Skip to main content

​
Skills

Skills are approved workflows that agents can run through Consuelo OS. A skill is more than executable code. It defines what the agent is allowed to do, what context it needs, which capabilities it depends on, what output it returns, and when a human decision is required.

​
When to use a skill

Use a skill when an agent needs a repeatable workflow. Examples:
  • generate a daily revenue brief
  • review recent calls and produce coaching notes
  • draft a follow-up after a call
  • review a campaign and produce recommendations
  • create a landing page draft
  • prepare a weekly manager report
  • summarize workspace activity into an artifact

​
Skill anatomy

PartPurpose
NameStable identifier the portal can call.
DescriptionClear explanation of what the skill does.
InputsStructured fields the agent or user provides.
ContextWorkspace records, files, memory, or artifacts the skill needs.
CapabilitiesConnected systems the skill requires.
ScriptsExecutable implementation behind the workflow.
ApprovalsHuman decision points for high-impact actions.
OutputsStructured result and artifact records.
Failure modesClear statuses for missing setup, invalid input, or execution problems.

​
Skills vs scripts

A script is implementation. A skill is the product workflow. 
skill
  -> instructions
  -> input contract
  -> capability requirements
  -> approval rules
  -> one or more scripts
  -> structured result
Most users should think in skills. Builders and coding agents can inspect scripts.

​
Skill status

Public docs should not list unfinished skills as if they are ready.
StatusUse when
AvailableThe skill can run in at least one supported mode.
Local onlyThe skill works locally but does not have cloud storage or app visibility yet.
Cloud readyThe skill works for team workspaces with app-native records.
PlannedInternal docs only unless there is a public roadmap reason.

​
Skill docs template

Each skill page should answer:
  1. What does this skill do?
  2. When should I use it?
  3. What context does it read?
  4. What can it create?
  5. Which capabilities does it require?
  6. What approval can it request?
  7. What does the result look like?
  8. What are common failure states?

​
Result shape

Every skill should return a structured result. Free-form prose can be part of the result, but it should not be the only result. A good result includes status, trace ID, summary, artifacts, source references, approval state, and safe errors.