No trailing indentation in prompts; preserve fenced body whitespace and Python helpers. Accept semantic HTML attribute/text equivalents; browser behavior and parser syntax are reported independently. Add diagnostic hidden-test pass rate; primary all-tests-required macro score unchanged. Fixed 350 data/test/reference files unchanged. Live browser validation still outstanding.
Use base-model code completion for SmolScript 0.4.0
Task comments and fixed headers/scaffolds; deterministic prefix assembly and first-function scope extraction without semantic repair. Same 350 tasks/tests and 96–256 token budgets. All 350 reference bodies pass static roundtrip checks. Full runtime validation and model score improvements remain unverified.
Fix CSS injection with disabled page scripting and browser workspace setup
Use synchronous style DOM insertion; add runtime computed-style preflight; report exact setup stage; use writable private HOME/XDG/temp paths and subprocess UID/GID rather than Python preexec hooks. Separate read-only requests from writable Docker workspaces. Fixed 350-task dataset unchanged. Live Modal verification remains outstanding.
Bound browser candidate hangs with a phase-aware watchdog
210 easy, 105 medium, 35 moderate tasks across nine languages. Preserve the previous 1000 tasks as previous1000 and original edition as full2000. Fast decoder settings unchanged. Full reference execution remains unverified.
Make fixed 1000-task fast edition the default (0.2.0)
Batch 64 with OOM backoff, length/budget buckets, cached decoding and no per-token CPU streamer. Versioned 96–256 output budgets. Preserve full2000 as an archived config. Full sandbox reference execution and GPU throughput remain unverified.
Allow cached model resolution during offline evaluation
2000 generated examples; exact language/difficulty allocation, isolated runners and Hugging Face evaluation. Static audit passes. Full reference and negative-control execution remains unverified because a Docker/Podman sandbox is unavailable.
