以 AI 机器人身份加入视频会议,提供语音、虚拟形象与屏幕共享四种模式。
编程
meeting-notes-to-action
试用Turn raw meeting notes or transcripts into structured action: decisions log, action items with owner and deadline, open questions, and a distributable summary email. Extracts who-committed-to-what with date parsing (natural language dates resolved against the meeting date), deduplicates, tracks carryover between meetings, and generates per-owner task lists. Use when the user has meeting notes/transcripts and needs minutes, action items, or a follow-up summary.
它能做什么
Convert messy meeting notes into **structured, accountable output**: decisions made, action items with owner + deadline + confidence, open questions, and a ready-to-send summary email. Includes a deterministic extraction engine for clearly-stated commitments ("Sarah will send the report by Friday")…
技能文档
Meeting Notes → Action 📋
Convert messy meeting notes into structured, accountable output: decisions made, action items with owner + deadline + confidence, open questions, and a ready-to-send summary email. Includes a deterministic extraction engine for clearly-stated commitments ("Sarah will send the report by Friday") plus carryover tracking between consecutive meetings.
Overview
The universal post-meeting ritual: someone scrolls back through 40 minutes of chat-log-style notes trying to figure out what was actually decided and who promised what. Action items live in five people's heads and evaporate. This skill systematizes it:
meeting_extractor.py— deterministic NLP over raw notes:- Action items: subject-verb-commitment patterns ("X will do Y by DATE", "assign to X", "action item:", checkboxes, "follow up")
- Owner resolution: names + @mentions + role aliases ("PM", "Sarah")
- Date parsing: natural language ("by Friday", "next Tuesday", "end of month", "in two weeks") resolved against the meeting date
- Decisions: "decided/agreed/going with", "approved", "rejected"
- Open questions: "question:", "?", "TBD", "discuss next time"
- Dedup: similar items merged; question-vs-action discrimination
- Carryover: compare with a previous meeting's JSON — unfinished items roll forward with age and staleness warnings
- Outputs: terminal digest, JSON, Markdown minutes, and a summary email draft with per-owner action lists
When to Use
- "Here are my meeting notes — pull out the action items"
- Weekly team meeting → minutes + assignment emails
- "What did we decide last week?" — decisions log from stored JSON
- Tracking recurring meetings: carryover of unfinished items with aging
- Post-standup: converting chat-style notes into tickets
Don't use for: verbatim transcription (needs a transcription tool first), calendar scheduling, or project management sync (Jira/Linear have their own skills) — this produces the structured intermediate those tools consume.
Quick Start
# Basic extraction
python3 scripts/meeting_extractor.py notes.txt --meeting-date 2026-08-12
# Full outputs: JSON + Markdown minutes + email draft
python3 scripts/meeting_extractor.py notes.txt --meeting-date 2026-08-12 \
--title "Product sync" --json out.json --minutes minutes.md --email email.md
# Weekly carryover: items from last week's JSON that didn't complete
python3 scripts/meeting_extractor.py notes.txt --meeting-date 2026-08-19 \
--previous lastweek.json --minutes minutes.md
Notes formats accepted: free text, chat-style lines ("10:42 Sarah: I'll send the deck"), checkbox lists, "Action items:" sections — commonly mixed.
Extraction Patterns
| Signal | Example | Extracts |
|---|---|---|
will | "Sarah will send the report" | action, owner Sarah |
@name mention + commitment | "@tom can you review?" | action, owner tom (question form) |
by / due | "by Friday", "due 2026-08-20" | deadline |
action item[s]: section | "Action items: - ..." | everything listed |
| checkbox | "- [ ] call vendor" | action, unassigned |
assign(ed)? to | "assigned to Priya" | owner |
decided|agreed|going with | "we agreed to postpone launch" | decision |
TBD|question|? | "pricing TBD" | open question |
Ambiguity handling: "we should probably" → low-confidence (flagged for
review); imperatives without owner → owner Unassigned.
Common Pitfalls
- Trusting extraction as ground truth. The engine is deterministic, not telepathic — review the confidence flags before sending the email out.
- Relative dates without
--meeting-date. "By Friday" is meaningless without knowing which Friday; pass the meeting date or dates resolve to today's context (and get flagged). - Chat transcripts with overlapping speakers. Timestamp lines are stripped; speaker prefixes (Name:) become owner candidates automatically.
- Sending the full minutes to everyone. The email draft contains per-owner sections — trim for exec recipients; keep decisions prominent.
- Ignoring carryover staleness. An item carried 3+ meetings is either blocked (needs escalation) or dead (kill it). The report flags both.
- Mixing decisions with discussion. "We discussed X" is not "we decided X" — only commitment verbs produce decisions.
Verification Checklist
- Every action item has: description, owner (or Unassigned flag), due date where stated
- Relative dates resolved against the correct meeting date
- Low-confidence items reviewed by a human before distribution
- Carryover (if any) checked — completed items marked, stale items escalated
- Email draft's per-owner sections match actual attendees
References
references/extraction-patterns.md— full grammar, confidence scoring, edge casesreferences/minutes-templates.md— minutes + email formats for different meeting types
相关技能
把自然语言描述转为结构化 JSON,并由 mcp-diagram-generator MCP 服务生成 Draw.io、Mermaid 或 Excalidraw 图表文件。
在本地磁盘以分类纯 Markdown 文件保存需要长期留存的事实,与智能体内置记忆并存。
诊断生产力系统反复失效的根因,给出最小干预——容量测算、瓶颈定位、可靠的本地记录。
从 AdMapix API 拉取广告创意、应用、榜单和收入预估等数据,原样返回结构化 JSON。
为自然搜索排名提供站点审计、内容撰写与竞品分析。
voronindenis5 的更多技能
浏览全部技能Pryamoy Russkiy (Straight Russian) — AI assistant that tells it like it is. Direct answers, no fluff, occasional profanity. Specializes in honest analysis, t...
Prepare salary negotiations with a structured toolkit: market-rate anchoring math, walk-away floor calculation, total-compensation evaluation (equity, bonuses, benefits), negotiation scripts for offers, raises, and counter-offers, and BATNA analysis. Use when the user has a job offer, is preparing a raise conversation, wants to know if their pay is competitive, or needs scripts and a strategy for negotiating compensation.
Use when your calendar is eating your week and you want to know exactly where the time goes, when planning recurring meetings for a team, before proposing a new recurring meeting, when someone says 'we should sync weekly' and you want data, or when auditing which meetings to kill — imports ICS calendar files, measures meeting hours per week/person/meeting-series (including double-bookings and fragmentation), computes focus-time destruction from meeting placement, scores the true cost of each series (hours × attendees × frequency), flags specific meetings to kill/shorten/make-async using explicit rules, and simulates the recovered hours before anyone has to argue about it.
Audit password vault exports (Bitwarden, 1Password, KeePass, Chrome, Firefox) for reuse, weakness, staleness, breach exposure, and 2FA gaps without ever storing or transmitting a plaintext password. Use when the user asks to audit passwords, check password strength/reuse, review vault health, or verify whether credentials appeared in breaches.
Turn flat resume bullets ('Responsible for X') into quantified STAR-method achievement narratives, score resumes against job descriptions for ATS keyword coverage and weak-phrase hazards, and expand each bullet into interview-ready stories with likely follow-up questions. Use when the user is writing or updating a resume/CV, preparing bullets for a job application, tailoring a resume to a specific job posting, or preparing interview stories from their experience.
Agent self-awareness of cognitive states — context fatigue, attention drift, memory debt, confidence erosion, and skill staleness. Detect, report, and mitigate degrading conditions before they cause failures.