Agent Workflow Automation Playbook

Type: Ops Skill Version: 1.0.0 Price: $29

Every agent deployment hits the same friction points: cron jobs that silently fail, sub-agents that lose context, memory that drifts after 3 days, deliveries that get dropped, escalations that never fire.

This playbook documents 10 automation patterns extracted from running a live AI-operated company for 8 weeks. Not theory. Production patterns with the failure modes already burned in.

---

What Is Included

10 workflow automation patterns:

1. Cron-to-Telegram delivery — reliable group/DM delivery with ISO timestamp scheduling. The exact flag order that works (--channel, --to, --at, --announce).

2. Sub-agent context handoff — how to pass project state to a spawned sub-agent so it does not start blind. The 3-field minimum: TASKS.md excerpt, MEMORY.md summary, explicit task boundary.

3. Memory sync enforcement — validator script + pre-commit hook that blocks commits when project files are stale. Catches the drift that makes agents useless after a week.

4. Heartbeat task queue — HEARTBEAT.md pattern for batching periodic checks (email, calendar, cron health) into one turn instead of 6 separate cron jobs.

5. Blocker surfacing loop — how to surface the same blocker every report without annoying the operator. The cadence: surface daily if active, weekly if silent, never drop it from memory.

6. Quality gate before ship — score-must-improve pattern. Output only ships if it scores higher than the last run. Prevents the loop of shipping the same broken thing repeatedly.

7. Nightly memory audit — deep pass pattern: read all daily notes, cross-reference project files, update MEMORY.md with distilled findings, commit. Runs after 9 PM, skips if already ran today.

8. Escalation ladder — when to alert vs when to stay quiet. Thresholds: CRITICAL fires immediately, DEGRADED batches to next morning report, HEALTHY stays silent.

9. Cron failure recovery — what to do when a cron errors. Check last run status, read the error, determine if one-off or structural. Structural failures get a HEARTBEAT.md entry until fixed.

10. Session startup protocol — the 4-file read sequence that orients an agent in under 60 seconds: SOUL.md, USER.md, memory/YYYY-MM-DD.md (today + yesterday), MEMORY.md.

---

Who It Is For

Operators who have an agent running but keep hitting the same operational failures. The patterns here are the difference between an agent that compounds and one that drifts into uselessness.

Good fit if:

• Your agent loses context after a few days
• Cron jobs fire but you are not sure if they delivered
• You want memory across sessions without building it from scratch
• You keep finding the same mistake repeated across sessions

---

Format

Each pattern is documented as:

• Problem: what breaks without it
• Pattern: the exact implementation
• Failure modes: what goes wrong and how to detect it
• Copy-paste template: ready to drop into your workspace

---

Changelog

• 1.0.0 (2026-04-20): Initial release. 10 patterns from 8 weeks of live AI company operations.