Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/ykdojo/yt-transcript/claude-mdgit clone --depth 1 https://github.com/ykdojo/yt-transcriptWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00498 | $0.00498 |
| Opus 5 | $0.00249 | $0.00249 |
| Sonnet 5 | $0.00100 | $0.00100 |
| Haiku 4.5 | $0.00050 | $0.00050 |
Grade A, and why
yt-transcript CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
yt-transcript
Plugin providing a yt-transcript skill: plain-text transcripts for video URLs.
Layout
scripts/transcript.sh— the whole implementation. Captions first, Parakeet fallback.skills/yt-transcript/SKILL.md— how Claude invokes it..claude-plugin/— plugin and marketplace manifests. Bumpversionin both.
Constraints that are easy to break
- Never raise
CHUNK_SECSabove ~400. The model's context isn_audio_ctx=5000atsubsampling_factor=8, roughly 400 seconds. Beyond itparakeet-cliexits 0 and writes an empty file. This failure is silent, so the script checks for empty output rather than trusting the exit code. Keep that check. - Chunks overlap by
OVERLAP_SECSand the repeat is not stitched out. It is labelled in the output instead. Do not "fix" this by adding de-duplication:-otxtcarries no timestamps, so a dedupe is a fuzzy suffix/prefix match whose failure mode is dropping a whole sentence — worse than the single boundary word the overlap exists to save. The consumer is a model, which handles a marked repeat trivially. - The chunk loop cuts each piece with its own
-ss/-t. Thesegmentmuxer cannot produce overlapping output, so it is not an option here. - Do not swap in a Whisper model as a "safer" default. It was measured 7x slower than Parakeet on the target hardware at worse accuracy. See README for numbers.
- Captions must stay the first path. Local ASR is minutes; captions are seconds.
Testing
There is no test suite. Verify by hand against two URLs:
- One with captions, which should return in seconds and never invoke
parakeet-cli. - One without, which should chunk and report progress per chunk.
A useful regression check is that the transcript is non-empty and its word count is plausible for the video length, roughly 130-160 words per minute for speech. Note that overlap inflates the count by ~5%, so discount the marked spans before comparing.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 40 lines · 498 tokens per session scan A 1fd4a25fb922
yt-transcript CLAUDE.md is an instructions file published in the GitHub repository ykdojo/yt-transcript (2 stars, last pushed 1mo ago), licensed MIT. It adds 498 tokens to every session, about $0.0025 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
speech-swift AGENTS.md
Instructions for soniqo/speech-swift, covering agent instructions, communication style, workflow, running tests and benchmarks — sequential and memory-aware and git conventions.
open-edit AGENTS.md
Instructions for veedstudio/open-edit, covering agents.md, to run a video, how the pieces relate, hard rules (do not drift — these protect output quality) and conventions.
vzt-flow CLAUDE.md
Instructions for vonzelle-vzt/vzt-flow, covering claude.md — vzt flow, workspace layout, build / test / run, critical gotchas and verification norms.
speech-swift copilot-instructions.md
Instructions for soniqo/speech-swift, a project described as: AI speech toolkit for Apple Silicon — ASR, TTS, speech-to-speech, VAD, and diarization powered by MLX and CoreML.
speech-swift CLAUDE.md
Instructions for soniqo/speech-swift, a project described as: AI speech toolkit for Apple Silicon — ASR, TTS, speech-to-speech, VAD, and diarization powered by MLX and CoreML.
open-edit CLAUDE.md
Instructions for veedstudio/open-edit, a project described as: Open-source, agent-driven editing pipeline: create subtitles, motion graphics, slides, edit and render videos.