Workflow orchestrator

Workflow Orchestrator

Purpose

Coordinate the execution of multiple skills in sequence or parallel to achieve a complex goal. Handles dependency resolution, data passing between skills, error recovery, and progress reporting. The top-level orchestrator that makes multi-step agentic workflows reliable.

When to Use

• A complex goal requires 3+ skills in sequence (e.g., research → plan → build → test)
• Multiple independent subtasks can run in parallel
• A pipeline needs retry logic or error recovery
• The user wants a multi-step workflow executed end-to-end

When Not to Use

• Only one skill is needed — invoke it directly
• The steps are manual and need user approval at each stage
• The workflow is a simple linear script — use a script instead

Inputs

Input	Type	Required	Description
goal	string	Yes	The overall objective
pipeline_definition	array	No	Explicit skill order if known (auto-generated if omitted)
fallback_strategy	enum	No (default: stop)	stop, retry, skip

Expected Outputs

Output	Description
Pipeline result	The output of the final skill
Execution trace	What ran, in what order, what each produced
Status report	Success, failures, and retries

Exact Workflow Steps

1. Accept the goal from the user or from a command
2. Decompose the goal into ordered skills using planner skill
3. Identify dependencies between skills (blocking, parallel, sequential)
4. Execute skills in dependency order:
   • Parallel skills run concurrently
   • Sequential skills run in order, passing outputs
5. After each skill, validate output using validation-engine
6. On failure: retry, skip, or stop based on fallback strategy
7. Report results to user with execution trace

Required Checks

• All skills in the pipeline are available
• Dependency order is correct (no cycles)
• Each skill's output is valid before passing to next
• Fallback strategy is defined for every pipeline step

Failure Modes

Failure Condition	Response
Required skill not available	Report missing skill and suggest alternatives
Pipeline step fails after retries	Report failure with partial results
Circular dependency detected	Restructure to eliminate the cycle

Examples

Example: Feature development pipeline

1. planner → decompose: code design, implement, test, document
2. architecture-designer → design module structure
3. code-builder → implement (depends on architecture-designer)
4. test-generator → create tests (parallel with docs)
5. validation-engine → validate all outputs
6. Report results

References

Reference	Path
Planner skill	../planner/SKILL.md
Validation Engine skill	../validation-engine/SKILL.md
Command Executor skill	../command-executor/SKILL.md