Workflow Help

Workflow Help Skill

Overview

This skill operates as an educational guide for repository workflows. It answers questions about how the agent/skill/routing architecture works, what tools and components are available, and when to use each workflow phase (brainstorm, plan, execute). The skill prioritizes accuracy over speed by reading actual SKILL.md and agent files rather than relying on memory.

Instructions

Phase 1: UNDERSTAND THE QUESTION

Goal: Determine exactly what the user wants to know about.

Parse the user's topic and $ARGUMENTS. Common categories:

• brainstorm / plan / execute - Workflow phases
• skills / agents / hooks - Component types
• routing / do - How routing works
• subagent - Subagent-driven execution
• No argument - Provide system overview

Constraint (Over-Engineering Prevention): Answer only what was asked. Do not dump the entire system architecture when the user asks about one skill. Scope your response to the question asked, then offer to explain related concepts.

Gate: Topic identified. Proceed only when you know what to explain.

Phase 2: GATHER ACCURATE INFORMATION

Goal: Read actual files before explaining anything.

Step 1: Get authoritative data from the catalog script

This constraint (Accuracy Over Speed) is non-negotiable. Counts and listings come from scripts/list-capabilities.py, which reads the generated INDEX files — deterministic, single source of truth. Match the question to the right subcommand:

--category X filters skills by keyword in name or description (e.g. voice, game, kubernetes); agents filter by exact category field. show <name> exits 1 when the name is absent — fall back to search <name> to suggest the closest match.

Step 2: Read the actual file for deep questions

The script gives authoritative counts, names, and one-line descriptions. For anything deeper — phases, gates, capabilities, when-to-use — read the file the script names in its File: field:

• For a specific skill: Read skills/{path-from-show}/SKILL.md
• For a specific agent: Read agents/{agent-name}.md
• For routing overview: Check the /do router configuration

Extract: name, description, version, what it CAN and CANNOT do, how to invoke it, related skills or agents.

Staleness: summary, skills, and agents exit 2 and print a stderr warning when source files are newer than the INDEX. If you see that warning, tell the user to regenerate: python3 scripts/generate-skill-index.py (skills) or python3 scripts/generate-agent-index.py (agents), then re-run.

Constraint (No Fabrication): If a skill or agent does not exist, say so rather than inventing capabilities. If a skill or agent was recently deleted or merged, search with Glob for similar names and suggest the closest match.

Gate: Information gathered from actual files, not memory. Proceed only when gate passes.

Phase 3: EXPLAIN CLEARLY

Goal: Present information in the format most useful for the user's question.

For system overview, lead with live counts from python3 scripts/list-capabilities.py summary, then present the execution architecture:

Router (/do) -> Agent (domain expert) -> Skill (methodology) -> Script (execution)

Then show key workflow:

1. BRAINSTORM - Clarify requirements, explore approaches
2. WRITE-PLAN - Break into atomic, verifiable tasks
3. EXECUTE - Direct or subagent-driven execution
4. VERIFY - Run tests, validate changes

For specific components, use this format:

## [Component Name]
**Type**: Skill / Agent / Hook
**Invoke**: /command or skill: name
**Purpose**: One-sentence description
**Key Phases/Capabilities**: Bulleted list
**Related**: Links to related components

For "when to use what", use a decision table:

Constraint (Show Real Examples): Reference actual skill names, commands, and file paths from this repository. Use tables for lists when presenting available skills, agents, and commands. Include invocation syntax for each component mentioned. Apply progressive disclosure: start with overview, offer deeper detail on request. Cross-reference related skills and agents when explaining one component.

Step: Offer next steps

After explaining, ask if the user wants to:

• Learn about a related component
• Actually execute a workflow (call the indexed workflow skill)
• See more detail on a specific aspect

Constraint (Route When Appropriate): If the user wants to execute a workflow, detect the execution intent and call the indexed workflow skill instead of explaining it. For example, if "how do I debug X" means "debug X for me," use this action: Call the Skill tool with workflow. Run its systematic-debugging pipeline.

Gate: User's question answered with information from actual files.