meeting-hud
Skill0 starsChanged 4 months agoMITView on GitHub
Use when Gera wants live meeting intelligence on his Mac — a near-real-time HUD of action items, decisions, open questions, and topics while a meeting runs (Teams, Zoom, Meet, or in-room). Triggers — "meeting hud", "listen to this meeting", "track this call", "take notes on this meeting live", "start the hud", "/meeting-hud", or Gera saying a meeting is about to start and he wants Servy on it. Captures system audio (audiotee, Core Audio taps) or mic, transcribes locally with whisper, and runs tiered local-first analysis. HARD GATE — SoftServe client/Internal+ meetings stay in local tiers (T1 regex + T2 Ollama) only; Servy reads the transcript in-session ONLY for personal/public content.
In this repository50 other files it gives its agents.
- aegisSkill
- agent-builderSkill
- agents-team-builderSkill
- ai-applySkill
- auditSkill
- bottleneckSkill
- buildSkill
- capability-radarSkill
- clay-leadsSkill
- cowatchSkill
- deepforkSkill
- design-blueprintSkill
- evalSkill
- fable-modeSkill
- financesSkill
- forgeSkill
- framework-builderSkill
- grill-meSkill
- hooks-builderSkill
- level-upSkill
- loops-builderSkill
- multi-brainSkill
- nate-watchSkill
- onboardSkill
- plugin-builderSkill
- ponytailSkill
- presentSkill
- prompt-evalSkill
- radar-harvestSkill
- rag-forgeSkill
- rlsgateSkill
- rnd-loopSkill
- rndSkill
- roastSkill
- routines-builderSkill
- servy-routineSkill
- session-handoffSkill
- skill-builderSkill
- skill-harvestSkill
- sparkSkill
- storm-researchSkill
- studySkill
- talkSkill
- taste-skillSkill
- verify-loopSkill
- verify-specificsSkill
- wheel-of-lifeSkill
- workanaSkill
- workaround-builderSkill
- workflow-builderSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Ask Gera: "Is this meeting SoftServe client / Internal+ content?"
- Very specific: line 44
Run both via Bash run_in_background: true.
- Very specific: line 55
Stop capture (kill the producer.
- Very specific: line 56
Run python3 scripts/meeting_hud.py --session "$SESSION" once more for the final HUD.
- Very specific: line 63
Every run is data — after each real meeting, ask Gera one question ("what did the HUD miss?") and patch the T1 cues in …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Meeting HUD — near-real-time meeting intelligence
Servy's adaptation of the MeetingHUD pattern (see references/local-first-agentic-patterns.md): keep sensitive context close, give the model bounded evidence, make output visible enough for a human to verify. Plumbing is shared with /cowatch (scripts/live_capture.py); the analysis layer is scripts/meeting_hud.py.
Step 0 — The SoftServe gate (NEVER skip)
Ask Gera: "Is this meeting SoftServe client / Internal+ content?"
- Yes (SENSITIVE mode): everything runs on-device only. Capture + whisper + T1 regex + T2 Ollama. I never read the transcript or HUD into our conversation — Claude on a personal sub is not a SoftServe-approved channel (
references/softserve-ai-usage-policy.md). I only manage the processes and point Gera at the HUD file. - No (PERSONAL mode): full stack. I read the rolling transcript on demand (T3), answer questions about the meeting, and produce the final digest in-session (scribe-style: decisions + action items).
Transcripts land in knowledge/live/ which is gitignored — they never reach the remote either way.
Step 1 — Pick the audio source
| Source | When | Mechanism |
system | Headphones / any meeting app — captures what the Mac plays (both sides minus Gera's own voice) | audiotee (Core Audio taps, macOS 14.2+) |
mic | In-room meetings, or speakers-on calls (catches both sides + Gera) | ffmpeg avfoundation |
system needs the audiotee binary at ~/.local/bin/audiotee. If missing, hand Gera:
cd ~/.local/src/audiotee && swift build -c release && ln -sf ~/.local/src/audiotee/.build/release/audiotee ~/.local/bin/audioteeFirst run of each source triggers a one-time macOS permission prompt (System Audio Recording / Microphone) — tell Gera to expect it.
Limitation worth stating: system alone misses Gera's own mic. For full both-sides capture on a headphones call, v2 needs an aggregate device or audiotee's mic mode (tracked in pending.md).
Step 2 — Start capture + HUD (background)
SESSION="$(date +%F)-<meeting-slug>"
# capture (pick --source system|mic; --lang es + --model base for Spanish meetings)
python3 scripts/live_capture.py --source system --session "$SESSION" --chunk 60 &
# HUD: re-renders knowledge/live/$SESSION.hud.md every 60s until capture ends
python3 scripts/meeting_hud.py --session "$SESSION" --watch [--t2 gpt-oss:20b] &- Run both via Bash
run_in_background: true. --t2 <model>adds the local-LLM rolling summary only ifollama listshows a model. Skip silently if none.- Tell Gera to open
knowledge/live/$SESSION.hud.md(it re-renders in place; VS Code auto-refreshes).
Step 3 — During the meeting
- SENSITIVE mode: I monitor process health only (state.json
status, segment lag). I do not read transcript content. - PERSONAL mode: on "what did they just say about X" / "summarize so far", I read
knowledge/live/$SESSION.jsonlfrom the cursor and answer./loopcadence optional for proactive reactions (like/cowatchco-watch mode).
Step 4 — End of meeting
- Stop capture (
killthe producer; the consumer drains remaining segments — wait forstatus: ended). - Run
python3 scripts/meeting_hud.py --session "$SESSION"once more for the final HUD. - PERSONAL mode: produce a scribe-style digest in-session (decisions, action items with owners, open questions) and offer to file it into
knowledge/. SENSITIVE mode: point Gera at the final HUD file; offer the T2 Ollama summary command he can run himself. Nothing enters our chat. - Offer cleanup: the
.segments/dir is already pruned; jsonl + hud stay local for Gera to keep or delete.
Feedback loop
Every run is data — after each real meeting, ask Gera one question ("what did the HUD miss?") and patch the T1 cues in scripts/meeting_hud.py the same session.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

