Iiwi

Iiwi · /ˈiː.wiː/ "ee-wee" — turns your coding-agent sessions into a weekly engineering report, without anything leaving your machine.

English | 繁體中文

The weekly status update is work you already did once. Iiwi reads the sessions your coding agent already recorded, synthesizes evidence-backed outcomes, and lets you review the result before it writes the report. Quick Review targets a 30–60 second pass from candidate outcomes to a shareable update. Works with OpenCode, Claude Code, and Codex.

• Nothing leaves your machine. Your local opencode run writes the narrative — no network, no API key.
• Finds your sessions anywhere. Every project, whichever folder you run from.
• Groups by repository. Worktrees collapse into one entry; child and subagent sessions stay with it.
• Redacts first. Common secret patterns are stripped locally before anything is written.

Quick start

Needs Python 3.11+ and git. Plus one transcript source: OpenCode (the default, needs an opencode executable) or a readable Claude Code / Codex transcript store at ~/.claude/projects or ~/.codex (their CLIs are not needed for reading). Quick Review outcome synthesis still uses your local opencode run for every transcript source; if it is unavailable, the recovery screen offers Retry and an explicit session-based report fallback.

pipx install iiwi                 # or: pip install iiwi

iiwi doctor                       # is the harness ready?
iiwi report --period last-week    # write the report

The report lands under reports/. Pass --dry-run to print the report to the terminal instead of writing a file.

The interactive menu

Run iiwi with no arguments if you would rather not remember flags:

$ iiwi
 ___ _        _
|_ _(_)_ __ _(_)
 | || \ V  V / |
|___|_|\_/\_/|_|                                v0.9.1
══════════════════════════════════════════════════════
Turn coding-agent sessions into engineering reports
github.com/mike840609/iiwi

▶ Generate Report
  Browse Sessions
  Check Setup
  Settings

↑↓ jk │ Enter Select │ 1-4 │ ? Help │ q Quit

Choosing Generate a report shows every setting at once instead of asking one question at a time — ↑↓ moves, ←→ changes a value, and a line under the list explains the setting you are on. Press g to generate, or r to open Review sessions first, which weighs each repository by how much of the period it actually accounts for:

Review Sessions   6 / 6 selected │ 252 / 252 msgs
══════════════════════════════════════════════════════
Select sessions to include in the report:

  1. ▾ ████████████  71% ● iiwi   3 / 3    Aug 5 │ 180 msgs
▶      ████░░░░░░░░  24% ● Add the interactive menu Aug 5 │ 60 msgs
       ████░░░░░░░░  24% ● Redact before writing    Aug 5 │ 60 msgs
       ████░░░░░░░░  24% ● Group worktrees          Aug 5 │ 60 msgs
  2. ▸ ████░░░░░░░░  24% ● obsidian-wiki   2 / 2    Aug 4 │ 60 msgs
  3. ▸ █░░░░░░░░░░░   5% ● dotfiles   1 / 1         Aug 3 │ 12 msgs

↑↓ jk │ ←→ hl │ Space Toggle │ p Preview │ e Exclude │ a All │ g Generate │ / Search │ ? Help │ b Back

Space toggles a repository or a single session, p previews a transcript (redacted), and e excludes a repository from every future scan. Your selection is remembered per period. Press g to synthesize the selected sessions into Quick Review instead of writing the session-based report immediately.

Quick Review keeps the evidence traceable while you decide what belongs in the update. The first five candidates are selected; additional work stays under More candidates, and extraction failures stay visible under Ungrouped candidates. User-added outcomes, optional Blockers, and optional Next week text are included in the same in-memory draft.

Space Include/exclude │ e Edit │ J/K Reorder │ v Evidence │ s Split │ a Add
p Preview │ g Generate │ b Back

p Preview renders the current draft without writing a file. g Generate writes that reviewed draft. If synthesis fails, retry it or explicitly use the labeled session-based report fallback. See the evidence-first Quick Review guide for the full flow and the Manager/Engineering and Brief/Full responsibilities.

Commands

iiwi doctor                       # is the harness ready?
iiwi scan --period last-week      # preview how sessions group
iiwi report --period last-week    # write the report
iiwi history                      # reports already written
iiwi update                       # check PyPI for a newer release
iiwi run                          # the same questions, one at a time

iiwi --help lists everything, and run is the linear wizard if you prefer being asked. In scripts, name a subcommand directly — with no terminal to prompt at, the menu exits with status 3 rather than reading stdin.

Configuration

Settings come from an environment variable first, then the settings file, then the default:

iiwi config init                                          # walk through all
iiwi config set harnesses.opencode.cli.model deepseek-r1  # write one
iiwi config list                                          # show all, with sources
iiwi config unset report.timezone                         # back to the default

Every setting and its environment-variable name is in the configuration guide.

Privacy

Everything happens on your machine: sessions are read from disk, common secret patterns are redacted, and the redacted transcript is handed to your locally installed opencode run. No API key, and update is the only command that touches the network.

Reports may still contain private goals, filenames, commands, and full working paths — always review one before sharing it. See Privacy and security for the full data flow and current limits.

Documentation

The name

The ʻiʻiwi is a scarlet Hawaiian honeycreeper whose long curved bill reaches nectar others cannot — which is what this tool does with the sessions your agent left behind. Pronounced the anglicised way, "ee-wee".

Development

git clone https://github.com/mike840609/iiwi.git
cd iiwi
uv sync --locked --extra dev                    # install the project and its dev tools

uv run iiwi                                     # run your working copy
uv run pytest --cov=iiwi --cov-fail-under=80    # tests, with the coverage gate CI enforces
uv run ruff check .                             # lint
uv run pyright                                  # type check

The last three are exactly what CI runs, so a green local run means a green PR.

Name forms

The name is cased by where it appears:

License

MIT