qwen38-iq3s reasons extensively before emitting the skill body. A max_tokens
cap truncated the hidden reasoning (finish_reason: length) leaving content
empty, so the body write failed with 'skill body is empty'. Omit max_tokens
so the model runs to completion (~7 min); reasoning is filtered automatically
since only content is read. Client timeout default raised to 1200s.
create_skill now writes a real predict/act body: the small decision model
still authors title + description (engine.generate), then a larger
OpenAI-compatible model (default qwen38-iq3s) writes the runnable code
against the SKILL.md contract. Bodies persist to data/skills/<cat>/<name>.py,
are hot-loaded via importlib, merged into the running tree, and the request
re-dispatches to the new leaf. The dashboard decision-flow view shows the
title/description with a writing badge while the body is being written.
Codegen failure degrades to a navigable stub.
The model follows the repo's existing dotted naming (email.compose,
tracking.check) and proposed titles like tracking.status_lookup, which the
strict alnum-only check rejected. Accept dotted snake_case names.
An empty category produced a one-option SemIf leaf decision; the backend rejects
<2 options, the ValueError unwound past the current-process reset, and every
later request queued forever behind a phantom current. Navigation now
short-circuits empty trees/categories straight to the create branch (no SemIf
decision when there's nothing to choose), and the scheduler clears its current
process even when dispatch raises. Authoring generation is capped at 128 tokens.
Navigation now offers create_category at the category level and create_skill
at the leaf. Both are stubs that log a suggestion event (state, question,
SemIf output) to the trace instead of returning an authoring sentinel.
Denton Social