Expand description
The resource flags a session worker is started with (SessionWorkerOptions of experimental/session-worker.ts,
from the command line through native-command.ts and the server’s model options):
--system-prompt <text>: the system prompt is exactly this text (currentSystemPromptreturns it unrendered);--append-system-prompt <text or file>(repeatable): appended to the rendered prompt in place of a discovered APPEND_SYSTEM.md. The CLI reads a file input in its own directory and sends the text (the worker’s cwd may differ). The TS server never passed it to a worker; ultron-rs does, asappendSystemPrompt;--no-skills: no skills in the prompt, theskills.*catalog or the command list (bundled ones included);--no-context-files: no AGENTS.md / CLAUDE.md in the prompt;--skill <path>(repeatable,skillPaths; absolute, the CLI resolves them in its own directory): skills loaded after the discovered ones (resource-loader.tsadditionalSkillPaths), the first skill of a name kept.
The tool flags (--tools, --no-tools, …) are not here: the model’s tool list is exactly [rlm]. Neither are
-e/--extension and --no-extensions: Pi extensions are not supported by ultron-rs (the CLI warns and ignores
them), and a worker ignores extensionPaths and promptPaths options.
Structs§
Functions§
- filter_
commands - The
/commands without the skill commands (skill:<name>) when--no-skillsleft the skills out.