running-readiness

name	running-readiness
description	Produce a bounded RunningProposal for today's run by consuming the runtime-computed `classified_state` + `policy_result` and applying judgment-only steps — action matrix, rationale prose, vendor cross-check, follow-up composition. The runtime already did every band, every score, and every R-rule; this skill does not re-derive them.
allowed-tools	Read, Bash(hai state snapshot *), Bash(hai state read *), Bash(hai clean *), Bash(hai propose *), Bash(hai intent list *), Bash(hai target list *)
disable-model-invocation	false

Running Readiness

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.running you receive:

• today / history — daily rollup rows (distance, intensity minutes).
• activities_today / activities_history — per-session intervals.icu rows (type='Run'), newest-first. Each carries distance_m, moving_time_s, hr_zone_times_s ([Z1..Z7] seconds), interval_summary, trimp, warmup_time_s, cooldown_time_s, feel, icu_rpe. Empty on rest days.
• signals — runtime-derived dict. Classic: weekly_mileage_m, weekly_mileage_baseline_m, recent_hard_session_count_7d, acwr_ratio, training_readiness_pct, sleep_debt_band, resting_hr_band. Structural (v0.1.4): z4_plus_seconds_today, z4_plus_seconds_7d, last_hard_session_days_ago, today_interval_summary, activity_count_14d. Context only; never re-derive.
• classified_state — source of truth. Carries weekly_mileage_trend_band, hard_session_load_band, freshness_band, recovery_adjacent_band, coverage_band, running_readiness_status, readiness_score, uncertainty.
• policy_result — source of truth for 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_running_state_daily; an ACWR-spike firing additionally cites accepted_recovery_state_daily.acwr_ratio.
• missingness — per state_model_v1.md §5.

Reach into activities_today / activities_history for qualitative context — "did today's session match what I planned?" (compare today_interval_summary to evidence.planned_session_type), or session-level colour in rationale ("Z4 for 4:42"). Never recompute a band the classifier already decided.

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.

2. Otherwise, pick from the action matrix

Keyed on classified_state.running_readiness_status × planned session character. The planned character comes from evidence.planned_session_type if present, else infer from history (last hard day, weekly progression).

status	planned character	action + action_detail
ready	any	proceed_with_planned_run — if active_goal present, action_detail = {"active_goal": <goal>}
conditional	intervals | race	downgrade_intervals_to_tempo with {"target_zone": "tempo", "reason_token": "conditional_readiness"}
conditional	tempo | long	downgrade_to_easy_aerobic with {"target_zone": "easy_aerobic"}
conditional	easy | recovery	proceed_with_planned_run with {"caveat": "keep_effort_conversational"}
hold	hard | intervals | tempo | long	cross_train_instead with {"reason_token": "hold_status_avoid_impact"}
hold	easy | recovery	rest_day_recommended with {"suggested_activity": "walk_or_mobility"}

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. The vendor cross-check below may lower further.

4. Vendor cross-check

The snapshot also carries Garmin's own running signals under recovery.today and the running signals block:

• acwr_status (on recovery.today / raw_summary) — Garmin's vendor band (e.g. "PRODUCTIVE", "OVERREACHING"). If acwr_status is overreaching/unproductive but freshness_band is fresh or neutral, cap confidence at moderate and add agent_vendor_acwr_disagreement to uncertainty.
• training_readiness_pct (on signals) — locally-computed mean of Garmin's component pcts. The vendor's categorical level can disagree because Garmin applies internal weighting; if it does, add training_readiness_weighting_disagreement.

5. Rationale (5–8 lines)

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

Examples: weekly_mileage_trend=<band>, hard_session_load=<band>, freshness=<band>, recovery_adjacent=<band>, acwr_status=<vendor_label> (informational), active_goal=<goal> (if present), agent_vendor_acwr_disagreement (if applicable).

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, vendor-disagreement tokens). Re-sort alphabetically; deduplicate.

7. Follow-up

Running emits a RunningProposal, not a recommendation, so it has no follow_up field. The synthesis layer assigns review semantics per finalised plan. Skip this step.

When the action is defer_decision_insufficient_signal, synthesis uses the running-domain template "Did you go for a run yesterday? How did it feel?" (owned by core.narration.templates.DEFER_REVIEW_QUESTION_TEMPLATES) so the question never leaks recovery session-language.

Output

Emit a RunningProposal JSON and call hai propose --domain running --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>_running_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, or a ratio. classified_state is the source of truth.
• You never evaluate an R-rule (require_min_coverage, acwr_spike_escalation, no_high_confidence_on_sparse_signal). 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; this skill emits one domain's bounded proposal.
• You never emit an action outside the v1 enum (proceed_with_planned_run, downgrade_intervals_to_tempo, downgrade_to_easy_aerobic, cross_train_instead, rest_day_recommended, defer_decision_insufficient_signal, escalate_for_user_review). hai propose enforces this.
• You never fabricate values for missing evidence; missing stays missing.
• If a decision isn't reasoned in rationale[] or policy_decisions[], it didn't happen.