Blue-B/pisesh

Pisesh solves a clear usability problem for frequent pi coding-agent users. The keyboard-driven TUI adds favorites, search, project filtering, custom titles, working-directory overrides, session details, and model-aware resume options without adding runtime dependencies. The standalone CLI and /sesh extension share a sensible handoff design, with the extension using pi’s official session-switching API instead of launching a nested process. The implementation handles terminal cleanup, CJK display widths, custom session directories, stale favorites, cancelled title generation, and malformed model output. Sidecar JSON files preserve most metadata without touching session history. The README is detailed, includes real terminal captures, distinguishes the two installation methods, explains storage locations, and discloses that title generation sends session text to the selected model provider. The MIT license, changelog, issue templates, cross-platform workflow, npm provenance publishing, and focused Node tests add confidence. The current 17 tests pass and cover title generation, resume settings, custom paths, selection handoff, and interrupted tool-call repair.

The main weakness is that CI does not run those tests. It checks syntax, package metadata, and --help across Node 18, 20, and 22 on three operating systems, but never executes the node --test command used by npm test. That leaves the most sensitive behavior, including session-file repair, unverified on pull requests. This matters because pisesh can modify session JSONL files when repairing interrupted tool calls. It creates a one-time backup and tests idempotency, which is good, but atomic writes and broader corruption or permission-failure cases would reduce risk further. The 1,700-line bin/pisesh script also combines storage, scanning, rendering, input handling, model generation, and repair logic. Splitting these concerns into internal modules would make future changes easier to review and test. Adding npm test to CI, testing the TypeScript extension handoff, and covering real TUI input sequences would be the highest-value next steps.