Scheduled-task resilience audit — 2026-05-01

What changed

Sam articulated a foundational issue: scheduled tasks are written in a brittle, step-locked style ("Do X, then Y, stop if anything is missing"). When one dependency moves — API key rotates, MCP renamed, path drifts — the whole run fails or produces a degraded "I couldn't do my job" report.

The fix: rewrite every scheduled task in an outcome-first, resource-aware, fallback-aware, agent-trusting style. State the goal. List the available resources ranked. Give the agent a fallback ladder. Trust it to navigate.

What was built

Standalone skill at /Users/samaguiar/Documents/Projects/Skills/task-resilience-auditor/SKILL.md. Covers seven antipatterns, anatomy of a good prompt, audit checklist, rewrite procedure, six standard fallback ladders by data type (GBP, GSC, Google Ads, WordPress, Litify, Notion), agent-empowerment block, mandatory header badges (brittleness + last_successful_run + deliverable_path), four linter rules.
Slash-command prompt at /Users/samaguiar/Documents/Claude/Scheduled/_prompts/task-resilience-audit-prompt.md. Aliases: /task-audit, /audit-scheduled-tasks, /resilience-rewrite, /agent-trust-audit.
Weekly Saturday 5am audit task — DRAFT at /Users/samaguiar/Documents/Claude/Scheduled/_drafts/scheduled-task-resilience-audit-DRAFT.md. Cannot create from inside scheduled session; Sam creates in next live session.

Sam-approved decisions (2026-05-01)

Q1: Full sweep, all 30 tasks — but get explicit approval on the first 2.
Q2: First 2 require approval, then full pre-approval for the rest.
Q3: Yes recurring — weekly Saturday 5am.
Q4 (multi): Add brittleness score header, add last_successful_run badge, add cross-platform path linter, force defined deliverable file path.
Approved competitive-review-monitor v2 push.
Approved daily-briefing v2 push.
Approved bulk sweep of remaining 28 under pre-approval.

What was applied

Task	Score	Action
competitive-review-monitor	1 (post-rewrite)	Full v2 rewrite — 7-tier data-source fallback ladder
daily-briefing	1 (post-rewrite)	Full v2 rewrite — per-section upstream/live fallback table
28 other tasks	3	Resilience overlay prepended (preserves original body, adds badges + standard fallback rules + agent-empowerment + cross-platform paths)

All modified SKILL.md files have a .v1-backup-2026-05-01 sibling for rollback.

Known scheduler-cache caveat

The scheduled-task system loads the prompt from /Users/samaguiar/Documents/Claude/Scheduled/<taskId>/SKILL.md. SKILL.md files are updated. The scheduler may also have an internal prompt cache populated via update_scheduled_task — that API is blocked from inside a scheduled-task session. Sam should run update_scheduled_task once for competitive-review-monitor and daily-briefing in the next live session to confirm sync. The overlay tasks pull SKILL.md at runtime (per the prompt body), so those don't need re-syncing.