History

Nothing else. No plans, no wiki pages, no preferences.

CSV Layer

The CSV files are the source of truth the agent reads for coaching. JSON is the forensic source. YAML in programs/ is a derived view.

These CSVs are sorted in ascending chronological order:

• strength: by date, start_time, title, exercise_title, set_index
• running activities: by date
• running laps: by date, activity_id, lap_index

/history status is the canonical read path over this CSV layer. Use the CSVs directly before opening YAML or JSON.

Regenerate YAML from CSV after a manual CSV edit:

uv run health-coach-import-hevy --input history/strength/hevy-workouts.json
uv run health-coach-import-strava --input history/running/activities.json

Or directly:

from health_coach.history.strength.ingest import import_hevy_file
from health_coach.history.running.ingest import import_strava_file

Mental Commands

• /history sync hevy
• /history sync strava
• /history review nutrition
• /history age nutrition
• /history sync all
• /history capture weight — append a new entry to history/body/weight.yaml (user-reported weight, optional composition).
• /history status

Support Tools

All heavy lifting here requires code — OAuth, external APIs, binary format parsing.

CLI equivalents:

• health-coach-sync-hevy
• health-coach-import-hevy --input <path>
• health-coach-sync-strava
• health-coach-import-strava --input <path>
• health-coach-capture-weight --weight-kg <kg> [--date YYYY-MM-DD]
• health-coach-age-nutrition

Command: /history sync hevy

Fetches all Hevy workouts from the logged-in Firefox session.

Steps:

1. resolves the Firefox profile with cookies.sqlite
2. reads the auth2.0-token cookie for the Hevy API
3. fetches all workouts in paginated batches
4. writes raw JSON to history/strength/hevy-workouts.json
5. converts JSON → per-set CSV at history/strength/hevy-workouts.csv
6. derives denormalized per-set YAML at programs/strength/program.yaml
7. the next expected read is /history status, which inspects the fresh CSV rows

Notes:

• requires Firefox with an active Hevy session
• use health-coach-import-hevy --input <path> when syncing is not available

Command: /history sync strava

Fetches Strava run activities via OAuth.

Steps:

1. loads tokens from .env or data/running/strava_tokens.json
2. refreshes the access token if expired
3. fetches all run activities (or incremental with --incremental)
4. merges with existing raw payload
5. fetches laps for each run activity (idempotent — skips if history/running/laps/<id>.json already exists)
6. writes raw activities to history/running/activities.json
7. converts JSON + laps → per-activity CSV at history/running/activities.csv (with kind, laps_text)
8. writes history/running/laps.csv with one row per lap
9. derives YAML view at programs/running/program.yaml
10. the next expected read is /history status, which inspects the fresh CSV rows

Notes:

• requires Strava API credentials in .env
• run health-coach-sync-strava --setup to perform OAuth setup
• use health-coach-import-strava --input <path> for manual imports

Command: /history age nutrition

Ages the current nutrition week draft into the history layer.

Steps:

1. reads programs/nutrition/week-draft.yaml
2. extracts week_start_date
3. tags the record with provenance: aged_proxy and aged_at timestamp
4. writes to history/nutrition/<week_start_date>.yaml

Why:

Nutrition has no reliable consumption log. The week plan (programs/nutrition/) represents intended meals, not confirmed consumption. Aging it into history/nutrition/ preserves it as a proxy record that the planning skill can use for replanning. The provenance tag ensures downstream tools know it is intention, not measurement.

Command: /history review nutrition

Canonical week-closing gesture for nutrition.

Purpose:

• capture what really happened during the week
• preserve the final week as proxy history
• give the next nutrition build a grounded adherence signal

Steps:

1. read programs/nutrition/week-draft.yaml
2. ask briefly what was actually eaten, swapped, skipped, too heavy, too repetitive, or too long to cook
3. write the feedback to today's memory/YYYY-MM-DD.md
4. age the current nutrition week into history/nutrition/<week_start_date>.yaml
5. keep the aged record marked as proxy history, not confirmed consumption

Rules:

• keep it short
• never block the next planning step if the user declines
• if the user has little memory of the week, write that as the signal

Notes:

• this is the canonical command the planning skill should call before drafting the next week
• /history age nutrition remains available as the lower-level aging gesture

Command: /history capture weight

Appends a new body metric entry reported by the user (no external sync path exists yet — weight is manual).

Steps:

1. reads the reported value from the user
2. prefers the support tool health-coach-capture-weight for the append-only write
3. appends an entry to history/body/weight.yaml with date, weight_kg, source: user_input, optional note
4. if body composition was reported alongside, appends a snapshot to history/body/composition.yaml with method (e.g. impedance_scale, dexa)

Why:

Body weight is a primary input for strength (volume targets), running (load), and nutrition (TDEE calibration). Without a dedicated log it can only be inferred from scattered memory notes. The append-only file is the canonical factual timeline.

Notes:

• never overwrite prior entries — this is an append-only log
• source is user_input today; reserve scale_sync for a future integration and inferred only when explicitly derived

Command: /history sync all

Runs sync hevy + sync strava in sequence. Nutrition aging is a separate gesture because it requires a deliberate decision to close the week.

Command: /history status

Summarize the current state of the history layer:

Purpose:

• this is the canonical read protocol for history/
• it reads the CSV layer first, then summarizes freshness, recent facts, and gaps
• it should prefer a medium horizon (2-3 weeks) over a tiny snapshot, so irregular days do not dominate the read

Steps:

1. verify the expected files exist:
   • history/strength/hevy-workouts.csv
   • history/running/activities.csv
   • history/running/laps.csv
   • history/nutrition/ aged weeks if relevant
2. start with a date-based window

Default horizon:

• 14 days when the question is tactical
• 21 days when the goal is to infer pattern, rhythm, or adherence

3. filter the CSV layer by date first:

cutoff=$(date -v-14d +%F)
awk -F, -v cutoff="$cutoff" 'NR==1 || $1 >= cutoff' history/strength/hevy-workouts.csv
awk -F, -v cutoff="$cutoff" 'NR==1 || $2 >= cutoff' history/running/activities.csv
awk -F, -v cutoff="$cutoff" 'NR==1 || $2 >= cutoff' history/running/laps.csv

4. if the view is too noisy, move to 21 days or add a targeted filter:

cutoff=$(date -v-21d +%F)
awk -F, -v cutoff="$cutoff" 'NR==1 || $1 >= cutoff' history/strength/hevy-workouts.csv
awk -F, -v cutoff="$cutoff" 'NR==1 || $2 >= cutoff' history/running/activities.csv
awk -F, -v cutoff="$cutoff" 'NR==1 || $2 >= cutoff' history/running/laps.csv

5. if the question is targeted, use rg or a focused awk:

rg "bench_press_barbell" history/strength/hevy-workouts.csv
rg "interval_session" history/running/activities.csv
awk -F, '$2 == "2026-04-01"' history/running/laps.csv

6. open JSON only if the CSV layer is insufficient or contradictory
7. summarize:
   • last sync date for Hevy (history/strength/hevy-workouts.json)
   • last sync date for Strava (history/running/activities.json)
   • strength pattern across the last 2-3 weeks
   • running pattern across the last 2-3 weeks
   • whether a recent structured running session is visible in laps.csv
   • aged nutrition weeks in history/nutrition/
   • any obvious gaps or drift

Write Policy

This skill writes only to:

• history/strength/
• history/running/
• history/nutrition/
• history/body/
• programs/strength/program.yaml (as a side effect of import — this is the structured program derived from history)
• programs/running/program.yaml (same)

This skill never writes to:

• wiki/
• memory/
• programs/nutrition/ — only the planning skill modifies nutrition plans

Principles

1. History holds facts. Plans hold intentions. Keep them separate.
2. Sync is the only path from external reality into history/.
3. Aging is explicit — it requires a deliberate gesture to close a nutrition week.
4. The provenance tag on aged records is not optional — it is the semantic contract.
5. When in doubt, sync before replanning.