Saved Reels Research Pipeline

Key ideas

Incremental collection

A seen-state file means only newly saved Reels download; empty runs cost nothing.

Local transcription

Speech is transcribed by an on-device Whisper model; no audio is uploaded anywhere.

Interactive onboarding

One wizard sets up an isolated environment, your session, and your collection — no manual configuration.

Claim verification

Every tool a Reel promotes is checked against its official repository or documentation before it counts as real.

Safety gate

Only pure prompt/workflow skills may auto-install; anything with code, network, or credentials is proposed to a human first.

A practical starting sequence

1. Install the skill from the official repository and run the onboarding wizard in a real terminal.
2. Let the wizard create an isolated Python environment, verify your Instagram session, and pick the saved collection to watch.
3. Schedule the incremental collector — it prints NO_NEW_REELS and stops when nothing is new.
4. Analyze each run locally: bounded frame samples plus a timestamped on-device transcript per Reel.
5. Extract every claimed tool, skill, or project and verify it against its primary source; label claims verified, partly verified, unsupported, or misleading.
6. Apply the safety gate: auto-adopt only prompt-only skills; write up everything else as a proposal with its blocking reason.
7. Record dated findings and sources in a durable research log — never credential values.