Skills
The full library of HotCRM AI skills — what each one does, where it appears, and what data it uses.
Skills
A skill is a bite-sized AI ability the platform assistant can perform — like "qualify this lead" or "draft a customer reply". Skills are all HotCRM contributes to the assistant; they are documented in two groups, Sales Skills and Service Skills.
Why skills, not "one big AI"
Packaging the capability as skills rather than one prompt makes it:
- Predictable — each skill has a defined input, output, and behavior.
- Auditable — admins can see what each skill reads from and writes to.
- Improvable — you can refine one skill (e.g., better lead qualification) without touching others.
- Permission-aware — skills inherit the user's record visibility.
Built-in skills
HotCRM ships with six core skills:
| Skill | Used by | What it does |
|---|---|---|
| 🔄 Live Data | Sales + Service | Reads the live schema and the live records — every turn |
| 🎯 Lead Qualification | Sales | Scores a lead on BANT, recommends next action |
| ✍️ Email Drafting | Sales + Service | Drafts a context-aware email or reply |
| 📈 Revenue Forecasting | Sales | Summarises pipeline by stage and forecasts a range, naming the deals at risk |
| 🔭 Customer 360° | Sales + Service | One-page summary of an account or contact |
| 🚦 Case Triage | Service | Suggests a priority with the reason that drove it |
🔄 Live Data
| Input | Any natural-language question about CRM data ("which accounts…", "show me the top…") |
| Reads | The live object metadata (describe_object, list_objects) and the live records (query_records, get_record, aggregate_data) at the moment of the question — no cached snapshot |
| Output | Direct answer with cited record IDs |
| Writes | Nothing — read-only |
| Knowledge base | None — purely metadata + records |
This is the skill behind Wow #1 — Live Schema: add a field in the admin console, and the assistant uses it on the very next turn. Every other Sales / Service skill composes on top of Live Data, so the same live-schema property applies transitively — Lead Qualification can score a lead on a custom field added 10 seconds ago.
🎯 Lead Qualification
| Input | A lead record — the skill activates on any crm_lead |
| Reads | The lead itself (describe_object for crm_lead, then get_record) and its related activity history (query_records). Not campaign engagement, and not matched-account history: neither is in the skill's instructions |
| Output | A 0–100 BANT score, one line of justification per dimension naming the field or activity it read, and the single recommended next step — with the lead ID cited. It does not return a Hot/Warm/Cold rating or a recommended status |
| Writes | Nothing itself — the two outcomes are real Actions it calls. At 70+ with your agreement, Convert Lead (action_convert_lead), whose flow creates the account, contact and optional opportunity and stamps the lead is_converted / status: converted; otherwise Schedule Follow-up (action_schedule_followup), whose flow files a crm_task and stamps the lead's next follow-up date. rating is never written by either — it is the 1–5 star Lead Score a person sets |
| Knowledge base | None. No knowledge step — and no Sales Knowledge library exists to have one (see Knowledge Bases) |
✍️ Email Drafting
| Input | A goal sentence plus the record it is about — the instructions ground the copy in the contact, and in the related account or opportunity when the request names one |
| Reads | The contact record (get_record), plus that related account or opportunity; query_records for the recent history |
| Output | A draft subject and body — subject under 60 characters, one clear ask, no more than 150 words — plus a second subject-line variant with a recommendation on which of the two to use |
| Writes | Nothing, and it has no way to: Send Email on the contact carries no AI exposure, so no send tool is materialised for the skill. It shows the draft and points at that button, where a person reviews and sends |
| Knowledge base | None. The drafting instructions never reach for an article, and the three libraries once named here do not exist |
📈 Revenue Forecasting
| Input | A question about pipeline health, forecast, risk or slippage + the period it covers (default: the current quarter) |
| Reads | crm_opportunity — its current shape via describe_object, count and summed amount grouped by stage via aggregate_data, and the at-risk deals via query_records (close date already past, no activity in 30+ days, stage unchanged too long). No field-history read is in the instructions; stage age comes off the record's own Days in Stage |
| Output | Pipeline by stage with weighted value (amount × probability computed from what it read), the at-risk deals named with the signal behind each, and a forecast given as a range — commit-only (high-probability stages) to full weighted pipeline — plus a chart via visualize_data when one carries the answer better. The Closed / Commit / Best Case / Pipeline buckets are not this skill's output: they are crm_opportunity's Forecast Category field and the amounts the scheduled forecast snapshot writes onto Forecast records |
| Writes | Nothing — read-only |
| Knowledge base | None. The methodology is in the skill's own instructions, not in a library |
🔭 Customer 360°
| Input | An account or contact |
| Reads | crm_account, then filtered to it: crm_contact, open crm_case (is_closed false), open crm_opportunity, and published crm_knowledge_article — with totals from aggregate_data rather than added up row by row. Contracts, recent activities and campaign memberships are not read: the skill carries query_records and could reach crm_contract, crm_campaign, crm_event and crm_task, but its instructions never tell it to (same finding as the Service Skills page) |
| Output | Three sections — Account Snapshot · Active Work · Risks & Notes — with record IDs cited inline for deep linking. Every risk names its record and the signal that raised it; a section with no records is reported as empty. There is no contracts or engagement section |
| Writes | Nothing — read-only |
| Knowledge base | Published Knowledge Articles, matched on the category or tags of the cases it just read — the only skill that reads them |
🚦 Case Triage
| Input | A case — the skill activates on crm_case |
| Reads | The case itself: describe_object for crm_case, then get_record for the case at hand, plus the customer context its rubric weighs first (Customer Tier, contract value). It declares no query tool, so prior cases are out of reach — and there is no product to read: crm_case has no product field under any name and no lookup to crm_product |
| Output | A priority (Low / Medium / High / Critical) with the ONE reason that drove it, citing the case ID and the field values used, then a pointer to Escalate Case with a ready-to-paste reason or Close Case with a resolution summary. No category, no team and no article list; the customer-facing reply is handed to Email Drafting |
| Writes | Nothing. Its tools are describe_object and get_record only, and the two Actions that would change the case — escalate_case / close_case — carry no AI exposure, so no write tool is materialised for it to call. The category and queue this row used to name are not fields on crm_case under any name (its own classification field is Case Type: Question / Problem / Feature Request / Bug, set by a person). Triage supplies the judgement; the agent clicks the button |
| Knowledge base | None. It declares no query tool, so it cannot search articles at all |
How skills work together
Skills do not call each other. The assistant selects the skill that fits the question you asked, and one question runs one skill — nothing chains a second skill behind it.
The only handoff written into a skill's own instructions is Case Triage's: for the customer-facing reply it names Email Drafting by name instead of writing the copy itself. That handoff is a pointer for you, so ask for the draft as your next question and that skill takes it. Case Triage does not invoke Customer 360 to weigh how valuable the customer is, and Email Drafting does not invoke it for account context — neither call exists. When you want the account history behind a triage or a draft, ask Customer 360° for it yourself; it is the skill that carries query_records.
What every skill does share is the data floor: each one reads through the same live metadata and record tools, so Live Data's live-schema property applies to all of them.
What every skill respects
- Sharing rules — a skill running for user X can only read records visible to X.
- Field-level security — restricted fields are hidden from skill input.
- Audit log — every skill invocation, with input and output, is logged for review.
- Confirmation — Lead Qualification is the only skill with a write path at all, and it writes through real Actions rather than fields: Convert Lead is irreversible, so it asks for confirmation and an agent-invoked conversion waits for human approval instead of running unattended; Schedule Follow-up is additive and reversible, so it carries no approval gate. The other five skills declare no write tool.
Adding your own skills
Admins and developers can build custom skills — for example:
- Generate a renewal pitch for a contract about to expire.
- Score account health combining usage, NPS, and case volume.
- Draft a release-notes email when a product hits a milestone.
See Customization › AI Skills for how to define a new skill, what knowledge bases it can use, and how to test it.
Tips for admins
- ✅ Monitor skill usage in the AI dashboard — which skills are used most, which have low acceptance.
- ✅ Tune knowledge bases for the skills that under-perform — most accuracy issues are knowledge issues.
- ✅ Disable skills that don't fit your business — each skill can be toggled in admin settings.