SkillsFlow

SkillsFlow lets you define deterministic, multi-step pipelines in workflows/ as YAML. Chain skill:, agent:, and tool: steps with dependency ordering, template data flow, and per-step context via inputs.

SkillsFlow Example

name: code-review-flow
description: Full code review pipeline

steps:
  - id: lint
    action: "Run static analysis on changed files"
    skill: static-analysis
    inputs:
      path: ${{ trigger.changed_files }}

  - id: review
    action: "Review code for security and performance issues"
    agent: code-reviewer
    depends_on: [lint]
    inputs:
      findings: ${{ steps.lint.outputs.issues }}
      focus: "Focus on security and performance. Flag any use of eval() or raw SQL."

  - id: test
    action: "Run the test suite with coverage"
    tool: bash
    depends_on: [lint]
    inputs:
      command: "npm test -- --coverage"

  - id: report
    action: "Summarise review findings and test coverage"
    skill: review-summary
    depends_on: [review, test]
    conditions:
      - ${{ steps.review.outputs.severity != 'none' }}
    inputs:
      review: ${{ steps.review.outputs.comments }}
      coverage: ${{ steps.test.outputs.report }}

error_handling:
  on_step_failure: escalate
  escalation_target: "eng-reviews"

Key Concepts

Concept	Description
Deterministic Execution	Skills run in declared order, not LLM discretion. Every run follows the same path.
Data Flow via Templates	Chain outputs to inputs with `${{ steps.X.outputs.Y }}` template syntax.
Per-Step Context via inputs	Pass extra guidance to a step through `inputs:`: the receiving skill or agent reads it as a named input.
Mixed Step Types	Combine `skill:`, `agent:`, and `tool:` steps in a single flow for maximum flexibility.

Execution Pipeline

lint   (skill)    — runs immediately (no depends_on)
review (agent)    — runs after lint completes
test   (tool)     — runs after lint completes (parallel with review)
report (skill)    — runs after both review and test complete

Steps with depends_on block until their dependencies finish. Steps without depends_on run immediately.

Step Types

Type	Key	Description
Skill	`skill: <name>`	Runs a SKILL.md module
Agent	`agent: <name>`	Delegates to a sub-agent defined in agent.yaml
Tool	`tool: <name>`	Executes a tool directly (e.g. `bash`, `read`, `write`)

SkillsFlow workflows live in the workflows/ directory of your agent repo: version-controlled alongside everything else.

Skills Overview

SKILL.md format and the skill discovery system

Agent Structure

workflows/ directory in the full repo layout

Design Patterns

SkillsFlow and other git-native patterns

CLI

Run and trigger workflows via CLI

Getting Started

Agent Structure

Skills & Workflows

CLI & Adapters

Compliance

FAQ

SkillsFlow Example

Key Concepts

Execution Pipeline

Step Types

Skills Overview

Agent Structure

Design Patterns

CLI

​SkillsFlow Example

​Key Concepts

​Execution Pipeline

​Step Types

Skills Overview

Agent Structure

Design Patterns

CLI

SkillsFlow Example

Key Concepts

Execution Pipeline

Step Types