StateLog

• Every entry the user states lands in the log in one tool call and gets one short acknowledgment, with no follow-up questions or analysis.
• Ratings come only from the user's own words; inferred events reset freshness counters and never write a rating.
• A status read shows each dimension's latest value and age, and names what is stale or never logged in one line.
• The journal stays the user's: it lives under LIFEOS/USER/STATE/ and never enters this skill or any public artifact.

<Letter><Digit 1-5> plus optional note (M2 slept badly). Bare W or K resets that freshness counter.

Scale (star-rating semantics, higher is better): 1 really bad · 2 bad · 3 fine (shown yellow, never treated as a crisis) · 4 and 5 the preferred range (shown green). The parser clamps 0/6-9 into range and preserves the raw value. Every dimension reads the same direction — there are no inverted dimensions.

What each dimension means to the user, where the creativity pair's framing comes from, the affirmation practice file, and the workout signal source: <configRoot>/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/StateLog/Config.md (configRoot is $CLAUDE_CONFIG_DIR when set, else ~/.claude). First-time setup is in Setup.md. Read it when discussing the creativity pair, logging an A, or deciding whether an event counts as a workout signal. Logging itself works without it; when Config.md is missing, say so once, point the user at Setup.md, skip the streak bump and signal resets, and still log the entry. The letters and 1-5 scale are the tool's defaults in State.ts; Config.md only records what they mean to the user.

The deterministic engine is ~/.claude/LIFEOS/TOOLS/State.ts (cross-consumer: statusline, interview, this skill):

bun ~/.claude/LIFEOS/TOOLS/State.ts log "M4"              # rate
bun ~/.claude/LIFEOS/TOOLS/State.ts log "M2 rough day"    # rate + note
bun ~/.claude/LIFEOS/TOOLS/State.ts log "W"               # freshness reset
bun ~/.claude/LIFEOS/TOOLS/State.ts log "H5" --source interview
bun ~/.claude/LIFEOS/TOOLS/State.ts reset W               # signal-driven reset
bun ~/.claude/LIFEOS/TOOLS/State.ts current [--json]      # snapshot with freshness tiers
bun ~/.claude/LIFEOS/TOOLS/State.ts statusline            # compact segment
bun ~/.claude/LIFEOS/TOOLS/State.ts rebuild               # re-derive current.json from log

Data: LIFEOS/USER/STATE/log.jsonl (append-only) + current.json (derived). The statusline reads current.json directly with jq — never spawn the tool from render paths.

• A bare rating letter with no digit (M) is an error, not a default — only W/K/A accept bare-letter resets.
• Bare A = "did today's affirmation reps". Config.md names the practice file (Affirmation practice file); on an A entry, also bump the streak line in that file's Active section. With no practice file configured, log the A and touch nothing else.
• --source records where an entry came from, for later correlation: manual (default), interview, signal. reset W|K stamps signal itself, so an entry inferred from an event ("just worked out") goes through reset (Intents/Log.md), never through log with a hand-passed source.
• Freshness is per-dimension, three tiers (fresh/aging/stale) with windows declared in State.ts DIMENSIONS: mood and energy decay in hours, relationships and financial in days. State.ts current computes each tier; a stale dimension is the prompt to log, not an error.
• Entries are case-insensitive, and a message may carry several at once (e4 cf3 = energy 4, creative freedom 3). A message that is nothing but shorthand routes here silently.
• Don't route goal-percentage talk here: "health 68%" is a life-goal gap score, not a journal entry.

User: "e4 cf3"
→ Log intent
→ Energy 4 and creative freedom 3 are in the log; the reply is one acknowledgment line

User: "just worked out"
→ Log intent
→ The workout counter resets with source signal, and no rating is written

User: "how am I tracking"
→ Status intent
→ A snapshot of each dimension's value and age, closing on one line that names what is stale