induct-media

namespace	aiwg
platforms	["all"]
name	induct-media
description	Induct acquired audio/video, transcript sidecars, and source metadata from media-curator into research-complete REF artifacts with timestamp citations and storage policy records
userInvocable	true
triggers	["induct media research source","video audio research induction","media research handoff","timestamp citation transcript","media REF template","turn acquired media into research corpus REF"]
commandHint	{"argumentHint":"<media-metadata.json|media-path> --transcript <path> --source-url <url> [--storage copied|lfs|object-storage|hash-only] [--ref-id REF-XXX] [--dry-run]","allowedTools":"Read, Write, Glob, Grep, Bash","model":"sonnet","category":"research"}

Induct Media

Create research-complete artifacts from media-curator acquisitions: a media REF, citation sidecar, radar sidecar, index/search metadata notes, and optional profile backlinks. This is the media-curator to research handoff for videos, lectures, podcasts, talks, and interviews.

Triggers

• "induct media research source"
• "video audio research induction"
• "media research handoff"
• "timestamp citation transcript"
• "media REF template"
• "turn acquired media into research corpus REF"

Inputs

Required:

• Acquired media metadata or local media path from media-curator.
• Transcript sidecar path from transcribe-media.
• Source URL.

Optional:

• --ref-id REF-XXX to use an assigned identifier.
• --storage copied|lfs|object-storage|hash-only to choose archive policy.
• Title, platform, duration, publication/upload date, speakers, channel/venue, license, acquisition ID, media hash, transcript hash.
• Existing citation mentions extracted from the transcript.

Storage Policies

Choose one policy and record it in the REF frontmatter:

Policy	Use when	Required record
copied	File is small enough for normal corpus storage	media/video/ or media/audio/ path and SHA-256
lfs	Git LFS is enabled for large binaries	LFS path, pointer committed, SHA-256
object-storage	S3/WebDAV/object backend stores binaries	Object URI, backend name, SHA-256
hash-only	Corpus cannot store or redistribute the media	Source URL, media SHA-256 when locally available, acquisition note

Never copy media when license or platform terms do not allow redistribution. Hash-only records are valid when source integrity and transcript provenance are enough for corpus analysis.

Output Artifacts

Write or update:

• .aiwg/research/findings/REF-XXX.md from templates/reference-media.md
• .aiwg/research/citations/REF-XXX-citations.md from templates/citation-sidecar.md
• .aiwg/research/radar/REF-XXX-radar.md from templates/radar-sidecar.md
• .aiwg/research/index.md or the configured index/search metadata
• Speaker/channel/venue profile backlinks when profiles already exist

Recommended media layout:

• media/video/REF-XXX-speaker-year-short-title.mp4
• media/audio/REF-XXX-speaker-year-short-title.mp3
• media/transcripts/REF-XXX-speaker-year-short-title.transcript.json

Execution Flow

1. Read acquisition metadata and transcript sidecar.
2. Verify the transcript sidecar has schema: aiwg.media.transcript.v1, segments[], source.sha256, and transcript.sha256.
3. Resolve or assign REF-XXX.
4. Determine source type (video, audio, podcast, lecture, interview, or talk) and select reference-media.md.
5. Apply the selected storage policy. Compute missing SHA-256 values when the local file is readable.
6. Generate the media REF with source URL, platform, duration, upload date, speakers, channel/venue, transcript path/hash, media storage, and license.
7. Generate citation sidecar:
   • normal paper-style outgoing/incoming tables when formal references exist
   • spoken-reference table for timestamped verbal mentions and unresolved informal references
8. Generate radar sidecar with media-specific GRADE rationale.
9. Update index/search metadata so research-query can find the REF by title, speakers, source URL, channel/venue, source type, and transcript text when available.
10. Report verification limits and required human review before publication quotes are used.

Citation Rules

Timestamp citations use:

@.aiwg/research/findings/REF-XXX.md @ HH:MM:SS - "exact transcript quote"

The timestamp MUST exist in the transcript sidecar, and the quote MUST exactly match the transcript text. Keep spoken references distinct from formal paper citations; unresolved verbal mentions remain valid sidecar rows with Inducted REF set to -.

Worked Fixture

Minimal media-curator output:

{
  "schema": "aiwg.media.acquisition.v1",
  "title": "Example Lecture",
  "source_url": "https://example.invalid/watch?v=lecture",
  "platform": "YouTube",
  "media_path": ".aiwg/media/acquisitions/example-lecture.mp4",
  "sha256": "sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
  "duration": "00:42:00",
  "speakers": [{"name": "Doe, Jane"}],
  "license": "platform ToS only"
}

Transcript from transcribe-media:

{
  "schema": "aiwg.media.transcript.v1",
  "source": {
    "path": ".aiwg/media/acquisitions/example-lecture.mp4",
    "url": "https://example.invalid/watch?v=lecture",
    "sha256": "sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
  },
  "transcript": {
    "sha256": "sha256:bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb",
    "language": "en",
    "tool": "human",
    "quality": {"limitations": []}
  },
  "segments": [
    {"id": "seg-000001", "start": "00:00:05", "end": "00:00:12", "speaker": "SPEAKER_00", "text": "Exact words from the lecture."}
  ]
}

Expected research outputs:

• REF-XXX.md records source URL, platform, duration, speaker, transcript hash, media hash, license posture, and key timestamps.
• REF-XXX-citations.md includes any formal references plus timestamped spoken mentions.
• REF-XXX-radar.md records media GRADE and refresh cadence.
• Index/search metadata includes source type, speaker, channel/venue, source URL, transcript path, and storage policy.

A complete, runnable worked example — sample.acquisition.json (from media-curator /acquire) and sample.transcript.json (from /transcribe-media) plus the three expected output artifacts — lives in examples/. See examples/README.md. The acquisition fixture is byte-identical to media-curator's acquire/examples/sample.acquisition.json, so the input contract is exercised end to end.

References

• @$AIWG_ROOT/agentic/code/frameworks/media-curator/skills/transcribe-media/SKILL.md
• @$AIWG_ROOT/agentic/code/frameworks/research-complete/templates/reference-media.md
• @$AIWG_ROOT/agentic/code/frameworks/research-complete/templates/citation-sidecar.md
• @$AIWG_ROOT/agentic/code/frameworks/research-complete/templates/radar-sidecar.md