Breaking: Java4Shell (CVE-2026-114938) - Less Than 0.12% of JVMs Estimated at Risk of RCE
Read the Report
Learn How MCP Is Bypassing a Decade of Cloud Security Best Practices
Read the Report
Are you exposed? Connect a repo in minutes.
Scan your repos free

transcript

Latest version: vgit:85b19e8c7bbd85e959ff59e116ff28fdb77fc899

Overview

Last scanned
18 Nov 2026
Agent skill Publisher: pascalandy
0 Stars
1 Forks
1 Versions
Python Repo language

Scan summary

transcript was last scanned on 18 Nov 2026 at commit 85b19e8c. Risk score 1 of 100 (low). No malicious behavior was detected.

  • Releases: Single version.
  • Findings: 1 rule fired in commit 85b19e8c (3 instances: 3 low). The score describes this commit only.

Versions

1 version published
Versions and the rules that fired in each
VersionPublishedStatus
85b19e8c 1 rule fired

Rules fired in 85b19e8c: 1

Findings

3 Low

Findings for commit 85b19e8c. The score is a capped sum of the weights of the rules that fired — not CVSS, not exploitability, and not a probability. It is not comparable across packages with different scan coverage.

Skill mentions tunnel, paste or webhook services

The skill's text or scripts mention a tunneling service such as ngrok or trycloudflare, a Discord webhook, a raw paste link or a public IP address. Legitimate tools reference these too, so this is context for review rather than evidence of harm.

LOW
Checking access…

🔒 Create a free account to see all 3 findings with descriptions and instance counts

The score describes 85b19e8c; other versions have their own rules, not their own score.

README

The publisher’s own README, as collected from the registry.

Transcript

Transcribe YouTube, Zoom, or local audio with Deepgram, then optionally create a Markdown summary through a named inference profile, run by a tool-free, session-free claude or pi process.

For transcription commands, resolve scripts/transcript.py relative to this skill directory and run that absolute path with uv run. Allow a 600-second process timeout for a real run, multiplied by the input URL count for a queue.

For a request such as « il faut rafraîchir le cookie pour Transcript », follow Refresh YouTube cookies. This maintenance request uses just transcript-auth-refresh --dotfiles-worktree <worktree> on the source Mac, a dotfiles PR and just merge, and makes no paid transcription call.

Choose the command

# YouTube
uv run <skill_dir>/scripts/transcript.py run youtube --url "<youtube-url>"

# YouTube queue
uv run <skill_dir>/scripts/transcript.py run youtube --url "<youtube-url-a>" "<youtube-url-b>"

# Latest Zoom meeting
uv run <skill_dir>/scripts/transcript.py run zoom --latest

# Specific Zoom meeting folder
uv run <skill_dir>/scripts/transcript.py run zoom --path "<folder-name-or-full-path>"

# Local audio, keeping the source file intact
uv run <skill_dir>/scripts/transcript.py run file --path "<audio-file>" --no-summary

Pass --profile only when the user selects a profile; list profiles marks the default.

YouTube runs read the session from the fnox vault entry YOUTUBE_COOKIES_TXT and verify it before paid work. No machine keeps a plaintext copy. Add --cookies FILE to override the vault. Neither mode falls back anonymously. Read Use the vault session for the source order, renewal, and the free transport check.

Manage inference through profiles

Treat every request to use, change, or add a model as a profile-management request. Do not translate the request directly into --provider, --model, or --effort.

First run:

uv run <skill_dir>/scripts/transcript.py list profiles

Show the current profile names, providers, models, and efforts. Then ask whether the user wants to select an existing profile, update one, or create one. If the user already named a profile, show its configuration and use --profile <name>.

The list order records preference only. The CLI runs one profile and never falls back to another profile.

To update or create a profile, edit the INFERENCE_PROFILES registry in scripts/transcript.py, the only list of profiles, then run the validation procedure in README.md.

For Zoom, --path names the meeting folder, not its audio*.m4a file. A plain folder name resolves under ~/Documents/Zoom.

help run youtube, help run zoom, and help run file list every run option and its default. Add an option only when the request calls for it. --provider, --model, and --effort form an advanced custom target for diagnostics only. --open and --preview are opt-in, so a normal run has no GUI side effect.

Change a prompt

Each bundled prompt is a Markdown file in references/prompts/, and its file name without .md is the --prompt value. Edit a file there to change a prompt, or add a .md file to create one; list prompts shows it. In a managed installation, edit the source package, because the next install overwrites installed copies.

Only summary_with_quotes receives the timestamped transcript; every other prompt receives plain text. Zoom summaries use the synthese-rencontre prompt from andy-mode's distill-prompt route.

Agent operation

Every command answers in one JSON line. Success is {"ok":true,…} on stdout: a run lists the files it saved, list its values, and a dry run its plan. A failure leaves stdout empty and ends stderr with {"ok":false,"errors":[…]}. While a run works, stderr shows each result folder as soon as it appears.

For an unfamiliar machine or after a preflight failure, run:

uv run <skill_dir>/scripts/transcript.py doctor --source youtube
uv run <skill_dir>/scripts/transcript.py doctor --source zoom
uv run <skill_dir>/scripts/transcript.py doctor --source file --no-summary

doctor checks local dependencies and credentials without calling Deepgram, a summary model, or another paid API.

Before a run with a non-default profile, prompt, output, or source setting, resolve the plan without secrets, network calls, or writes:

uv run <skill_dir>/scripts/transcript.py run youtube --url "<youtube-url>" --prompt short_summary --dry-run

Discover valid values through the command tree:

uv run <skill_dir>/scripts/transcript.py list prompts
uv run <skill_dir>/scripts/transcript.py list profiles
uv run <skill_dir>/scripts/transcript.py list models --provider claude
uv run <skill_dir>/scripts/transcript.py list models --provider codex
uv run <skill_dir>/scripts/transcript.py list models --provider openrouter
uv run <skill_dir>/scripts/transcript.py --help

Exit 2 means the invocation or source is invalid. Exit 1 means runtime work failed. Exit 75 means a temporary failure before any paid request; rerunning the same command is safe. Exits 130 and 143 mean the run was interrupted. Each error ends with the command that fixes it, after fix:, retry:, or rerun:; a usage error without one adds a help command instead. Never rerun an exit 1 run automatically, because Deepgram may already have billed the audio.

Report the result folder, the one holding the files, and whether a summary .md is among them. A failure after the folder appears keeps it, and its answer still lists the files saved, including the metadata that names the failed stage. A run with several URLs lists the files of each URL, and its errors name each failed URL before a command that reruns only those. Do not paste the generated summary into chat unless the user asks.

YouTube transport check

Run uv run <skill_dir>/scripts/youtube_smoke.py for the free transport check on https://www.youtube.com/watch?v=EIEc43CxIvY from the README's Test videos table. It uses the same source as a run: the explicit cookies file, else the vault session, else anonymous access. All modes require pinned yt-dlp 2026.7.4. The check validates temporary audio with ffprobe and never calls Deepgram or a summary model. A pass proves the selected transport downloaded audio, not that YouTube authenticated the account.

A run never reads a browser. The vault session comes from Helium's CLI profile, a profile only tools use, signed in to the account Transcript should use, such as YouTube Premium. When YouTube refuses the vault session, follow Refresh YouTube cookies.

Changing this skill

Read README.md in this skill directory before changing code, tests, prompts, or documentation. It owns the CLI contract, runtime requirements, output formats, and validation procedure.

Always finish every transcript change with the README's E2E closeout. Its Test videos rule pre-authorizes the paid runs with --profile sonnet, so run the closeout without asking. Keep Checks, Automated tests, CI, Agent QA, and E2E separate. E2E is the last gate and passes only when the relevant coded validation and Agent QA pass.

OX Radar and Response Center

Find threats in your stack.

OX Radar can check high-priority threats, like critical package CVEs, against your context.

Each incident is marked Detected or Not Detected with a remediation plan and proof all tracked in the Response Center.

Where it’s listed

The probe returned only these fields for this server — auth requirement, protocol and tool list weren’t observed, so they’re not shown (never guessed).

  • GitHub pascalandy/skills/transcript

References

The publisher’s own links, as the registry reports them. These are not sources for the assessment on this page.

Scan your whole stack

Group 1261154050 1
Group 1261154050
 

From OX Research Newsroom

Nodejs Brotli Flaw
CVE

Node.js Brotli Decoder Out-of-Bounds Write (RCE) – CVE-2026-111497

Ethersphere Hijacked With Crypto Stealing Malware (1)
npm

Ethersphere Hijacked With Crypto Stealing Malware Via Github & npm

Microsoft's Official agent framework Hijacked
AI & VibeSec

BREAKING: Hijacked Releases of Microsoft’s Official agent-framework Steals Azure and CI Credentials

Frame 2085668439 (1)
Group 1261154229