Releasesv0.14.0
v0.14.0 — integration harmony: skills wired into the loop
Dies ist die Release Note so, wie sie veröffentlicht wurde, keine Neufassung davon. Release Notes werden in der Sprache veröffentlicht, in der sie geschrieben wurden.
Integration-harmony pass. A full-repo audit (every tool, skill, CLI command, backend, subsystem checked for wiring) found the system largely coherent — pluggable backends conform to one contract, the hierarchy orchestrator is wired classify→decompose→verify→synthesize, the evolution loop's governance gate really fires, memory is read+written with a taint gate, all modules import — but surfaced two real disconnects, now fixed.
Fixed
- Built-in skills now actually reach the agent loop. They registered and listed via `chimera skills`, but nothing at runtime consulted them (`Agent` was tool-only; the retrieval helpers had zero callers). `Agent` now surfaces the most task-relevant built-in skills into the system prompt, across every generic path — so `data_analysis`/`data_visualization` are reachable, not dead weight.
- Skill retrieval no longer matches on stopwords (which surfaced every skill for any task). Content-word matching only: a chart task surfaces `data_visualization`; a geography question surfaces nothing.
- `download_media` is now a governed FETCH tool — it pulls untrusted internet content but was in no governance set, so a downloaded file wasn't tainted. Added to `FETCH_TOOLS`.
Gate: ruff clean, 1227 passed / 2 skipped, mypy clean.
`pip install chimera-agent==0.14.0`