background-workers

name	background-workers
description	Formalized background worker system for automated audit, memory management, state sync, and optimization tasks. Workers run on intervals during active sessions to maintain system health without manual intervention. Use when: configuring workers, debugging worker failures, adding new automated tasks. Skip when: workers are running correctly.
tags	["automation","workers","system-health"]
triggers	["background workers","use background workers","run background workers"]
tier	standard

Background Workers — Automated System Maintenance

Purpose: Manual maintenance doesn't scale. Background workers handle audit, memory compression, state sync, and optimization automatically during active sessions.

Worker Registry

Four workers, each with a defined priority and interval:

1. Audit Worker (CRITICAL — every 5 min)

Purpose: Continuous security and compliance monitoring.

Tasks:

Task	What It Does	On Failure
Secret scan	Grep staged files for API keys, tokens, passwords	BLOCK commit, alert CC
Credential file check	Verify .env.agents not in git staging	Unstage immediately
Hook audit review	Check tmp/hook_audit.log for anomalies	Flag suspicious entries
Permission verify	Confirm no agent exceeded its claims	Log violation to MISTAKES.md

Implementation:

# Secret scan (runs on staged files)
git diff --cached --name-only | xargs grep -l -E "(sk_live_|Bearer |api_key|password\s*=)" 2>/dev/null

# Credential file check
git diff --cached --name-only | grep -E "\.env" && echo "BLOCKED: .env file staged"

# Hook audit log size check
wc -l tmp/hook_audit.log 2>/dev/null | awk '{if ($1 > 1000) print "WARN: audit log bloated"}'

Config: .agents/config.toml [workers.audit]

2. Memory Worker (HIGH — every 30 min)

Purpose: Prevent memory bloat and maintain freshness.

Tasks:

Task	What It Does	Threshold	Action
SESSION_LOG.md size	Count lines	>200 lines	Compress entries >14 days → ARCHIVES/
ACTIVE_TASKS.md count	Count items	>50 items	Remove completed tasks >7 days old
brain/ combined size	Count total lines	>500 lines	Flag for manual review
Confidence decay	Check fact freshness	>30 days unverified	Reduce confidence by 0.1
Duplicate scan	Check for repeated entries	Any duplicates	Merge and deduplicate

Compression protocol:

1. Read SESSION_LOG.md
2. Identify entries older than 14 days
3. Move them to memory/ARCHIVES/sessions-YYYY-MM.md
4. Keep the 10 most recent entries in the active file
5. Update any cross-references

Config: .agents/config.toml [workers.memory]

3. Sync Worker (HIGH — every 10 min)

Purpose: Keep all AI interfaces working from the same ground truth.

Tasks:

Task	What It Does	On Failure
STATE.md update	Refresh operational state from current session	Reconstruct from ACTIVE_TASKS.md
Supabase flush	Push pending traces to agent_traces table	Queue for next cycle
ACTIVE_TASKS.md verify	Ensure task statuses reflect actual progress	Flag stale tasks
Cross-agent check	Verify no conflicting state from other agents	Merge (newer wins)

Sync protocol:

1. Read current brain/STATE.md
2. Compare against actual session activity
3. Update any stale fields
4. If Supabase available: flush pending agent_traces
5. Verify ACTIVE_TASKS.md hasn't drifted from actual progress

Config: .agents/config.toml [workers.sync]

4. Optimize Worker (LOW — every 60 min)

Purpose: Long-term system health and efficiency.

Tasks:

Task	What It Does	On Finding
Unused skill detection	Check skill_activation table for dormant skills	Report (don't delete)
Duplicate memory entries	Scan LONG_TERM.md + PATTERNS.md for duplicates	Merge automatically
Referential integrity	Grep for broken wiki-links in markdown	Fix broken links
Pattern promotion	Check [PROBATIONARY] patterns with 3+ uses	Promote to [VALIDATED]
SOP success rates	Review SOP execution counts and failure rates	Flag SOPs with <70% success

Config: .agents/config.toml [workers.optimize]

Worker Execution Model

Workers are NOT separate processes — they are logical checkpoints that run during active sessions:

1. Session start: All workers run their checks once (initial health scan)
2. During session: Workers are triggered at their configured intervals
3. Session end: All workers run final checks (cleanup pass)
4. Between sessions: No workers run (system is idle)

Priority determines execution order when multiple workers trigger simultaneously:

• CRITICAL (audit) → runs first, always
• HIGH (memory, sync) → runs second
• LOW (optimize) → runs last, skipped if session is ending

Adding New Workers

To add a new background worker:

1. Define it in .agents/config.toml under [workers.new_name]:

[workers.new_name]
enabled = true
priority = "high"     # critical | high | low
interval_seconds = 600
tasks = [
    "description of task 1",
    "description of task 2"
]

2. Add the implementation details to this skill file
3. Update brain/CAPABILITIES.md with the new worker count

Worker Health Monitoring

Track worker execution in the session:

WORKER STATUS:
  Audit: Last run [time] — [PASS | WARN: details]
  Memory: Last run [time] — [PASS | COMPRESSED: N entries]
  Sync: Last run [time] — [PASS | SYNCED: N traces]
  Optimize: Last run [time] — [PASS | FOUND: N issues]

If any worker consistently fails, log to MISTAKES.md and investigate root cause.

Integration Points

• Self-healing skill: Workers ARE the automated arm of self-healing
• Memory management skill: Memory worker implements the bloat prevention rules
• Hooks automation: Post-task hooks trigger worker-like checks per-operation
• Brain Loop Step 10 (HEAL): The HEAL step runs a subset of worker checks

Obsidian Links

• [[skills/self-healing/SKILL.md]] | [[skills/memory-management/SKILL.md]]
• [[skills/hooks-automation/SKILL.md]] | [[brain/BRAIN_LOOP]]
• [[brain/CAPABILITIES]] | [[brain/STATE]]