hdls Skills

hdls Skills is a headless, MCP-native agent skill library backend. It manages folders, skills, versions, and shares, operated entirely through named tools your AI assistant calls. It stands in for tools like prompt libraries, GPT Store — same job, no UI to run.

💬 Just ask

"Create a skill for Acme, then show me my skills."

You don't call these tools yourself — just tell your assistant in plain English. Everything below is the reference for when you (or your assistant) want the exact details.

How Skills is organized

The headline entity is the skill — start there, then attach the rest to it. Records connect by reference: a skill_share links to a skill. The skill_version table is an append-only timeline — every change and note lands there and is never edited or deleted.

EntityPurposeRelates to
skillA reusable agent skill: instructions/prompt plus metadata, semantically searchable.folder
folderFolder hierarchy for organizing skills (OneDrive-style).
skill_shareGrants other tenants/users/agents access to a skill (read/use/edit).skill
skill_versionAppend-only history of skill instruction revisions.skill

Common workflows

Each line is one real sequence of tool calls — your assistant chains them for you.

  • Set up a skill end to end: create_skillcreate_folder

Tools

Call these at https://hdls.ai/api/mcp/skills. Required fields you must supply; optional fields refine the call. You never pass tenant_id — it is stamped server-side.

Create

ToolWhat it doesRequiredOptional
create_folderCreate a new folder.nameparent_id, path, data
create_skillCreate a new skill.name, slug, instructionsfolder_id, description, tags, data

Update & advance

ToolWhat it doesRequiredOptional
update_skillUpdate an existing skill by id.idname, slug, instructions, folder_id, description
share_skillShare a skill.id

Find & read

ToolWhat it doesRequiredOptional
search_skillsSearch skills by free-text and/or column filters (tenant-scoped, paginated).search, filters, limit, orderBy
list_skill_versionsSearch skill versions by free-text and/or column filters (tenant-scoped, paginated).search, filters, limit, orderBy

Field lists come from the product's live schema and are embedded in each tool's own description — read the tool description for the exact, current fields.

Field reference

Every field you can set on each record. Custom fields you add live alongside these in data.

skill

FieldTypeRequiredNotes
folder_ididLinks to a folder.
nametextYesDisplay name.
slugtextYes
descriptiontext
instructionstextYes
tagslistFree-form labels.
datajsonFree-form JSON — custom fields live here.

folder

FieldTypeRequiredNotes
parent_ididReference to a related record.
nametextYesDisplay name.
pathtext
datajsonFree-form JSON — custom fields live here.

skill_share

FieldTypeRequiredNotes
skill_ididYesLinks to a skill.
audiencetextYes
audience_idtextReference to a related record.
scopetext

Tailor it with custom fields

Add fields without a schema change. add_custom_field defines one, list_custom_fields shows what's defined, and promote_custom_field (admin) shares a personal field with the whole workspace. The value lives in each row's data (JSON) and is set and read through the normal record tools. See Products, tools & custom fields for the full model.

Roles & safety

  • Tenant isolation is automatic. Your credential is pinned to one workspace; you never pass tenant_id, and you can only ever see your own data (enforced by Postgres row-level security).
  • Role-gated. Permissions run reader < member < admin < owner. Installing a product and promoting a custom field workspace-wide need admin/owner.
  • History is append-only. The skill_version timeline is never edited or deleted — it's your audit trail.

Connect

On the concierge (https://hdls.ai/api/mcp), run install_product({ slug: "skills" }) (admin/owner) to enable it for your workspace, then add https://hdls.ai/api/mcp/skills as a connector in your assistant — see Connect your assistant.

A worked example

The literal call your assistant makes when you ask it to create the headline record:

Tool: create_skill
Arguments: {
  "name": "Acme Corp",
  "slug": "example",
  "instructions": "example"
}

See All products · Connect your assistant · Automation