# Changelog

## 0.3.0 — 2026-11-11

GitHub release. npm publication is separate and remains pending.

- Support session format v3 only. Show unsupported sources in aggregate coverage and refuse mixed-format preparation rather than silently omitting evidence.
- Keep one complete investigation active per Pi session, with duplicate-start refusal, branch-bound approval, manual cancellation and explicit interrupted-runtime recovery.
- Add TUI and model-facing status, saved-run inspection, same-ID resume and exact-ID-confirmed deletion. A different analysis of an existing cut receives a new Run ID.
- Resume partial preparation from validated copied evidence, reading only missing originals after renewed approval. Preserve saved task identities, coverage and valid results.
- Record coordinator-reviewed output digests and ordered synthesis dependencies. Permit explicitly reviewed stale synthesis to use newly grouped inputs; require validated coverage and finished-worker reports before completion.
- Protect shared cuts and checkpoints with bounded interprocess transactions, and move cutting into owned, interruptible Node workers.
- Bound tool inputs and responses, make optional preference-read failures nonfatal, and require complete paginated evidence reads with explicit context limits.
- Exclude internal alignment plans and reviews from the npm package.

## 0.2.0 — 2026-10-22

Initial public GitHub release. npm publication is separate.

- Start the same guided workflow through a natural-language request or `/miner`, using one shared entry instruction source and Pi's native tool selection.
- Guide source selection with explicit workdir, project, timeframe, analysis goal and coverage confirmation.
- Prepare ordered, bounded source packets and self-contained task instructions from historical Pi sessions.
- Run Miner operations sequentially within each Pi session through Pi's native tool execution mode; separate sessions remain independent.
- Inventory session headers in bounded batches through the existing technical workers, one session at a time per thread, and report resource failures instead of silently omitting sources.
- Support assistant-coordinated analysis, separate synthesis, saved-run browsing and reuse without duplicating source snippets.
- Use the existing paginated workdir browser for all run listings and searches; remove the redundant unbounded run-list tool.
- Remember the working directory after successful cut reuse as well as fresh preparation; report preference-write failures without invalidating a ready run.
- Report incomplete run folders that block saved-run listing and deletion, and ask for a user-approved delete or move through Pi's existing tools.
- Keep result files separate from evidence and distinguish technical file completeness from semantic correctness.
- Add public entry documentation, security guidance, MIT licensing, logo metadata and a thin Pi entrypoint.