agentleFS
mantleCurvemantleCurve/claude-skills/skills/episodic-video-merge/SKILL.md

episodic-video-merge

Skill0 starsChanged 10 days agoMITView on GitHub

Download an accessible episodic video series, inspect and trim frozen endings, test selected episodes, and merge a complete MP4 with one chapter per episode. Use for numbered episode downloads, smooth episode joins, or chaptered compilations from existing local videos.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read download-and-recovery.md for stream discovery, HLS variants, resumable downloads, and the HEVC parameter-set failure encountered in …

    line 18
  2. Distinctive:

    Read timing-and-validation.md before changing trim or merge logic.

    line 40
  3. Distinctive:

    After the user requests applying an accepted setting to the complete series:

    line 54

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Episodic video download, trim, and merge

Produce a numbered, verified source collection and a new MP4 with accurate episode chapters. Use Python 3.10+, FFmpeg, ffprobe, and curl locally. The scripts need no Python packages. They do not require a video-authoring framework or a hosted service.

Establish the series and scope

Use the current conversation and existing project files to determine the series, episode count, output folder, language/audio choice, and requested range. Preserve original episodes and earlier deliverables. A request for “just episodes 4–8 to test” produces only that test; an acknowledgment of the test records the accepted setting without silently launching a full rebuild.

Inspect existing manifests, downloads, and scripts before revisiting a site. Do not infer episode URLs by incrementing an opaque asset ID. Record the actual page/stream mapping and verify that every expected number occurs once. An accessible subset is not a complete series.

Keep videos paused and autoplay disabled unless the user asks for playback. Prefer direct reads and local tools. When browser interaction is needed, use the environment's permitted browser tool and its documented APIs. Do not repeatedly click play or follow advertising redirects. Existing approval for a particular site's required interaction applies only within that task. Do not bypass login, payment, DRM, or browser security blocks.

Read download-and-recovery.md for stream discovery, HLS variants, resumable downloads, and the HEVC parameter-set failure encountered in this workflow.

Download and verify the sources

Create a local sources.json with observed HTTP(S) media URLs and sequential episode numbers. Keep signed URLs, cookies, headers, and media out of the skill repository. The helper supports direct files and simple, unencrypted, muxed MPEG-TS VOD HLS; use the reference's alternate path for other formats.

python3 /path/to/episodic-video-merge/scripts/download_episodes.py \
  --manifest sources.json --expected-count 60 --output episodes

The downloader selects the largest HLS variant, fetches segments with bounded retries, retains in-band HEVC configuration using hev1 when appropriate, and fully decodes every completed episode. It records hashes and only reuses a verified file whose hash still matches. Do not re-download good originals merely to change a trim.

Inspect endings and choose the cut

python3 /path/to/episodic-video-merge/scripts/episode_media.py audit \
  --episodes episodes --output ending-audit.json --range 4:8

The audit reports stream start times, frame rates, exact repeated decoded tail frames, and near-static detection events over the last four seconds. These are evidence, not an automatic verdict: subtle motion, compressed stills, or deliberate dramatic pauses can fool a threshold. A clean decode and continuous timestamps do not establish that an ending looks smooth.

Read timing-and-validation.md before changing trim or merge logic. Keep trim choices in frames, with their corresponding seconds. Round an additional requested cut up to a whole frame and add it to the previous cut; do not repeatedly round a nominal total and lose an increment.

The originating 25 fps series used 25 frames (1.00 s), then 38 (1.52 s), then 51 frames (2.04 s) from every ending. The user accepted the 51-frame episodes 4–8 test. This is a known preset for that series, not the default for unrelated footage. A separate episode had a 42-frame exact freeze; inspect outliers rather than assuming all tails match.

Make the requested test or complete merge

Build directly from original episodes to avoid repeatedly encoding an already lossy test. Choose a new filename for each revision. --trim-end means the total cut from each original ending, not an incremental cut. --trim-overrides accepts a JSON object mapping episode numbers to total trim seconds for outliers.

python3 /path/to/episodic-video-merge/scripts/episode_media.py build \
  --episodes episodes --range 4:8 --trim-end 2.04 --normalize-timing \
  --output Series_Episodes_04-08_Test_Trim_2p04s.mp4 --title "Series — trim test"

After the user requests applying an accepted setting to the complete series:

python3 /path/to/episodic-video-merge/scripts/episode_media.py build \
  --episodes episodes --expected-count 60 --trim-end 2.04 --normalize-timing \
  --output Series_All_60_Trim_2p04s.mp4 --title "Series — complete"

The default encoder is portable libx264 at CRF 18. On a supported Mac, --encoder h264_videotoolbox uses a configurable hardware bitrate (default 4 Mbps). This is a re-encode; do not call it lossless. The helper requires matching dimensions and constant frame rates, strips inherited container metadata/chapters, normalizes video starts, preserves audio's offset relative to picture, and uses PCM audio intermediates so individual AAC priming delays do not accumulate. It encodes the final AAC track once and embeds one chapter per included episode, retaining the original episode numbers.

--normalize-timing repairs small source timestamp jitter, capped at half a frame of cumulative deviation from the native frame grid. This was needed for the originating series. Omit it for strict timing checks; larger gaps or variable frame rates still stop the build. The report records the measured deviation.

“Stripping” here means the requested tail cuts and replacement of inherited container metadata. Do not silently remove subtitles burned into the picture, watermarks, opening scenes, credits, or speech. Separate audio/subtitle tracks require explicit selection before using the helper, which selects the first video and audio stream.

Validate and hand back

The build checks all output video timestamps, frame count, duration, chapter titles/starts/ends, and full audio/video decoding before renaming its .partial.mp4. It writes a verification report and chapter text beside the result. If validation fails, retain logs and intermediates; do not label the output complete. Fix the specific cause rather than suppressing warnings globally.

For a trim test, deliver the file, actual extra and total cut, included episodes, and chapter count. Let the user judge the remaining pause in their player. For a full merge, also state duration, size, and any missing episodes or unverified conditions. Do not autoplay the deliverable.

The helper removes only its own intermediates after success unless --keep-work is set. Originals, earlier cuts, manifests, and final reports remain. Never commit downloaded videos, local source manifests, or authentication material when sharing this skill.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.