CLI Harnesses
Route commands for beads, gwc, commiat, docker, kubectl, gh, aws, terraform, and more via manifests or passthrough adapters.
Skill Discovery Layer
Agents query a single skill graph.
supercli caches schemas, inputs, outputs, and rules for every harness. Agents explore skills via search, inspect action metadata, and trust consistent envelopes and exit codes.
Example
supercli skills search "database"
supercli skills get beads.issue.create
supercli plan beads issue create --title "Task"Skill Sources
Every tool becomes a skill.
supercli ingests multiple ecosystems so the router stays agnostic to execution details.
OpenAPI & HTTP
Import specs, auto-generate skills, and let agents execute normalized HTTP requests with schema validation.
MCP + Workflows
Expose MCP tools, AI-driven plans, and declarative workflows as first-class skills with consistent envelopes.
Architecture
From agent request to tool execution.
Skills flow through a discovery layer, router, and harness/adapters that target CLIs, APIs, or custom runtimes.
Agent → Skill Discovery → Router → Harness / Adapter → Tool- Discovery – catalog every skill with metadata & policy.
- Routing – resolve namespaces to adapters automatically.
- Execution – normalize inputs/outputs and exit codes.
- Extension – drop in new harnesses, plugins, or specs.
Skill Mesh Vision
Toward a universal skill mesh.
supercli is growing into a mesh that handles discovery, routing, execution, composition, and governance for every capability in your environment.
- Graph-native discovery & tagging
- Execution DAG observability
- Agent-facing HTTP/WebSocket endpoints
- Policy controls across harnesses
Follow along in the docs for upcoming mesh capabilities and long-term direction.