CodeStrain CLI

Your AI coding recovery score, from the terminal.

What is this

CodeStrain parses the Claude Code JSONL session logs already on your disk (~/.claude/projects/) and prints cost, token usage, a Developer Recovery Score (DRS) estimate, and a per-project breakdown. Zero dependencies — Python stdlib only. Read-only — your JSONL never leaves the machine.

Install

The CLI is a single command. Pick the install path your machine prefers — all three put codestrain on your $PATH:

# macOS — recommended (Homebrew)
brew install codestrain/tap/codestrain
codestrain --help

# any OS — pipx, isolated venv + symlink on PATH
pipx install codestrain
codestrain --help

# any OS — uv tool, fastest
uv tool install codestrain
codestrain --help

# one-liner installer — detects pipx/uv/pip and adds PATH automatically
curl -fsSL codestrain.dev/install | sh

pip install codestrain works too, but leaves the binary off your $PATH on macOS (PEP 668 / user-install convention). See troubleshooting below.

That's the macOS / PEP 668 quirk — pip installed the wheel but didn't put the entry-point on PATH. Three fixes, in order of preference:

# (a) run it as a module — works without any PATH change
python3 -m codestrain_cli --all

# (b) switch to pipx (installs in isolated venv, fixes PATH)
brew install pipx        # or:  python3 -m pip install --user pipx
pipx ensurepath          # adds ~/.local/bin to PATH
pipx install codestrain

# (c) find where pip put the binary and add that dir to PATH
pip show -f codestrain | grep -E 'bin/codestrain'
# typical macOS user-install path:
echo 'export PATH="$HOME/Library/Python/3.11/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

Quick start

# today's stats (default)
codestrain

# all-time, every session ever logged
codestrain --all

# all-time, with project names hashed
codestrain --all --anonymize

Example output

   ______          __     _____ __             
  / ____/___  ____/ /__  / ___// /__________ _( )___
 / /   / __ \/ __  / _ \ \__ \/ __/ ___/ __ `/ / __ \
/ /___/ /_/ / /_/ /  __/___/ / /_/ /  / /_/ / / / / /
\____/\____/\__._/\___//____/\__/_/   \__._/_/_/ /_/

  Your AI coding recovery score.

--- All Time ------------------------------------------

  Sessions:  1454
  Duration:  137h 21m  (span 15352h 27m)
  Turns:     61007
  Tokens:    2.0M in / 25.4M out
  Cost:      $21948.61
  Models:    claude-haiku-4-5, claude-opus-4-5, claude-opus-4-7 +5 more

  DRS Estimate  (avg per active day · 52 days · 2.6h/day)
  Strain:    9.0/21
  Recovery:  82%
  Readiness: GREEN — Recovered. Good to go.

--- Per-Project Breakdown -----------------------------

  project-1   31h  2m  13638 turns  $7193.92
  project-2   21h 40m   8684 turns  $3652.80
  project-3   15h 32m   4789 turns  $1212.63
  ...

Flags reference

DRS — what it actually measures

Strain (0-21, per active day). The CLI sums the gaps between consecutive turns that are ≤ 5 minutes — that's the "active coding" duration. Each hour contributes 2.1 strain points, capped at 21. The 5-minute threshold matches the ccusage / Claude Code Usage Monitor convention and is configurable via CODESTRAIN_GAP_MIN. Debug-heavy sessions (high error ratio), late-night work (after 22:00), and weekend coding add small penalties.

Recovery (0-100%). Recovery moves inversely to strain and is modulated by hours since the last session (sleep proxy). Eight hours off lifts the baseline; high recent strain pulls it down. The local heuristic doesn't have biometric input — it's purely behavioral.

Readiness. A traffic-light derived from recovery: GREEN at ≥ 67%, YELLOW between 34% and 66%, RED below 34%. The thresholds match the macOS app and the WHOOP-inspired DRS spec.

This is a heuristic estimate from JSONL logs, not medical advice. The full CodeStrain app refines DRS with ML models, wearable data (HealthKit / WHOOP / Oura), and per-user calibration.

Why this is privacy-first

• All parsing runs locally. No data ever leaves your machine.
• No telemetry, no opt-in pings, no usage analytics — not even crash reports.
• Your JSONL files are read-only. They are never uploaded, copied, or modified.
• Respects NO_COLOR and FORCE_COLOR / CLICOLOR_FORCE conventions for piping and CI.

Related projects

• ccusage — the npm reference for parsing Claude Code JSONL. Friend, not foe. We follow its session model so numbers line up.
• Claude-Code-Usage-Monitor — Python alternative with ML burn-rate prediction and a live dashboard.

Roadmap (v0.1)

• CreatureView — a tiny macOS menu-bar companion that surfaces DRS without opening a terminal (private beta).
• Souls Studio — paid persona pack and custom-character marketplace (Drill Sergeant, Gentle Princess, Sarcastic AI...).
• Magenta-key sprite pipeline v1.2 — clean alpha extraction for community-created creatures.
• Wearable integration — Apple HealthKit, WHOOP, Oura → unified HealthSnapshot.

More at codestrain.dev.

Contributing

PRs welcome. Sign your commits with git commit -s (DCO) and run the suite before opening one:

python -m pytest tests/
tests/smoke.sh

See CONTRIBUTING.md for the contribution workflow and TESTING.md for the full test matrix.

License

CodeStrain CLI is MIT-licensed and free for everyone — individuals, teams, companies, and forks — forever for this and every prior release. The CodeStrain hosted service (DRS predictions, ML models, encrypted sync) is a separate paid product; the CLI works fully offline without it.

If we ever introduce a commercial license for a future major version, we will give at least 90 days' notice, keep individuals and small organizations free, and never apply new terms retroactively. See CONTRIBUTING.md for the maintainer's relicensing posture and the DCO sign-off contributors use.

Copyright (c) 2026 LLP HubLab (codestrain.dev).

Star this repo if codestrain told you something you didn't know about your last week of AI coding. → codestrain.dev