# Sanity Check

> Verify all InboxMate demo agents are properly configured — checks greetings, quick questions, knowledge, colors, and CRM linkage. Run after batch demo creation or before sending outreach.

- Skill: `psquared-development/sanity-check` (Agent Skill)
- Install (CLI): `npx skillmds@latest add psquared-development/sanity-check`
- Raw SKILL.md: https://api.skillmd.com/api/skills/psquared-development/sanity-check/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: psquared-development (https://skillmd.com/u/psquared-development)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/psquared-development/sanity-check

---


# Sanity Check — InboxMate Demo Agents

> **Announce:**
> ```
> ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
> Sanity Check started.
> Checking environment...
> ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
> ```

---

## STEP 0 — Check Environment

Read the `.env` file in the current working directory:

Required tokens:
- `EMAIL_DRAFT_ONLY_BEARER` — for calling the notification service sanity check endpoint
- `PSQUARED_CRM_TOKEN` — for querying CRM opportunities

If either is missing, **stop** and ask the user.

---

## STEP 1 — Select Targets

If the skill was invoked with a parameter (e.g., `/sanity-check PENDING_REVIEW` or `/sanity-check campaign:<id>`), skip the prompt below and use the parameter directly — map it to the appropriate option (A or E).

Ask the user what to check. Options:

**A) By CRM stage/status** — query CRM for opportunities matching a filter:
- `demoStatus = PENDING_REVIEW` (pre-send QA)
- `demoStatus = OK_TO_SEND` (about to send)
- `demoStatus = SENT` (already sent, verify nothing broke)
- A specific campaign ID

**B) All visible drafts** — check everything currently in the notification service draft queue

**C) Specific opportunity IDs** — user provides a list

**D) Specific draft IDs** — user provides draft UUIDs from the notification service

**E) Campaign ID** — user provides a CRM campaign UUID

### Querying CRM (for option A only)

For option A, query the CRM to collect opportunity IDs:

```
POST https://crm.psquared.dev/graphql
Authorization: Bearer $PSQUARED_CRM_TOKEN
Content-Type: application/json

{
  "query": "{ opportunities(first: 200, filter: { stage: { eq: SCREENING }, demoStatus: { eq: <STATUS> } }) { edges { node { id company { name } } } totalCount } }"
}
```

Collect all `id` values from the response.

---

## STEP 2 — Run Sanity Check via Notification Service

The endpoint accepts **three input formats** — use whichever matches the user's input. The backend resolves everything to demo page IDs internally.

### Option 1: By CRM opportunity IDs

```
POST https://notifications.psquared.dev/drafts/sanity-check
Authorization: Bearer $EMAIL_DRAFT_ONLY_BEARER
Content-Type: application/json

{
  "crm_opportunity_ids": ["uuid1", "uuid2", ...]
}
```

### Option 2: By campaign ID

```
POST https://notifications.psquared.dev/drafts/sanity-check
Authorization: Bearer $EMAIL_DRAFT_ONLY_BEARER
Content-Type: application/json

{
  "campaign_id": "campaign-uuid"
}
```

The backend resolves the campaign → all linked opportunity IDs via CRM, then checks each.

### Option 3: By draft IDs

```
POST https://notifications.psquared.dev/drafts/sanity-check
Authorization: Bearer $EMAIL_DRAFT_ONLY_BEARER
Content-Type: application/json

{
  "draft_ids": ["draft-uuid1", "draft-uuid2", ...]
}
```

The backend resolves draft IDs → `crm_opportunity_id` from the drafts table, then checks each.

**Max 200 IDs per request.** If more than 200, batch into multiple calls. Only provide ONE of the three fields per request.

The endpoint:
1. Resolves input → CRM opportunity IDs (if not already)
2. Resolves each CRM opportunity → demo page ID (via CRM `demoId` field)
3. Calls the agenthub Supabase RPC to validate each demo agent
3. Returns results mapped back to CRM opportunity IDs

### Response format

```json
{
  "success": true,
  "data": [
    {
      "crm_opportunity_id": "uuid",
      "demo_found": true,
      "agent_found": true,
      "agent_published": true,
      "claimed": false,
      "bucket_count": 1,
      "total_bucket_items": 7,
      "bucket_has_items": true,
      "all_items_url_type": true,
      "non_url_items": 0,
      "all_items_embedded": true,
      "failed_items": 0,
      "processing_items": 0,
      "not_found_items": 0,
      "low_token_items": 0,
      "no_source_url_items": 0,
      "avg_tokens": 847,
      "demo_api_reachable": true,
      "draft_footer_ok": true,
      "draft_button_ok": true,
      "issues": [],
      "warnings": [],
      "healthy": true
    }
  ]
}
```

### What gets checked

**Health failures (unhealthy):**
- Demo page / agent not found
- Agent not published
- No knowledge buckets / empty buckets
- Text-type items (should be URL-sourced via Tavily)
- Failed or still-processing embeddings
- 404 / "Seite nicht gefunden" content in knowledge items
- >50% of items have < 50 tokens (AI stubs, not real scraped content)
- Demo API unreachable
- Email draft link mismatches

**Warnings (healthy but flagged):**
- Demo already claimed
- Some items missing source URL (no citations in chat)
- Some items have < 50 tokens (but less than half)
- Average tokens per item < 100 (thin knowledge)
```

---

## STEP 3 — Report Results

Group results into three categories:

### Healthy
All checks pass. Show count and list company names.

### Warnings (healthy but with notes)
- `claimed = true` — demo was already claimed by a prospect (still healthy, but worth noting)

### Unhealthy
Show each unhealthy item with:
- Company name (from CRM query in Step 1)
- List of issues from the `issues` array
- List of warnings from the `warnings` array (if any)
- Suggested action:
  - `"No demo linked in CRM"` → demo was never created, run `/inboxmate-demo`
  - `"Demo page not found"` → demo page deleted or demoId mismatch in CRM
  - `"Agent not found"` → agent was deleted, needs rebuild
  - `"Agent not published"` → run `publish_agent` via MCP
  - `"No knowledge buckets assigned"` → run `/fix-demos` or rebuild
  - `"Knowledge bucket is empty"` → knowledge items failed to scrape, re-scrape
  - `"N knowledge items are text type"` → run `/refurbish-demos` to replace with URL-sourced content
  - `"N knowledge items have failed embeddings"` → re-index needed
  - `"N knowledge items still processing"` → wait and re-check
  - `"N knowledge items contain 404/not-found page content"` → run `/refurbish-demos` — URLs were guessed wrong
  - `"N/M knowledge items have < 50 tokens (likely stubs)"` → run `/refurbish-demos` — content is AI summaries, not real scraped pages

### Summary line

```
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Sanity Check Complete
✓ Healthy: N    ⚠ Warnings: N    ✗ Unhealthy: N
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
```

---

## STEP 4 — Offer Next Steps

Based on results, suggest:

- If unhealthy items exist: "Run `/fix-demos` to attempt auto-fixes, or `/inboxmate-demo <company>` to rebuild specific demos"
- If all healthy and status is OK_TO_SEND: "Ready for `/setup-email-drafts`"
- If all healthy and status is SENT: "Pipeline looks good. Run `/check-outreach-status` to monitor responses"