DoubleX AgentesDocs

API reference › Resources

Skills

Create and maintain reusable instructions to attach to any agent.

The Skill object#

JSON
{
  "id": "skl_3e8f1a2b4c5d",
  "name": "Sales proposal",
  "description": "Builds sales proposals following the company standard, with scope, timeline and price.",
  "instructions": "# Sales proposal\n\n1. Start with the problem the customer has…",
  "scope": "team",
  "teamId": "team_9c8b7a6d5e4f",
  "ownerId": "usr_1a2b3c4d5e6f",
  "canEdit": true,
  "usedBy": [{ "id": "agt_8f3a2b1c9d0e", "name": "CRM Analyst" }],
  "createdAt": "2026-10-23T09:00:00.000Z",
  "updatedAt": "2026-10-26T17:45:00.000Z"
}

GET/api/skills#

Lists the skills you can see: { "skills": [Skill, ...] }. To read a single skill, filter the list by id.

curl https://doublex.ai/api/skills -H "Authorization: Bearer $DOUBLEX_API_KEY"

POST/api/skills#

FieldTypeDescription
namestring, 2 to 80Required.
descriptionstring, 2 to 1024Required. When the agent should use the skill.
instructionsstring, 2 to 30,000Required. The instructions, in Markdown.
scopepersonal, team, organizationDefault personal.
teamIdstringThe team, when scope is team.
curl -X POST https://doublex.ai/api/skills \
  -H "Authorization: Bearer $DOUBLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Sales proposal",
    "description": "Builds sales proposals following the company standard.",
    "instructions": "# Sales proposal\n\n1. Start with the problem the customer has…"
  }'

Response 201: { "skill": Skill }.

PATCH/api/skills/{id}#

Changes name, description and instructions (all optional, with the same limits). Agents that use the skill switch to the new version on their next message. Requires edit permission; without it, the response is 404.

Response: { "skill": Skill }.

DELETE/api/skills/{id}#

Deletes the skill and removes it from the agents that used it. Response: { "ok": true }.

Skill packages#

Skills also support command (without /), invocation (auto, manual, both), dependencyIds, files and actions. Related skills are procedures loaded by the same agent; subagent delegation remains separate. The general assistant can discover visible skills; specialist agents use attached skills and their dependencies.

New files use { path, mediaType, encoding: "utf8" | "base64", content }. Updates can retain existing files using their metadata (path, mediaType, size, sha256); this can only reference files already belonging to the same skill. Limits: 100 files, 2 MB per file, 10 MB total. References stay outside the instruction text limit.

  • GET /api/skills/:id: skill metadata, instructions, file and action indexes.
  • GET /api/skills/:id/files?path=references/example.md: authorized file download.
  • GET /api/skills/:id/package: export .skill with SKILL.md, attachments and doublex.skill.json.

Actions have id, name, description, enabled and kind. Script actions add path and runtime (python, node, shell); connector actions add connectorId and tool. Connector actions require an active, attached account available to the runner and obey connection and agent restrictions. Importing a skill grants no account access.

Scripts receive a JSON object on stdin and return text or JSON on stdout. Package files are mounted read-only at /skill. Isolated execution has no network or host credentials. Limits: 64 KB input, 128 MB memory, half a CPU, 32 processes, 35 seconds and two concurrent executions. Python provides its standard library; Node.js provides built-in modules. Additional libraries are not installed during execution.

In chat, /command request invokes the skill for the current request. Tools listar_skills, usar_skill, ler_arquivo_skill, executar_acao_skill provide discovery and on-demand loading. Each call rechecks access, dependency cycles are rejected, and the conversation records skill, file and action usage. Existing skills keep working.