Ask: Update all routines, scheduled tasks, and scheduled events so every run ends with a post-run QA reflection (8-point self-audit + 4-way MC questions in Sam's format, MC block inside the Notion export, context-aware handoff, QA: Open / QA: Clean tag).

What shipped

1. Canonical skill created

• Path: /Users/samaguiar/Documents/Projects/Skills/post-run-qa-reflection/SKILL.md
• Purpose: Single source of truth for the QA reflection routine. 8-point self-audit (scope gaps, missing tools, unnecessary work, skill gaps, location access, permissions, downstream effects, future-agent notes), Sam's 4-way MC format (A recommended, B, C, D=Other), MC block placement in both the run report and the Notion export, QA: Open / QA: Clean tag rule, compaction/handoff rules.
• Effect: Future QA format changes = one edit. No more copy-pasting an inline QA block into every individual SKILL.md.

2. CLAUDE.md universal hook

• Path: /Users/samaguiar/Documents/Projects/CLAUDE.md
• New section: Post-Run QA Reflection (MANDATORY for every scheduled task and real-work session) with a pointer to the canonical skill. Applies to all tasks, even the ones whose SKILL.md was never edited.

3. Active scheduled tasks migrated (full prompt push)

Full scheduler prompt updated via mcp__scheduled-tasks__update_scheduled_task AND on-disk SKILL.md updated to match:

• gads-search-term-review — daily search-term review for SAIL - Car Accidents 2026 (customer 3813916687, MCC 8676599345, campaign 23723841732). Direct Google Ads API v30. Recommendation-only.
• google-ads-neg-cleanup — weekly negative-keyword dedup, same scope, recommendation-only unless pre-approved queue.
• gads-wasted-spend — disabled-but-active SKILL.md updated to reference canonical skill; scheduler prompt already points to the launchd migration note from 2026-04-21.
• gads-campaign-health — reconstructed prompt (task had no stored SKILL.md). Flagged for Sam: confirm reconstructed prompt matches the intended procedure (MC question below).

4. Disabled tasks — light-touch update

19 disabled tasks: description field overwritten with [Disabled — prompt not reconstructed] QA hook enforced via CLAUDE.md. On any manual run, load post-run-qa-reflection skill at /Users/samaguiar/Documents/Projects/Skills/post-run-qa-reflection/SKILL.md before ending. Prompts NOT touched — avoids clobbering original intent. Safe to re-enable; QA block will fire via the CLAUDE.md hook regardless.

5. Helper script

• Path: /Users/samaguiar/Documents/Codex/tools/get-scheduled-task.sh (chmod +x)
• Usage: get-scheduled-task.sh <taskId> — prints the on-disk SKILL.md, any sidecar task.json / metadata.json, and a reminder that the live scheduler prompt lives in the scheduler, not on disk.