SIM-ONE Alpha combines product-shipped Flue capabilities with runtime capabilities added by a user or the agent.
| Layer | Contents | Lifecycle |
|---|---|---|
| Built-in Flue layer | Product skills, tools, workers, workflows, and MCP connections | Shipped with the product |
| SIM-ONE runtime registry | User- or agent-added skills, tools, workers, and MCP servers | Stored outside the product artifact and loaded after restart |
Both layers enter the same Flue skill, tool, and subagent surfaces. The runtime registry adds extensibility without giving installed capabilities authority over protocols or approvals.
| Type | Purpose | Default |
|---|---|---|
| Skill | Reusable instructions, procedures, and supporting resources | Enabled when added |
| Tool | Typed executable action attached to an owning agent | Disabled unless enabled |
| Worker | Specialized executor loaded as a Flue subagent profile | Disabled unless enabled |
| MCP server | Remote HTTP or HTTPS service contributing tools | Disabled unless enabled |
Protocols are not capabilities. Protocols are mandatory runtime rules stored in SQLite and loaded through the Protocol Tool.
The authoritative registry is:
~/.gorombo/db/capabilities.sqlite
File-backed capabilities are materialized under:
~/.gorombo/capabilities/skills/<id>/
~/.gorombo/capabilities/tools/<id>/
~/.gorombo/capabilities/workers/<id>/
MCP definitions store their endpoint, transport, and token environment-variable name in SQLite. MCP tokens remain in the environment.
Capability records and managed files live outside the installed product artifact, so product upgrades preserve runtime additions.
Skills, tools, and workers accept:
The pre-release CLI accepts and stores --version, but materialization
shallow-clones the remote default branch. Reliable branch, tag, and commit
pinning remains a release gate. Local directory sources ignore the recorded
version.
Capability ids must be safe slugs and cannot collide with built-in or existing runtime capability names.
sim-one skill add <source> <id> "<name>" \
[--description "<text>"] [--version <requested-version>] [--enable]
Skills are enabled when added because they contain workflow knowledge rather than executable capability.
sim-one tool add <source> <id> "<name>" \
[--description "<text>"] [--version <requested-version>] [--enable]
Tools remain disabled unless explicitly enabled.
sim-one worker add <source> <id> "<name>" \
[--description "<text>"] [--version <requested-version>] [--enable]
Workers remain disabled unless explicitly enabled.
sim-one mcp add <id> "<name>" --url <url> \
[--transport <streamable-http|sse>] [--token-env <ENV_NAME>] \
[--description "<text>"] [--enable]
The URL must use HTTP or HTTPS. streamable-http is the default transport.
--token-env records the name of the secret-bearing environment variable.
Each capability family supports:
list
enable <id>
disable <id>
update <id>
remove <id>
Updating a skill, tool, or worker re-fetches its recorded source. Removing it deletes the registry record and managed files. In the pre-release CLI, MCP update changes only the record’s update timestamp. Change an MCP connection, name, or description by removing and re-adding it. MCP removal deletes the connection record.
Apply lifecycle changes by restarting the gateway through the process or service manager that launched it.
The agent can propose or add runtime capabilities through governed tools:
Registration does not grant unrestricted authority. Enabled capabilities remain subject to:
The release contract also subjects every capability path to active protocol scoring and orchestrator/critic enforcement. Complete activation of that boundary remains a release gate.
sim-one <skill|tool|worker|mcp> list
Restart the gateway, then open a new terminal session and confirm the capability is available to its owning agent.