nutrition-alignment

name	nutrition-alignment
description	Produce a bounded NutritionProposal for today by consuming the runtime-computed `classified_state` + `policy_result` and applying judgment-only steps — action matrix, rationale prose, honest handling of micronutrient unavailability. The runtime already did every band, every score, and every R-rule; this skill does not re-derive them. Macros-only v1 per the Phase 2.5 retrieval-gate outcome.
allowed-tools	Read, Bash(hai state snapshot *), Bash(hai state read *), Bash(hai propose *), Bash(hai intent list *), Bash(hai target list *)
disable-model-invocation	false

Nutrition Alignment

All arithmetic happens in code. Your job is: read the bundle, honour the policy result, pick an action, write the rationale, persist.

Load the bundle

hai state snapshot --as-of <today> --user-id <u> --evidence-json <hai clean output>

Under snapshot.nutrition you receive these blocks:

• today — today's accepted_nutrition_state_daily row (calories, protein_g, carbs_g, fat_g, hydration_l, meals_count, derivation_path), or null.
• history — trailing rows for lookback context.
• signals — runtime-derived dict the classifier consumed: today_row, goal_domain, plus (v0.1.15 W-A) optional is_partial_day (bool) and target_status ("present" | "absent" | "unavailable"). Context only; never re-derive.
• classified_state — calorie_balance_band, protein_sufficiency_band, hydration_band, micronutrient_coverage, coverage_band, nutrition_status, nutrition_score, calorie_deficit_kcal, protein_ratio, hydration_ratio, derivation_path, uncertainty. Source of truth.
• policy_result — policy_decisions[], forced_action, forced_action_detail, capped_confidence, and (v0.2.0 W-PROV-2) optional evidence_locators[]. The runtime always emits a row-level locator citing today's accepted_nutrition_state_daily; an R-extreme-deficiency firing additionally cites calories + protein_g. Partial-day suppression keeps the row-level locator intact and skips the column-level citations. Source of truth.
• missingness — per state_model_v1.md §5.

Protocol

1. If the policy forced an action, use it

If policy_result.forced_action is set, action is that value and action_detail is policy_result.forced_action_detail. Confidence: low for defer_decision_insufficient_signal, else moderate. Skip the action matrix; jump to rationale.

Specifically:

• defer_decision_insufficient_signal → R-coverage fired (no row / macros missing); record the decision verbatim.
• escalate_for_user_review → R-extreme-deficiency fired (big calorie gap AND very-low protein on the same day); record calorie_deficit_kcal + protein_ratio in action_detail so the user can inspect.

2. Otherwise, pick from the action matrix

Keyed on classified_state.nutrition_status:

status	action + action_detail
aligned	maintain_targets
deficit_caloric	reduce_calorie_deficit with {"reason_token": "<calorie_balance_band>", "calorie_deficit_kcal": <value>} — the band names the severity (moderate_deficit or high_deficit)
protein_gap	increase_protein_intake with {"reason_token": "<protein_sufficiency_band>", "protein_ratio": <value>}
under_hydrated	increase_hydration with {"hydration_ratio": <value>}
surplus	maintain_targets with {"caveat": "calorie_surplus_trend"} — surplus alone does not force a correction, only surfaces for awareness
insufficient_data	(v0.1.15 W-D arm-1) Don't reach this row in the matrix — the W-D arm-1 short-circuit also sets coverage_band='insufficient' and uncertainty=("partial_day_no_target",), so step 1's defer_decision_insufficient_signal forced-action path fires first. Surface honestly: "I'm holding off on classifying — partial-day intake against an unset nutrition target wouldn't be meaningful."

3. Confidence

Default from classified_state.coverage_band: full → high, partial → moderate, sparse → moderate, insufficient → low. If policy_result.capped_confidence is set, it lowers the default but never raises it.

4. Micronutrient honesty

classified_state.micronutrient_coverage is always unavailable_at_source in v1 because the daily-macros derivation carries no micronutrient evidence. Do not surface any claim about a specific micronutrient's status — no nutrient-name-qualified "low," "inadequate," or "deficient" prose, no mineral-ratio commentary. Those signals are structurally unavailable until a later release lands meal-level intake. The only allowable micronutrient-adjacent mention is the honest one: micronutrients_unavailable_at_source is already on classified_state.uncertainty; carry it through so downstream knows this domain's proposal speaks only to macros + hydration.

5. Rationale (4–7 lines)

One line per band or signal that informed the decision. Name the band; do not re-derive it.

Examples: calorie_balance_band=<band> with calorie_deficit_kcal=<value>, protein_sufficiency_band=<band> with protein_ratio=<value>, hydration_band=<band>, nutrition_status=<status>, extreme_deficiency_detected (if R-extreme-deficiency fired), micronutrients_unavailable_at_source (always, in v1).

6. Uncertainty

Start with classified_state.uncertainty (already sorted + deduped). Append any tokens you added (e.g. *_unavailable_at_source derived from the snapshot's missingness token). Re-sort alphabetically; deduplicate. micronutrients_unavailable_at_source is already included by the classifier under v1; do not duplicate it.

7. Follow-up

Nutrition emits a NutritionProposal, not a recommendation, so it has no follow_up field. Synthesis assigns review semantics per finalised plan. Skip this step.

On defer_decision_insufficient_signal, synthesis uses the nutrition-domain template "How did yesterday's eating go? Anything worth logging as macros?" (owned by core.narration.templates.DEFER_REVIEW_QUESTION_TEMPLATES).

X9 post-adjust contract

After synthesis commits the training recommendation, Phase B runs X9: if the final training action is a hard session, the runtime appends a protein/carb-target bump to the nutrition recommendation's action_detail — NOT to the action. This is runtime-owned; the skill must not second-guess it or fold its effect into the proposal. The proposal you emit is the pre-synthesis view; the X9 adjustment lives downstream, run by the runtime after synthesis.

Output

Emit a NutritionProposal JSON and call hai propose --domain nutrition --proposal-json <path>. The propose tool validates the shape and appends to proposal_log; it is your determinism check.

proposal_id = prop_<for_date>_<user_id>_nutrition_01 (idempotent on (for_date, user_id, domain); re-running on the same day does not produce a new row).

Copy policy_result.policy_decisions into the output's policy_decisions verbatim — the runtime decided them; you do not re-edit or add new ones. v0.2.0 W-PROV-2: if policy_result.evidence_locators is present, copy that list verbatim into the proposal's evidence_locators field — do NOT derive locators yourself; the runtime computed them. If the field is absent, omit evidence_locators from the proposal.

Invariants

• You never compute a band, a score, a ratio, or a deficit. classified_state is the source of truth.
• You never evaluate an R-rule (require_min_coverage, no_high_confidence_on_sparse_signal, extreme_deficiency_escalation). policy_result is the source of truth; you honour forced_action and capped_confidence.
• You never apply X-rule mutations. Synthesis owns all cross-domain reasoning (X2 nutrition deficit softens strength/recovery; X9 training intensity bumps nutrition targets post-commit). This skill emits one domain's bounded proposal; synthesis mutates the draft mechanically, based on x_rule_firing rows, before the skill ever sees it as "final."
• You never emit an action outside the v1 enum (maintain_targets, increase_protein_intake, increase_hydration, reduce_calorie_deficit, defer_decision_insufficient_signal, escalate_for_user_review). hai propose enforces this.
• You never fabricate micronutrient claims. The data layer does not carry that evidence in v1; silence about micros is the honest state.
• You never fabricate values for missing evidence; missing stays missing (a missing hydration log is unknown, not low).
• If a decision isn't reasoned in rationale[] or policy_decisions[], it didn't happen.