hubspot

Files (7)

• SKILL.md
• scenarios.json · evals/
• deal-properties.json · fixtures/
• hubspot-plan.cjs
• hubspot-requests.cjs
• hubspot-validation.cjs
• hubspot.cjs

SKILL.md (first lines)

# HubSpot

Use this skill for HubSpot CRM work when the operator has connected HubSpot to
HybridClaw with a HubSpot Service Key, legacy private app access token, or
OAuth credentials.

## Scope

- read contacts, companies, and deals
- search high-frequency CRM objects by common properties
- inspect HubSpot CRM properties for contacts, companies, and deals
- update a deal's `dealstage`
- update contact, company, or deal `lifecyclestage`
- create timeline notes associated with contacts, companies, or deals
- create tasks associated with contacts, companies, or deals
- plan common natural-language CRM requests before building API calls
- build ordered natural-language workflows with lookup, property-validation, and
  write steps for the high-frequency CRM operations
- validate internal HubSpot stage option values from saved property metadata
- interpret common HubSpot authentication, authorization, stage, and rate-limit
  errors
- measure cost per assistant run through normal HybridClaw `UsageTotals`

## Credential Rules

For normal single-account HubSpot use, create a HubSpot Service Key and store
it as `HUBSPOT_ACCESS_TOKEN` in HybridClaw encrypted runtime secrets. Service
Keys are HubSpot's recommended account-level, system-to-system bearer
credential for direct REST API requests when OAuth and webhooks are not needed.
Never paste tokens into a prompt.

Create the Service Key in HubSpot under **Development > Keys > Service keys**
or open <https://app.hubspot.com/service-keys>. Give it only the scopes needed
for the requested CRM operations, copy the key, and store that value as
`HUBSPOT_ACCESS_TOKEN`. Do not use a HubSpot Personal Access Key or Developer
Key for this skill; those are for other HubSpot developer workflows and are not
valid CRM REST bearer tokens for these calls.
HubSpot Service Key docs:
<https://developers.hubspot.com/docs/apps/developer-platform/build-apps/authentication/account-service-keys>.

Recommended setup order:

1. Browser admin: open the active HybridClaw admin URL ending in `/admin/secrets` and set
   `HUBSPOT_ACCESS_TOKEN`.
2. Browser `/chat` or TUI fallback:
   `/secret set HUBSPOT_ACCESS_TOKEN <token>`.
3. Local console fallback:

```bash
hybridclaw secret set HUBSPOT_ACCESS_TOKEN
```

Equivalent setup when passing the Service Key explicitly:

```bash
hybridclaw auth login hubspot \
  --access-token "<hubspot-service-key>" \
  --account sales@example.com
```

Legacy private app access tokens can also be stored as
`HUBSPOT_ACCESS_TOKEN`, but new single-account REST integrations should prefer
Service Keys.

OAuth client credentials are also supported for public app installations:

```bash
hybridclaw auth login hubspot \
  --client-id "<hubspot-oauth-client-id>" \
  --client-secret "<hubspot-oauth-client-secret>" \
  --refresh-token "<refresh-token>"
```

The gateway injects `HUBSPOT_ACCESS_TOKEN` only when an `http_request` uses
`bearerSecretName: "HUBSPOT_ACCESS_TOKEN"` against HubSpot API hosts. With OAuth
credentials, the gateway mints the access token from `HUBSPOT_CLIENT_SECRET` and
`HUBSPOT_REFRESH_TOKEN`; with Service Key or legacy private app setup, the
stored bearer credential is used directly.

Required HubSpot scopes depend on the task. The default OAuth login scope set
covers contacts, companies, deals, notes, tasks, CRM schema reads, and `oauth`.

## Default Workflow

1. Start with reads or `plan`. Do not mutate CRM state unless the user clearly
   asks for the exact write.
2. Run the bundled helper to build an `http_request` wrapper:
   ```bash
   node skills/hubspot/hubspot.cjs ...
   ```
3. Pass only the emitted `httpRequest` object to the built-in `http_request`
   tool when dry-running or when the helper cannot reach the gateway. For live
   HubSpot reads and writes, use the helper `run` command so the CJS script owns
   request construction, gateway submission, and auth-error handling.
4. Never handcraft HubSpot `http_request` calls from memory. The helper owns
   endpoint selection, method, payload, secret refs, and auth-error handling.
5. For natural-language writes, run `workflow` first. It emits ordered
   property metadata, lookup, operator confirmation, and write steps.
6. For writes, confirm the target record and intended field change, then pass
   either the exact `--grant` value or `--operator-grant`.
7. Use internal HubSpot IDs for write targets. If a name search returns multiple
   records, stop and ask for the exact record ID.
8. Use internal stage values for `dealstage`, `pipeline`, and
   `lifecyclestage`. Read `/crm/v3/properties/...` first when the internal
   value is unknown.
9. If a live HubSpot call returns 401 or 403, stop after that first failure. Do
   not retry, do not call more HubSpot endpoints, and do not guess from dates or
   epoch timestamps. Tell the operator to verify or replace
   `HUBSPOT_ACCESS_TOKEN`. For Service Keys, they should copy the current key
   from HubSpot's Service keys page or rotate it if HubSpot says it was revoked,
   expired, exposed, or invalid.

## Command Contract

Plan a request without authentication:

```bash
node skills/hubspot/hubspot.cjs --format json plan "Move the Acme Renewal deal to contractsent and add a follow-up task"
```

From HybridAIOne/hybridclaw, MIT.

More from HybridAIOne/hybridclaw