/hq-cowork-meetings — Read meeting intelligence from a sandboxed agent

Reads recorded meetings, their source material, and extracted signals from inside Cowork, where the hq CLI is unreachable. Routes through the host-side MCP server. All three tools are read-only.

Args: $ARGUMENTS — what to read (a meeting, a transcript, action items, etc.). Pick the right tool below from intent.

Meetings — mcp__hq__hq_meetings

Action	Purpose	Needs
list	Recorded meetings, newest first	limit optional (default 20)
get	Details for one meeting	meetingId
search	Match by title / participant	query
transcript	Full transcript	meetingId
notes	AI-generated notes	meetingId

{ "action": "notes", "meetingId": "<id>", "company": "<slug>", "json": false }

Sources — mcp__hq__hq_sources

Source material (meeting / email / slack / linear / notion) attached to a vault entity.

Action	Purpose	Needs
channels	Enumerate canonical channels	—
entities	Entities you can access	—
list	Sources for an entity, by channel	entity optional, type optional
get	One source by id	id (+ entity/type)

Signals — mcp__hq__hq_signals

Extracted signals for a vault entity.

Action	Purpose	Needs
types	Enumerate canonical signal types	—
entities	Entities you can access	—
list	Signals for an entity, by type	entity optional, type optional
get	One signal by id	id (+ entity/type)

Signal types: action_item, commitment, decision, key_point, risk, summary.

Notes

• entity defaults to the active company. Pass it explicitly for a multi-company user, and respect cross-company isolation.
• json: true (meetings) / json: true (sources/signals) returns raw JSON for programmatic use; omit for human-readable tables.

When to use this instead of host-side calls

Only inside Cowork or another sandboxed plugin host. On a host-side session, call hq meetings / hq sources / hq signals directly.

Why this skill exists

These commands authenticate with the user's Cognito session under ~/.hq and read the cloud vault — neither is visible from Cowork's Linux VM. The host-side MCP server runs the real commands and returns output to the sandboxed agent.