Stallari/plugins

Fork 0

Add Meeting Intelligence pack #1

Merged

piersdd merged 2 commits from contrib/meeting-intelligence into main

2026-04-18 11:25:31 +00:00

piersdd commented

2026-04-18 01:25:40 +00:00

(Migrated from github.com)

Summary

New community pack: Meeting Intelligence v1.0.0
Privacy-first meeting processing with on-device transcription via mlx-whisper
4 skills: transcribe-meeting, extract-minutes, weekly-commitments, process-meeting
1 agent: meeting-processor (operator role)
Audio never leaves the machine — zero data egress

Pack details

Field	Value
author	digitalassistants-au
author_type	community
readiness	experimental
licence	MIT
pack spec	1.6

Services

Requires: vault (read, search, create, append, files, lens_fts)
Recommends: calendar (events, today), notifications (send)
Transcription: mlx-whisper MCP (on-device, installed separately via uv)

Privacy model

All transcription is on-device via mlx-whisper (MLX backend, Apple Silicon)
No audio sent to cloud APIs
phones_home: false, stores: nothing

Future plans

Speaker diarisation (pyannote.audio)
Board minutes mode with resolution numbering
Compliance critique-refine loop

## Summary - New community pack: **Meeting Intelligence** v1.0.0 - Privacy-first meeting processing with on-device transcription via mlx-whisper - 4 skills: `transcribe-meeting`, `extract-minutes`, `weekly-commitments`, `process-meeting` - 1 agent: `meeting-processor` (operator role) - Audio never leaves the machine — zero data egress ## Pack details | Field | Value | |-------|-------| | author | digitalassistants-au | | author_type | community | | readiness | experimental | | licence | MIT | | pack spec | 1.6 | ## Services - **Requires:** vault (read, search, create, append, files, lens_fts) - **Recommends:** calendar (events, today), notifications (send) - **Transcription:** mlx-whisper MCP (on-device, installed separately via `uv`) ## Privacy model - All transcription is on-device via mlx-whisper (MLX backend, Apple Silicon) - No audio sent to cloud APIs - `phones_home: false`, `stores: nothing` ## Future plans - Speaker diarisation (pyannote.audio) - Board minutes mode with resolution numbering - Compliance critique-refine loop

piersdd commented

2026-04-18 07:06:40 +00:00

(Migrated from github.com)

Closing — pack agent trust model needs platform-level gating before community packs can define agents. Will revisit after hardening.

piersdd commented

2026-04-18 08:20:37 +00:00

(Migrated from github.com)

Hi — thanks for your Meeting Intelligence pack submission (PR #1). It's a solid piece of work and exactly the kind of contribution we want to see in the ecosystem.

During review we identified a gap in our pack spec around how community-defined agents are trusted and sandboxed at dispatch time. This isn't a problem with your pack — it's a platform-level issue we need to resolve before any community pack that defines agents can be safely merged.

We've temporarily made the repo private while we work through the design. Your PR and branch are intact and we'll reopen it once the spec update lands.

We'll be in touch with specific guidance on any changes you'll need to make to meeting-intelligence.yaml to conform to the updated spec. We expect the adjustments to be minor — likely a namespace or metadata change rather than a rewrite.

Apologies for the delay, and thanks for your patience.

Hi — thanks for your Meeting Intelligence pack submission (PR #1). It's a solid piece of work and exactly the kind of contribution we want to see in the ecosystem. During review we identified a gap in our pack spec around how community-defined agents are trusted and sandboxed at dispatch time. This isn't a problem with your pack — it's a platform-level issue we need to resolve before any community pack that defines agents can be safely merged. We've temporarily made the repo private while we work through the design. Your PR and branch are intact and we'll reopen it once the spec update lands. We'll be in touch with specific guidance on any changes you'll need to make to `meeting-intelligence.yaml` to conform to the updated spec. We expect the adjustments to be minor — likely a namespace or metadata change rather than a rewrite. Apologies for the delay, and thanks for your patience.

piersdd commented

2026-04-18 11:12:06 +00:00

(Migrated from github.com)

Thanks for your patience on this. We've shipped the trust model that was blocking your PR — it landed in v0.66.0.0 across two design decisions (DD-145, DD-146).

Here's what the platform now does with your pack:

Namespacing: Your agent becomes meeting-intelligence/meeting-processor at compile time — no collisions with platform or other community agents
Guardrails: Platform security rules are prepended to your agent's system prompt at dispatch. Your prompt runs in full — it's wrapped, not suppressed
Consent: Users see a one-time approval prompt before your agent first dispatches, showing the agent name, role, and services it accesses
Integrity: The guardrails are sealed (AES-256-GCM) and SHA-256 verified on every load, with beacon alerts if anything is tampered with

You don't need to change anything in your pack YAML. The agents: section and agent: field on your skills are valid pack-spec v1.6, and the platform handles trust boundaries transparently.

If you'd like to go further, pack-spec v1.7 supports optional pack-local guardrails — self-imposed constraints on your own agent (e.g., "only write to spaces/Personal/meetings/"). Entirely optional.

Reopening this PR now.

Thanks for your patience on this. We've shipped the trust model that was blocking your PR — it landed in v0.66.0.0 across two design decisions (DD-145, DD-146). Here's what the platform now does with your pack: - **Namespacing:** Your agent becomes `meeting-intelligence/meeting-processor` at compile time — no collisions with platform or other community agents - **Guardrails:** Platform security rules are prepended to your agent's system prompt at dispatch. Your prompt runs in full — it's wrapped, not suppressed - **Consent:** Users see a one-time approval prompt before your agent first dispatches, showing the agent name, role, and services it accesses - **Integrity:** The guardrails are sealed (AES-256-GCM) and SHA-256 verified on every load, with beacon alerts if anything is tampered with **You don't need to change anything in your pack YAML.** The `agents:` section and `agent:` field on your skills are valid pack-spec v1.6, and the platform handles trust boundaries transparently. If you'd like to go further, pack-spec v1.7 supports optional pack-local guardrails — self-imposed constraints on your own agent (e.g., "only write to `spaces/Personal/meetings/`"). Entirely optional. Reopening this PR now.

piersdd commented

2026-04-18 11:25:26 +00:00

(Migrated from github.com)

Welcome to the marketplace — this is our first community pack merge. 🎉

Validated against pack-spec v1.7 with the agent trust model (DD-145/DD-146) now live. Your meeting-processor agent will run as meeting-intelligence/meeting-processor with platform guardrails and user consent gating.

Merging as-is — no changes required.

Welcome to the marketplace — this is our first community pack merge. 🎉 Validated against pack-spec v1.7 with the agent trust model (DD-145/DD-146) now live. Your `meeting-processor` agent will run as `meeting-intelligence/meeting-processor` with platform guardrails and user consent gating. Merging as-is — no changes required.

No reviewers

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

Stallari/plugins!1

No description provided.

Rows
Columns