v0.10.21

Skills-focused release: 7 new workflow skills and a quality pass over the whole bundled-skill set. (Built from the v0.10.20 baseline + skills changes only.)

Added

• /tracking-health — preventive conversion-tracking audit: Meta pixel inventory + result_indicator audit, Google conversion-action status, GA4 divergence flags, per-platform scorecard with ranked fixes. (#390)
• /budget-pacing — month-to-date spend vs monthly budget target, run-rate and projected landing, ±10% pace alerts, Guardrails-aware daily-budget recommendations. (#390)
• /monthly-report — client-facing monthly digest: MoM comparison, Goal attainment, action-log recap with outcome verdicts, next-month recommendations. (#390)
• /experiment — designed A/B tests with evidence discipline: falsifiable hypothesis, one variable, fixed window, no peeking, per-variant mureo_outcome_evaluate verdicts (inconclusive is a valid outcome). Meta native split tests; honest manual paths on Google Ads. (#391)
• /audience-review — audits targeting against the STRATEGY.md Persona using the demographic/audience/device read tools; flags mismatches and placement waste; executes only supported mutations. (#391)
• /ad-fatigue-check — frequency (derived impressions÷reach) + CTR-decay fatigue scoring with noise guards; routes fatigued ads to /creative-refresh. (#391)
• /incident-postmortem — post-rescue timeline reconstruction, root-cause analysis, structured postmortem doc, /learn insights, preventive Guardrails proposals. No ad-platform writes. (#391)

Changed

• Skills polish (#389): the diagnostic preamble (learning insights + advisor consult) now lives once in _mureo-shared (skills carry a pointer); every bundled skill's version is normalized to the package version (with a parity test); stale Google Ads "v18" label corrected to v23; Codex host coverage documented; weekly-report gains the diagnostic preamble.

19 operational skills ship in this release.