mirror of
https://github.com/nearai/ironclaw.git
synced 2026-09-03 08:06:01 +08:00
11f006987d
* fix(engine): make FINAL/FINAL_VAR awaitable in CodeAct scripts LLMs frequently emit `await FINAL(answer)` by analogy with async tool calls. The handler returned `MontyObject::None` synchronously, so the subsequent `await` raised `TypeError: 'NoneType' object can't be awaited` and the answer was dropped on the floor. Route FINAL and FINAL_VAR through the existing `pending_futures` mechanism with a trivially-resolving task. `final_answer` is set the moment the call arrives, so both `FINAL(x)` (coroutine discarded) and `await FINAL(x)` (resolves to None) succeed. Two regression tests in `scripting::tests` pin both forms. * fix(engine): normalize typographic punctuation before skill activation iOS, macOS, and most rich-text inputs autocorrect `I'm` (ASCII U+0027) to `I'm` (curly U+2019). Skill activation patterns like `(?i)I'm a (CEO|...)` are authored with ASCII punctuation, so the curly form silently failed to match — `ceo-setup` and any other skill with apostrophes in its regex never activated for autocorrected input. Add `normalize_punctuation()` in `default.py` that folds 8 typographic quote variants and the en/em dashes to ASCII before scoring runs. Implemented with chained `.replace()` because Monty does not expose `str.maketrans`/`.translate()`. Runs once per turn on the goal text; user content is untouched everywhere else. Two regression tests under `executor::orchestrator::tests` drive `normalize_punctuation` and `select_skills` end-to-end with the exact ceo-setup pattern and a curly-apostrophe goal. * docs(codeact): note closure capture quirk and Rust regex limits Two recurring CodeAct failures got their own paragraphs in the preamble: - Function closures defined in one ```repl``` block do not reliably capture names (modules, top-level vars) defined in earlier blocks, producing spurious `NameError`. Authors must keep the helper, its imports, and its call sites in the same block. - The embedded `re` is the Rust `regex` crate, not CPython: positional- only flag args, no lookaround, no backreferences. Lead with string methods (`in`, `startswith`, `splitlines`) and only reach for `re` when truly needed. Pure docs change. * fix(engine): adapt FINAL-await tests to staging CodeExecutionResult API Replace `had_error` field references with `failure.is_none()` to match the staging struct where `had_error: bool` was replaced by `failure: Option<CodeExecutionFailure>`. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(engine): address gemini-code-assist review — normalize before extraction, seed explicit skills (#2531) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(skills): remove useless .into_iter() to satisfy clippy (#2531) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(clippy): collapse nested if blocks into match guards (#2531) Fixes collapsible_match lints in ironclaw_tui::render and ironclaw_engine::runtime::mission triggered by Rust 1.95. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(clippy): resolve collapsible_match and unnecessary_sort_by lints (#2531) Fix 5 collapsible_match lints (responses_api, cli/tool, rig_adapter, setup/prompts) and 2 unnecessary_sort_by lints (glob_tool, grep_tool) triggered by Rust 1.95 clippy. Also fix rustfmt formatting for match guards in render.rs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix(clippy): group wasm startup channel registration context --------- Co-authored-by: Illia Polosukhin <ilblackdragon@gmail.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>