Menu
Converts a video URL, local video/audio file, or existing transcript/text into structured, concept-first lecture notes.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Scans Jira for stale issues, duplicates, and missing fields, then fixes what the user approves. Use when the user says the board is a mess, asks for a backlog grooming pass, wants to clean up Jira, or asks about stale/duplicate tickets before sprint planning.
The front door to this plugin. Listens to what the user needs right now — vague or specific — and routes them to the right skill. Trigger whenever the user asks "what can you do," "help me with the team/board/sprint," "what should I focus on," or any open-ended project-management request that doesn't clearly match a single skill.
Update, create, or reprioritize the team's roadmap. Use when adding a new initiative and deciding what moves to make room, shifting priorities after new information comes in, moving timelines due to a dependency slip, or building a Now/Next/Later view from Jira epics.
Closes out a 2-week sprint — reconciles planned vs. completed scope, flags spillover and scope creep, writes a retro-ready summary, and gives a heads-up on the next sprint's readiness. Use when the user says the sprint is ending, asks for a sprint report, wants retro prep, or asks what's carrying over to next sprint.
Plan the next 2-week sprint — scope work from the Jira backlog, estimate team capacity (accounting for PTO), set a sprint goal, and draft a sprint plan. Use when kicking off a new sprint, sizing the backlog against team availability, deciding what's P0 vs. stretch, or handling carryover flagged by `sprint-close`.
Generates a one-page weekly status brief from Jira — sprint burndown, what shipped, what's blocked, and the week ahead. Built for a Scrum team running 2-week sprints. Use when the user asks for a status update, standup summary, sprint check-in, or "what's the team working on."
| name | to-notes |
| description | Converts a video URL, local video/audio file, or existing transcript/text into structured, concept-first lecture notes. |
| disable-model-invocation | true |
| license | Apache-2.0 |
| metadata | {"author":"Oleg Shulyakov","version":"1.1.0","source":"github.com/olegshulyakov/agent.md","catalog":"productivity","category":"notes","tags":["notes","transcript","video","lecture-notes"]} |
Turns a video URL, local video/audio file, or existing transcript into concept-first lecture notes — organized by idea, not by transcript chronology. No speech-to-text: this skill assumes a transcript already exists or can be fetched as captions. Does not handle .pdf/.docx/.pptx/.xlsx/.msg/.epub documents.
If the source is a file attached in this chat rather than a typed/local path, resolve it using this harness's convention for attached-file paths first.
python3 scripts/resolve_source.py <url-or-path>
The script handles all three source shapes:
yt-dlp (never downloads the video). Prefers manual captions over auto-generated; prefers en, falling back to whichever single language track is available. Requires outbound network access — unavailable in some sandboxed cloud sessions..vtt/.srt/.txt transcript next to it, matching either the exact stem (talk.vtt) or a language-coded variant (talk.en.vtt)..vtt/.srt/.txt/.md) — reads it directly, stripping timestamps/cues if present.If the input is pasted transcript text rather than a file or URL, skip the script and use the pasted text as-is.
The script prints LANGUAGE: <code> on the first line, a --- separator, then the plain transcript text.
Done when the script prints resolved transcript text, or exits with a specific error (no manual/auto captions found, no sibling transcript, unsupported file type, no network reaching the URL) — never fall back to transcribing audio yourself. If a URL fetch fails for lack of network access, ask the user to paste the transcript text or upload a caption file instead.
Read the resolved transcript and draft lecture notes organized by the source's actual ideas — group related points under topic headers, do not follow the transcript's timeline.
Save the notes as <kebab-case-title>.md:
Use this frontmatter:
---
title: <title>
author: <speaker/author, if identifiable, else omit>
source: <original URL or file path>
---
If the resolved language (from Step 1's LANGUAGE: line) isn't en, add a language: <code> field. Redact sensitive information (API keys, passwords, personally identifiable information) before writing the notes.
Done when the file is written with complete frontmatter and headers reflect the source's real concepts rather than its chronology.