How the architect operates during the weekly architecture-health-check CRON — what to scan, how to produce the health report, and how findings flow to the immune system and PM.
Your architecture-health-check CRON fires weekly. It runs in an isolated session with proper model and effort. Your job in that session is specific: scan the codebase for architectural drift, produce a health report, and surface findings.
When the CRON fires, you're in an isolated session. No task context. No heartbeat data. Just your CRON message telling you to run the health check. Your tools, skills, and sub-agents are available.
Use dependency-mapper sub-agent:
Agent: dependency-mapper
Prompt: "Scan fleet/core/ for imports from fleet/infra/, fleet/mcp/,
or fleet/cli/. Report any violations of inward dependency direction."
Expected: zero violations. Any match is structural drift.
Scan for modules approaching the 500-line convention:
Agent: code-explorer
Prompt: "List all Python files in fleet/ over 300 lines with line counts.
Flag any over 500 as split candidates."
Use pattern-analyzer sub-agent:
Agent: pattern-analyzer
Prompt: "Compare patterns used in fleet/infra/ clients (mc_client,
irc_client, gh_client, plane_client). Report any inconsistencies
in how they handle errors, configure clients, or expose interfaces."
Review recent commits across the codebase:
git log --since="7 days ago" --stat --oneline
Look for: different conventions in different areas, new patterns that deviate from established ones, config loading inconsistencies.
fleet_artifact_create("analysis_document", "Architecture Health — Week of {date}")
Structure:
## Architecture Health Report — Week of {date}
### Dependency Direction: {PASS/FAIL}
- Violations: {count}
- Details: {list if any}
### Module Size: {OK/WARNING}
- Files approaching limit: {count}
- Largest: {file} ({lines} lines)
### Pattern Consistency: {OK/DRIFT}
- Inconsistencies: {count}
- Details: {list if any}
### Cross-Agent Drift: {OK/DRIFT}
- Observations: {list}
### Overall: {HEALTHY/DEGRADING/NEEDS ATTENTION}
### Actions Needed:
- {specific tasks to create or flag}
# Post to board memory for PO and fleet-ops visibility
fleet_chat(
"Architecture health check complete. Status: {overall}. "
"{findings_summary}",
mention="project-manager" # PM creates tasks for issues
)
If critical issues found:
fleet_alert(
category="architecture",
severity="high",
details="Dependency direction violation in fleet/core/{module}"
)
The CRON session automatically records trail. But explicitly post:
# Board memory with architecture tags
# This feeds fleet-ops compliance checks and accountability patterns
From config/agent-crons.yaml:
- name: architecture-health-check
schedule: "0 8 * * 1" # Monday 8am weekly
model: sonnet # sonnet sufficient for scanning
effort: medium # not deep thinking, structured scanning
session: isolated # clean session, no task contamination
This CRON DETECTS. PM TRIAGES. Agents FIX. You separate detection from remediation.
This CRON uses knowledge from:
/fleet-architecture-health — what to look for/fleet-domain-boundary-enforcement — dependency direction rules/fleet-srp-verification — module responsibility checks/fleet-design-pattern-selection — pattern consistency referenceThe CRON is WHERE you apply these skills. The skills are WHAT you know. Together they produce the health report.