Skills — reusable playbooks

A skill is a battle-tested playbook your AI assistant can pull in on demand — a framework for reviewing a sales pipeline, a rubric for qualifying a lead, a disciplined recipe for scoring account health. Instead of re-explaining a whole process every time, you point your assistant at a skill and it applies the playbook.

💬 Just ask

"Find a playbook for reviewing my sales pipeline and walk me through it."

"Is there a skill for qualifying a new lead?"

You don't call these tools yourself — just tell your assistant; the technical reference below is for when you want the details.

Skills live on the concierge — the control-plane endpoint at https://hdls.ai/api/mcp. Connect there once and two tools appear alongside product discovery: search_skills and get_skill. Both are open to every role, including a read-only reader — discovering and reading a skill is pure, no-write discovery.

At a glance

You want to…	Tool	Who can use it
Find a relevant playbook	`search_skills`	everyone (reader+)
Read a playbook in full	`get_skill`	everyone (reader+)

Mental model. The skills registry is a global, read-only catalog — the same curated set is visible to every workspace, like a public package index. You discover and apply skills; you don't author them. There are around 45 curated skills today, and the catalog grows over time.

What a skill actually is

Every skill carries a stable slug, a name, a description, a list of tags, and (optionally) a product_slug tying it to one product (e.g. crm). It also has a kind that decides how your assistant consumes it:

prompt — an inline markdown playbook. get_skill returns the full text in a content field, ready to apply immediately. There is no download and no install step — the text is the skill.
bundle — a downloadable skills.zip. get_skill returns a download_url plus install instructions: download the zip and unpack it into your assistant's skills directory. Published bundles download without authentication — the download_url is public.

Either way, every result also includes a public download_url of the form https://hdls.ai/api/skills/<slug>, so you can always install from a URL.

Find a skill — `search_skills`

Search by keyword over a skill's name, description, tags, and product. Pass the plainest phrase that describes what you're trying to do.

// tool: search_skills   (on https://hdls.ai/api/mcp)
{
  "query": "pipeline review"
}

Optional arguments let you narrow or widen the search:

Argument	Type	Default	Notes
`query`	string	—	Keyword over name, description, tags, and product. An empty query lists the catalog (newest first) — the easiest way to browse what exists.
`productSlug`	string	—	Restrict to one product, e.g. `"crm"`.
`limit`	number	`10`	Max results. Capped at `50`.

Each result is a compact summary you can scan — slug, name, description, tags, kind, product_slug, and a download_url:

{
  "query": "pipeline review",
  "count": 1,
  "skills": [
    {
      "slug": "crm-pipeline-review",
      "name": "CRM Pipeline Review",
      "description": "Pipeline Review framework: run a disciplined weekly review of every open deal…",
      "tags": ["crm", "sales", "pipeline", "forecast", "deals"],
      "kind": "prompt",
      "product_slug": "crm",
      "download_url": "https://hdls.ai/api/skills/crm-pipeline-review"
    }
  ]
}

Tip. Run search_skills with no query to dump the whole catalog, then filter by eye. Add productSlug when you only want skills for the product you're working in.

Read a skill in full — `get_skill`

Once you've found a slug, read the whole playbook:

// tool: get_skill   (on https://hdls.ai/api/mcp)
{
  "slug": "crm-pipeline-review"
}

For a prompt skill, the response includes the complete markdown in content — your assistant applies it directly, no further steps.
For a bundle skill, the response gives you the download_url and install instructions: download the skills.zip and unpack it into your assistant's skills directory.

That's the whole loop: search → read → apply. No registration, no per-skill configuration, and nothing to maintain.

How your assistant pulls a skill in

In practice you rarely call these tools by hand. You describe the work, and your assistant does the discovery for you:

"Run a proper weekly pipeline review on my open deals."

Behind the scenes your assistant typically:

Calls search_skills({ query: "pipeline review", productSlug: "crm" }) to find a matching playbook.
Calls get_skill({ slug: "crm-pipeline-review" }) to pull the full framework in.
Applies that framework against your live data on the product endpoint (e.g. https://hdls.ai/api/mcp/crm) — reading deals, checking stage hygiene, and producing the forecast the playbook describes.

The skill supplies the method; your product tools supply the data. The result feels less like calling an API and more like working with a colleague who already knows the playbook.

Skills that ship today

The catalog spans the product portfolio; these CRM playbooks are seeded and searchable right now and are a good way to see the shape of a skill:

Slug	What it gives you
`crm-pipeline-review`	A weekly framework for reviewing open deals: stage hygiene, slippage, and a commit / best-case / pipeline forecast call.
`crm-account-health`	A 0–100 account health score with a Red / Yellow / Green band, derived from engagement, relationship depth, usage, support load, and commercial signals.
`crm-lead-qualification`	A BANT + MEDDIC rubric: the questions to ask, what to record on the lead, and the threshold to convert it into a deal.

A search_skills with an empty query always shows you the current set — treat the table above as a sample, not the whole catalog.

Good to know

Open to every role. search_skills and get_skill work for reader, member, admin, and owner alike — reading a playbook changes nothing in your workspace.
Global, not per-workspace. Skills are not tenant data. Everyone sees the same registry, and a skill never carries another workspace's information.
No tenant_id, ever. Like everything on the concierge, you never pass a workspace id — but skills don't touch your data at all, so there's nothing to scope.
Public download. A published skill's download_url is fetchable without auth, so you can share or script installs freely.

Where to go next

New here? Get connected to the concierge first, then browse the catalog with search_skills({}) (empty query) on https://hdls.ai/api/mcp.
See the full product catalog — skills apply their method against the live data those products hold.
Keep context between sessions with Workspace memory so your assistant remembers how your team works.
Bring colleagues in with Teams & roles.