Skip to main content
Skills are reusable knowledge packages — coding guidelines, domain expertise, workflow instructions. Instead of stuffing everything into your prompt, skills use progressive disclosure: the agent sees a brief manifest, then loads full instructions on demand.
LevelWhat the agent seesWhen
0 — ManifestName + descriptionAlways in system prompt
1 — InstructionsFull skill content + reference listAgent calls LoadSkill
2 — ReferencesDeep reference documentsAgent calls ReadSkillReference

​
Quick Start

Pass skill names in the skills array: 
run = client.run(
    engine="tim-gpt",
    input={
        "instructions": "Build a REST API with proper error handling",
        "tools": [{"type": "platform", "id": "web_search"}],
        "skills": ["api-design", "error-handling"],
    },
    options={"await_completion": True},
)
When skills are present, LoadSkill and ReadSkillReference tools are auto-injected — you don’t configure them. If you attach skills to a saved agent in the dashboard, they’re included automatically on every run (API, playground, cron, Composio) without passing skills in the request.

​
Visibility

LevelWho can use itCreated by
PlatformEveryoneSubconscious (curated)
PublicEveryoneAny organization
OrgYour org onlyYour organization
New skills default to org.

​
Skills API

CRUD endpoints under /v1/skills: 
GET    /v1/skills              # List all visible skills
POST   /v1/skills              # Create a skill
PUT    /v1/skills/:id          # Update a skill
DELETE /v1/skills/:id          # Delete a skill

​
Create Example

POST /v1/skills
{
  "name": "my-coding-standard",
  "displayName": "My Coding Standard",
  "description": "Enforces our team's coding conventions",
  "instructions": "When writing code, follow these rules:\n1. Use TypeScript strict mode\n2. ...",
  "author": "your-team"
}
name must be kebab-case and unique within your org. You can only edit/delete org-owned skills — platform skills are read-only.

​
Skill References

Reference documents are detailed specs or examples that the agent loads via ReadSkillReference. References support nesting via parentReferenceId. 
GET    /v1/skills/:skillId/references              # List references
POST   /v1/skills/:skillId/references              # Add a reference
PUT    /v1/skills/:skillId/references/:refId       # Update a reference
DELETE /v1/skills/:skillId/references/:refId       # Delete a reference

​
Dashboard

Manage skills at subconscious.dev/platform/skills:
  • Browse platform skills — pre-built, curated by Subconscious
  • AI Skill Builder — describe what you need, generate automatically
  • GitHub import — paste npx skills add commands or browse repos
  • Attach to agents — checkbox in the agent config panel

Tools

Tools give agents actions; skills give agents knowledge

Runs

How skills and tools work together in runs