| name | Sidecar Skill Governance |
| description | Architectural rules and standards for creating Sidecar Skills to support Workflows. |
| source | catalog |
Sidecar Skill Governance
Version: 1.0.0 | Type: Governance Skill
This skill defines the structural standards and OSS compliance rules for creating Sidecar Skills.
A Sidecar Skill is a data companion to a Workflow of the same name. It offloads static data (templates, checklists) and complex scripts from the workflow file, keeping the workflow "Zero-fluff" and strictly focused on process logic.
1. Core Principles
- Workflow = Process Logic: The
.agents/workflows/<name>.md file must only contain sequential steps and router calls. It should not contain heavy text blocks or large bash scripts.
- Skill = Data Companion: The
.agents/skills/<name>/SKILL.md file serves as the index and router for data, templates, and scripts.
- Lazy Loading (Just-In-Time): Data is only read by the workflow when a specific step requires it (e.g., using
cat).
- Graceful Degradation: A workflow should still function (or fail gracefully with a warning) if its sidecar skill or specific templates are missing.
2. OSS English-First Governance (CRITICAL)
All Sidecar Skills must comply with the English-First rule for Open Source Repositories:
- All internal files within
.agents/skills/ (including SKILL.md, templates in references/, and scripts in scripts/) MUST be written in English.
- This rule ensures the workspace structure can be shared, reviewed, and extended by global open-source contributors, regardless of the end-user's localized preferred language for code generation.
3. Directory Structure
A compliant Sidecar Skill must follow this exact structure:
.agents/skills/[workflow-name]/
├── SKILL.md ← Router & Governance Index (Required)
├── scripts/ ← Bash scripts to offload CLI logic (Optional)
│ └── example.sh
└── references/ ← Static data, markdown templates, checklists (Optional)
└── example-template.md
4. Extraction Thresholds (When to use a Sidecar)
When designing or updating a workflow, extract content to a Sidecar Skill if:
- Templates > 20 lines: Any markdown template, prompt skeleton, or formatted output block should be placed in
references/.
- Scripts > 10 lines: Any complex bash logic (loops,
jq parsing, heavy sed/awk replacements) must be extracted to an .sh file in scripts/.
- Checklists: QA or code review checklists should be maintained in the skill, not embedded in the workflow.
5. Implementation Guide
Defining the Router Table (SKILL.md)
The SKILL.md must contain a "Router" table so the Agent knows what files are available:
## Resource Router
| Resource | Relative Path | Description |
| :-- | :-- | :-- |
| Main Template | `references/main-template.md` | Used in step 3 to format the output |
| Build Script | `scripts/build.sh` | Executed to compile assets |
Invoking from Workflow
Inside the workflow file, use // turbo bash blocks to invoke scripts or load templates:
cat .agents/skills/[workflow-name]/references/template.md 2>/dev/null || echo "⚠️ Template not found"
bash .agents/skills/[workflow-name]/scripts/build.sh
6. Resource Router
This skill provides ready-to-use boilerplate templates to quickly scaffold new sidecar skills:
| Resource | Relative Path | Description |
|---|
| Workflow Sidecar Template | references/workflow-sidecar-template.md | Boilerplate for creating a sidecar skill attached to a global workflow. |
| Project Sidecar Template | references/project-sidecar-template.md | Boilerplate for creating a domain-specific context skill for a project. |