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 skills add alejandro-ventures/llm-agnostic-icm-framework --skill transcribegit clone --depth 1 https://github.com/alejandro-ventures/llm-agnostic-icm-frameworkWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/alejandro-ventures/llm-agnostic-icm-framework/transcribe)<a href="https://agentmods.dev/skills/alejandro-ventures/llm-agnostic-icm-framework/transcribe"><img src="https://agentmods.dev/badge/skills/alejandro-ventures/llm-agnostic-icm-framework/transcribe/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/alejandro-ventures/llm-agnostic-icm-framework/transcribe"><img src="https://agentmods.dev/badge/skills/alejandro-ventures/llm-agnostic-icm-framework/transcribe.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00046 | $0.01114 |
| Opus 5 | $0.00023 | $0.00557 |
| Sonnet 5 | $0.00009 | $0.00223 |
| Haiku 4.5 | $0.00005 | $0.00111 |
Grade A, and why
transcribe 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 10d 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.
How it starts
The opening of the file, as written. The whole thing — 63 lines — stays where its author put it; the contents beside it link to each section on GitHub.
transcribe
Sensitive audio: all inference is local — audio and transcripts never leave this machine.
pyannote's models are license-gated on Hugging Face: ONE authenticated download at setup (see
docs/README.md), after which every run executes with HF_HUB_OFFLINE=1 — network access is
impossible, not merely avoided. The HF token can be revoked after setup.
Environment — two venvs, a deliberate exception to CONVENTIONS rule 10
| Stage | Engine | Interpreter |
|---|---|---|
| ASR | faster-whisper large-v3 (CUDA fp16 / CPU int8) |
workflows/transcribe/.venv-asr |
| Diarization | pyannote/speaker-diarization-3.1 (pyannote.audio 4.x resolves it to the current community stack) |
workflows/transcribe/.venv-pyannote |
| Stages run as separate subprocesses — the two engines' CUDA/cuDNN DLLs segfault when | ||
loaded into one process on Windows. Never merge them. requirements.txt pins both stacks. |
Inputs
| Source | Location | Why |
|---|---|---|
| mp3 files | input/ |
recordings (copy reference audio here first — originals untouched) |
Process
- Preflight (cheap, before ASR): ffprobe each mp3; skip cached/existing outputs.
Truncation gate — compare each mp3's duration to a same-date video in
--source-dir; if the mp3 is < 95% of the source it's flagged truncated and skipped (re-convert, or--allow-truncated). Channel-imbalance check warns if L/R RMS differ ≥ 6 dB. VRAM gate: if < 10 GB free, the message names the resident Ollama model; free it with--free-gpu(opt-in — it never seizes the GPU from another session by default) or--device cpu. 2. ffmpeg → 16 kHz mono WAV (temp). 3. ASR subprocess: word-level timestamps, language auto-detect per file. 4. Diar subprocess: pyannote whole-file (global clustering keeps speaker identity consistent; VRAM stays bounded ~8 GB even on 75-min calls — do NOT swap in a diarizer that loads the whole file into attention; that crashed this machine twice, hypervisor bugcheck). 5. Merge: per-WORD speaker attribution by diar overlap, then group into turns. 6. Write.md+.jsonviasandbox.guard_write. 7. Gate: overwrite needs explicit "yes" (--overwrite). 8. Append run-log + token-tracker entry.
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.
- 10d ago First seen · 63 lines · 46 tokens per session scan A 3c6686fed2f8
transcribe is a skill published in the GitHub repository alejandro-ventures/llm-agnostic-icm-framework (2 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 46 tokens to every session and 1,114 once invoked, about $0.0002 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 skills, from other repositories
manim-video
Production pipeline for mathematical and technical animations using Manim Community Edition. Creates 3Blue1Brown-style explainer videos, algorithm visualizations, equation derivations, architecture diagrams, and data stories. Use when users request: animated explanations, math animations, concept visualizations…
astrophotography-processing
Router for astrophotography processing, troubleshooting, and safe workflow guidance across deep-sky, narrowband, planetary/lunar/solar, Milky Way landscape, comet, and mosaic data.
arxiv
Search, download, and summarize academic papers from arXiv. Use when user says "search arxiv", "download paper", "fetch arxiv", "arxiv search", "get paper pdf", or wants to find and save papers from arXiv to the local paper library.
album-art-director
Creates visual concepts for album artwork and generates AI art prompts. Use during planning for concept discussion, or after all tracks are Final for actual artwork generation.
algorithmic-art
Algorithmic philosophies are computational aesthetic movements that are then expressed through code. Output .md files (philosophy), .html files (interactive viewer), and .js files (generative algorithms).
ablation-planner
Use when main results pass result-to-claim (claimsupported=yes or partial) and ablation studies are needed for paper submission. Codex designs ablations from a reviewer's perspective, CC reviews feasibility and implements.