. Stop after the first evidence delivery. Only when the user explicitly asks

for another selection or format, use tile evidence IDs with materialize_job_evidence, or use create_evidence_board for a custom selection or next_start_rank continuation.

Actor scope

• Actor data is available through query_video, not name search. It represents

anonymous, video-scoped face clusters—not a named or cross-video identity.

• Treat its image as a representative full frame, not an exact face crop or

proof of continuous presence. Do not claim exhaustive named appearances.

• For named-person requests, surface the best scene candidates immediately and

label uncertain matches as candidates. Do not delay delivery while trying to prove identity through additional searches.

Output

• Lead with evidence, not a search narrative: first embed the returned board or

frame or provide its working resource link, then list the ready clips and keyframes. Use the returned local_path or download_url; never write an unlinked label such as “View evidence board.” Use get_artifact_download only if neither is returned.

• After the evidence, add at most a brief accuracy note. State uncertainty or

visible mismatches without launching another search. Accuracy feedback must not replace or precede the evidence.

• Preserve the source job and evidence IDs. Describe scores as retrieval scores,

and distinguish a visible appearance from a dialogue or caption mention.

• Stop waiting on success, failure, or cancellation. An empty result means no

matching indexed evidence was found, not that the event is absent from the original video.