release: 0.2.1

[OneNoted]

Summary

• unify public rewrite mode onto a single rewrite mode with backward-compatible config aliases
• move cue-led openings like my apologies, I meant ... to LLM adjudication instead of deterministic replacement
• remove the extra outer meter shell from the OSD meter and keep the inner pill
• remove stale public live-mode docs for the not-yet-shipped whispers voice command
• bump the crate to 0.2.1

Release Notes

Added

• structured rewrite cue context for utterance-initial and courtesy-prefixed correction phrases
• compatibility loading for deprecated rewrite mode names and legacy [agentic_rewrite] config fields

Changed

• canonical postprocess rewrite mode is now rewrite
• rewrite policy, glossary, and default correction policy now live under [rewrite]
• rewrite mode now always sends correction/session evidence to the rewrite model instead of short-circuiting deterministic replacement text
• rewrite prompts now treat strong edit cues as evidence rather than an automatic winner in utterance-initial courtesy cases
• setup now presents a single public rewrite mode

Fixed

• utterance-initial phrases like my apologies, I meant ... no longer get over-filtered by deterministic correction handling
• meter OSD no longer renders the extra outer pill background around the inner indicator
• README no longer documents the unsupported whispers voice command

Verification

• cargo test --quiet
• cargo publish --dry-run --locked
• published to crates.io as 0.2.1

Notes

• This PR supersedes the stale release-0.2.0 PR.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4ff9896bfd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve rewrite policy settings when selecting a model

Avoid resetting rewrite.policy_path, rewrite.glossary_path, and rewrite.default_correction_policy in update_config_rewrite_selection. After this release, those fields are part of the active unified rewrite mode, so running whispers rewrite-model select ... now silently overwrites user custom policy/glossary locations and correction policy (e.g., conservative/aggressive) just to change models, which can disable expected rewrite behavior.

Useful? React with 👍 / 👎.

[OneNoted]

Fixed in a follow-up commit on this PR branch. update_config_rewrite_selection() now preserves rewrite.policy_path, rewrite.glossary_path, and rewrite.default_correction_policy, and I added a regression test covering custom policy/glossary locations plus a non-default correction policy.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e4e437967f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Restore deterministic replacement for follow-up alias cues

Removing the deterministic replacement from SessionBacktrackPlan drops the only path that converted explicit follow-up aliases into clean replacement text (for example srajvat, hi / scratchfarts, hi, which cleanup::explicit_followup_replacement still recognizes). build_backtrack_plan can still recommend ReplaceLastEntry for these utterances, but without the precomputed replacement it falls back to correction_aware_text/raw text, so when rewrite is unavailable (or falls back) the prior entry is replaced with the literal cue phrase instead of the intended tail like hi.

Useful? React with 👍 / 👎.

[OneNoted]

Fixed in a follow-up commit on this PR branch. Explicit follow-up aliases now feed the session replace candidate text again, and the rewrite fallback path reuses that deterministic tail when a replace-last-entry session action is active. I also added regression tests covering the alias fallback path and the rewrite-plan fallback text.