Cross-cutting agents and skills not tied to a specific technology. Installed externally via sparkdock.
agents/system/<agent-name>/
├── copilot/ # GitHub Copilot profile (.agent.md)
└── opencode/ # OpenCode profile (.md)
skills/system/<skill-name>/
└── SKILL.md # Skill definition (plus optional bundled assets)
System agents support multiple tools (Copilot, OpenCode). Each tool gets its own file in a subfolder, but the prompt body is kept identical across tools -- only the YAML frontmatter differs to match each tool's configuration format. There is no shared standard yet.
| Tool | Install path |
|---|---|
| Copilot | ~/.copilot/agents/<agent-name>.agent.md |
| OpenCode | ~/.config/opencode/agents/<agent-name>.md |
System skills follow the standard skill format (folder with SKILL.md). They
may include bundled assets such as scripts, reference data, or eval definitions.
Skills are installed to a shared standard path used by all coding agent tools:
| Install path |
|---|
~/.agents/skills/<skill-name>/ |
- the-architect -- Conversational AI oracle for discovery, brainstorming, architecture, and general knowledge. Not a code agent.
- glab -- GitLab CLI skill for working with issues, merge requests, CI/CD pipelines, and repositories via the
glabCLI. - gh -- GitHub CLI skill for working with issues, pull requests, Actions workflows, releases, and repositories via the
ghCLI. - playwright-cli -- Browser automation with the
playwright-cliCLI tool for web testing, screenshots, form filling, and data extraction. Synced from microsoft/playwright-cli with custom output file conventions. - skill-creator -- Create, iterate, and benchmark agent skills with eval-driven workflows. Synced from anthropics/skills with custom tool-agnostic usage guidance.
- doc-coauthoring -- Structured workflow for co-authoring documentation, proposals, and technical specs. Synced from anthropics/skills.
- create-agentsmd -- Prompt for generating an AGENTS.md file for a repository. Synced from github/awesome-copilot.
- code-security-audit -- Multi-phase code security audit workflow (discovery, container generation, native scans, Docker scans, manual review) for web applications and APIs. Open-source tools only.
- agentic-security-audit -- Audit AI agent configurations, instruction files, and LLM integration code for security risks mapped to the OWASP Top 10 for Agentic Applications (2026).
- githuman -- Review AI-generated code before committing via GitHuman Docker instances managed by Just recipes (
sjust/ajust). Synced from mcollina/githuman-skills with SparkFabrik-specific commands, review workflow, and troubleshooting.