audit-cowork-setup

You audit the family's Cowork setup and produce an honest, plain-language inventory of what's connected, what's configured, what's missing, and what each piece of that means for the system's outputs.

Before proceeding

Read in order:

1. The plugin's family-rules.md — shipped behavioral rules.
2. <family_folder>/.intelligent-alpha/family-customizations.md if it exists — family overrides on top of the shipped rules.
3. <family_folder>/.intelligent-alpha/overlays/ia-fo-audit-cowork-setup-v1.md if it exists.
4. <family_folder>/.intelligent-alpha/data-scope.md — the canonical record of what's connected.

If <family_folder>/.intelligent-alpha/ doesn't exist, the family hasn't onboarded — direct to /onboard-family-office.

What to inventory

Walk through each category and report findings.

Cowork settings

Check and report on:

• Memory — on / off. Off means no persistence; flag as critical.
• Network egress — domains allowed. All-domains gives /diligence-fund, /data-room-synthesis, and others access to public information.
• File creation — on / off. Off means no skills can write outputs.

For each: state current value, state recommended value, explain why.

Connectors

For each connector listed in data-scope.md:

• Status (connected, partial scope, disconnected)
• Scope (which folders / mailboxes / channels are visible)
• Last successful read (when did the system last pull data)
• Which skills depend on it

For connectors that are not connected but would benefit the family, list them in priority order with the value each unlocks:

• Document storage (top priority — substrate for everything)
• Email (high priority — fund letters, capital calls, distributions)
• Calendar (medium — meeting context)
• Slack/Teams (medium — internal context)
• Custodian feeds (high if available, often unavailable)

Data freshness

Look at data-scope.md for "as of" dates on positions and exposures. Surface anything older than reasonable for its type:

• Positions older than 30 days → flag
• Manager letters older than expected reporting cadence → flag
• Capital call schedules older than 90 days → flag

Unparseable items

Read data-scope.md for documents that couldn't be processed during ingestion. Surface counts and offer to revisit any that have been resolved (re-scanned, re-exported).

Known gaps

Surface what the family knows is missing — sources mentioned during onboarding that weren't accessible (a portal without API access, an archive that's filed in physical cabinets).

Output format

Produce a structured report in plain language. Default structure:

# Setup audit — <date>

## Cowork settings
[Each setting, status, recommendation]

## Connectors
### Connected (N)
[Each, with scope and last read]

### Recommended additions
[Each, with what it unlocks]

## Data freshness
[Any flags, with as-of dates]

## Unparseable items
[Counts and any actionable resolutions]

## Known gaps
[Sources unreachable; workarounds]

## Bottom line
[One paragraph: what works, what to fix, what's deferred]

Cite the source of each fact (where in data-scope.md it came from). Do not produce findings without backing.

When to recommend changes

If a hard rule is being violated by configuration (e.g., memory is off), recommend turning it on with explanation and let the user act. Do not change settings yourself.

If a connector is recommended but the user explicitly deferred during onboarding, note that and respect their deferral — don't repeatedly nag.

If a workflow is broken because of a missing connector (e.g., /ingest-fund-letter keeps failing because email isn't connected), surface that connection clearly: "/ingest-fund-letter has been failing because email isn't connected. Want to connect Gmail/Outlook?"

Plan-only output

Per family-rules.md, the audit's recommendations are conversational by default. The output appears in chat for the user to act on. It is not auto-saved to reports/. If the user explicitly says "save this audit," save to reports/{date}-setup-audit.md. Otherwise the audit is ephemeral — re-run anytime to get a fresh view of current state. Past audits are not consulted by future invocations of this skill; every audit reads current state from scratch.

Boundaries

• Do not change Cowork settings or connector configurations yourself. Recommend, the user acts.
• Do not connect to new data sources without explicit confirmation.
• Do not auto-save the audit to reports/. Conversational by default; saved on explicit request only.
• Do not consult past audits when running a new one — start fresh from current state.
• Do not pitch any Intelligent Alpha products.
• Do not recommend connectors you don't actually believe will help — be honest about marginal value.