Repository navigation
SparseRead v0.1.4
November 14 update
Download the updated source installer ZIP
and checksums.
The updated installation guide
covers workspace and per-user installations. PyPI registration is pending.
- OpenCode V2 adds per-user installation via
--platform opencode --global. Fresh
global installs start in advisory mode and preserve native reads. Once 48 hours
have elapsed since the first successful SparseRead evidence read, the next such
read shows a one-time suggestion in the terminal to consider auto mode. - Reinstalling preserves the selected mode. To switch, run the installer with
--platform opencode --global --sparseread-mode autoand restart OpenCode.
BUILD_COMMIT-20261114.txt identifies the source of the updated attachments.
The original v0.1.4 tag and its attachments remain available.
Original release
- Core updates text request and source identity, workspace-relative paths, and bounded source-manifest checks.
- Scoped macOS arm64 host checks cover OpenCode 2.3.0 (plus V1 1.17.14/1.19.5), Claude Code 2.1.325, Codex CLI 0.169.0, and official NanoBot 0.4.1. NanoBot requires a stock
read_fileselection each persistent turn before direct SparseRead source access; shell operations remain native. - Pi 0.87.1 and OpenClaw 2026.7.1-2 passed fixed-version macOS source archive regressions. These checks do not claim latest-version upgrades.
- Limited synthetic DeepSeek workflows checked functional activation; Claude and Codex used test-only wire adapters. They are not performance benchmarks. The Claude model run preceded narrowly scoped final changes that were separately tested.
The fixed integration commit 05768c8 passed 323 Python tests, five OpenCode Node tests, 19 Pi Node tests and artifact checks. See the validation matrix and independent report for evidence, exact version and platform boundaries. BUILD_COMMIT.txt and SHA256SUMS identify each delivered artifact.
Windows, Desktop, Cloud, Web and remote support or newer-host performance are not established by these macOS CLI tests. Historical paper experiments retain their original scope.