Japanese reports cover daily work through 90-day trends
Nippo reads Claude Code and Codex histories, normalizes sessions in Rust, and asks the installed skill to write a Japanese report. The default daily view lists work, decisions, terminology, messages, and tool use. Other modes produce a short brief, reflection questions, a plan, a learning guide, a 7-day progress report, a 90-day self-review, or longer trend analysis. The repository does not include an English README or parallel English templates.
The design draws a useful line around automation. Rust collects dates, projects, session counts, messages, tools, and decision candidates. The skill handles prose and uses the templates in docs/templates. Reflection mode writes questions with blank answer spaces instead of composing answers for the user. Nippo therefore saves the clerical work of reconstructing a day while leaving the uncomfortable part, explaining why a choice was made and what should change, to the developer.
A 0.3 MB checkout passed all 106 tests
Our fresh Rust container installed 55 packages in 16 seconds. The repository contained 33 files, about 5,104 source lines, and occupied 0.3 MB before dependencies. The build succeeded in 44 seconds. That is a pleasantly small codebase compared with the agent histories it parses. Our scan found 1 CI workflow file, no Dockerfile, and no top-level tests directory. Rust tests can live beside source, so the missing directory is not evidence of missing tests.
Cargo test completed in 25 seconds with 106 passed and 0 failed out of 106. Those results cover commit b23b807 on 3 CPUs and 12 GB of RAM in an unprivileged container. We did not feed the collector a copy of a real personal Claude Code or Codex history, ask an agent to write a report, or judge whether its prose matched a working day. The measurement establishes install, compilation, and test health rather than summary quality.
What happened when we ran it
Our whole install, build, and test sequence took 85 seconds. Every measured step succeeded, and the test command returned a clean result. Nippo had no dependency-audit number in the supplied lab block, so we make no vulnerability claim. It also had no Dockerfile, which is unsurprising for a local Rust command installed through Cargo. Users should still review which session files it reads because those histories can contain prompts, tool calls, project names, and operational details.
The CLI can emit JSON or a compact text summary without calling an agent. It merges records sharing one session ID, removes duplicate prompts, and excludes the currently active session when the host provides an exact ID. Default filtering drops skill expansion, harness notices, image placeholders, interruption messages, compact introductions, and short acknowledgements. Flags can restore that material for debugging. These filters reduce noise, though they also define what the eventual report can see.
Focus time can become a 927-minute false block
Issue 23 shows a collector error in work-window statistics. A Claude Code session left open from early morning to evening can be treated as one continuous interval, even when the user was idle for hours. In the reporter's example, 29 sessions across 23 projects became 1 focus block lasting 927 minutes. The proposed fix divides focus blocks from individual message timestamps while preserving the existing project-switch calculation.
That defect does not erase the session list or written report. It does make the numeric focus result unsuitable for time sheets, billing, or productivity scoring. Nippo's own philosophy argues for restraint here: session timestamps show when records were written, not continuous human attention. Until the event-based correction lands and is tested, omit the longest-focus figure from reports shown to managers or clients. Use it only as a prompt to inspect the underlying day.
Windows v0.1.6 needs HOME even when USERPROFILE exists
Issue 20 reports that nippo v0.1.6 reads only the HOME environment variable to locate skill and session directories. Git Bash sets it, while native PowerShell and cmd often rely on USERPROFILE. On the reported Windows 11 setup, skill installation failed and collection looked under the filesystem root. Manually setting HOME to USERPROFILE worked around the problem. That is a first-run failure for a supported Codex or Claude user in a common shell.
Installation also has a versioning wrinkle. cargo install nippo updates the binary, then nippo skill install writes embedded skill and template files. A later binary upgrade does not refresh those files automatically; users must rerun the skill installer with --force. A repository checkout can instead install symbolic links so updates appear immediately. Windows packaging requires care because checkout symlinks may become ordinary files when the environment does not preserve them.
Version 0.1.6 is active, narrow, and still settling
Release v0.1.6 shipped on August 24, 2026. It added prompt-noise filtering, split-session merging, active-session exclusion, deterministic local-time helpers, and period metadata. GitHub showed 134 stars, 5 combined open issues and pull requests, and a last push on August 24. Issue 24 asks the skills to accept yesterday and other period words already supported by the CLI, while issue 22 proposes OpenCode as another source.
Nippo is easiest to recommend when its limits match the user: Japanese output, local Claude Code or Codex history, personal reflection, and reports that stay in the project directory. Our 106-test pass makes that path inexpensive to try. The small community and v0.1.6 label call for cautious expectations. Check the collected JSON, keep private reports out of version control, and treat automated statistics as clues about work rather than proof of effort.