workoutplan

Personal strength-training log and analytics. Run it in Claude Code; log set by set on the phone during the session.

• CLAUDE.md is the operating manual - parsing rules, the progression rule, the feedback contract, the hard rules. Read that, not this.
• log.csv is the append-only system of record (empty until the first session).
• routine.yaml holds one or more weekly plans and a schedule of which was active from when. Switching plans changes what is prescribed and which volume bands apply - and nothing else. Trends, the strength index, the bridge, stalls and balance read log.csv, which has no plan column, so a lift keeps its whole history across a switch.
• log.example.csv holds 12 weeks of generated history so the analytics produce output on day zero - real output of prescribe(), not hand-typed.

/start          open a session
bench 80x8 @2   log a set (exercise is sticky afterwards)
/end            validate, append to log.csv, 4-line summary
/review         volume vs target bands, 7d and 14d
/plan           next session, loads from the progression rule
/scan           drop a photo of a gym machine - resolves it against the library,
                or proposes a new entry and asks before adding it

python3 analyze.py validate --log log.example.csv
python3 analyze.py volume   --log log.example.csv --window 7
python3 analyze.py plans                       # the plans, and which is active
python3 analyze.py bands --plan gym80          # the bands a plan delivers, to paste

Requires Python 3.10+ and pyyaml.

web/ - prototype

A phone-first prototype: Today (the plan, prescribed loads, terse logging, the 3-line feedback contract), Trends (e1RM per lift, strength index, the volume-load bridge, stalls, balance, adherence), Plan (switch between weekly plans, edit one, export routine.yaml). Static HTML/CSS/JS, no build step, no backend.

web/data.js embeds the exercise library, the routine, the bands and a full run of analyze.py's analytics (web/analytics.json) - generated, never hand-edited. Every deep number in Trends is computed once by analyze.py and rendered as-is; nothing is recomputed in JavaScript. After editing exercises.yaml, config.yaml, routine.yaml or log.example.csv:

python3 web/build_data.py
python3 -m http.server 8000 -d web    # then open http://localhost:8000

An exercise can carry an optional reference photo (image: in exercises.yaml, a file under web/images/) - shown as a small thumbnail next to it in Today, click to expand. Purely cosmetic, never read by analyze.py. /scan attaches these automatically when it identifies a machine from a photo.

web/build_artifact.py bundles everything into one file (web/artifact.html, not tracked) for hosts that can't fetch sibling files, such as a Claude Artifact - any exercise photo is inlined as a base64 data URI in that build.

The prototype does not write log.csv. Finish prints the rows the CLI would append. Session state and any plan edits live in localStorage, per browser.