Agent Health Optimizer

Diagnose, score, and steadily improve your OpenClaw setup.

This skill is an audit toolkit, not a magic self-healing system. It is good at surfacing likely problems, weak spots, and opportunities. Its --fix mode is intentionally conservative.

Requirements

• - python3 (3.8+)
• openclaw CLI

Quick Start

CODEBLOCK0

What It Does

🏥 health_score.py — Agent Health Grade (0-100)

Scores 5 dimensions:

• - 🧠 Memory (25pts): MEMORY.md completeness, daily log activity, working buffer, anti-poisoning hygiene, source tags
• ⏰ Cron (25pts): job health, schedule diversity, use of isolated sessions, suspicious delivery setups, selective stagger usage
• 📦 Skills (20pts): count, overlap hints, ClawHub management ratio, metadata completeness
• 🔒 Security (15pts): safety rules, anti-poisoning policy, WAL protocol, external action controls
• 🔄 Continuity (15pts): SOUL.md, USER.md, HEARTBEAT.md, IDENTITY.md, git tracking

🔍 memory_auditor.py — Memory Hygiene

Detects:

• - Imperative rules that should be declarative facts
• Missing source tags on factual entries
• Stale entries >30 days with pending status
• External content stored as instructions
• Oversized files needing archival
• Daily log gaps

⏰ cron_optimizer.py — Cron Job Doctor

Detects:

• - Error states with job names and error messages
• Time collisions (multiple jobs on same schedule)
• Missing stagger on burst-prone recurring schedules
• Suspicious announce setups (for example explicit channel without explicit to)
• Timeout mismatches
• Session target recommendations (isolated vs main)

INLINECODE2 mode:

• - creates memory/cron-backup.json before changes
• only auto-adds stagger to recurring top-of-hour stampede-prone jobs
• does not force delivery on jobs using INLINECODE4
• does not modify exact-time jobs just because they lack stagger

📦 skill_comparator.py — Skill Landscape Checker

Via ClawHub API (https://clawhub.ai/api/v1/):

• - Fetches stars, downloads, installs for installed skills
• Lists top ClawHub skills you're missing
• Finds adjacent / overlapping skills with stronger community signals
• Category coverage analysis (what domains are missing?)

Important: these are comparison hints, not authoritative replacements.

🔄 self_optimize.py — Unified Runner

Runs all 4 tools and produces:

• - Combined report with prioritized action items (HIGH/MED/LOW)
• Trend tracking vs prior run (📈/📉)
• JSON reports in memory/ for historical review

What It Reads & Writes

Reads (non-destructive):

• - Workspace files: MEMORY.md, AGENTS.md, SOUL.md, USER.md, HEARTBEAT.md, IDENTITY.md
• Daily logs: INLINECODE7
• Skill metadata: INLINECODE8
• Cron config: INLINECODE9
• ClawHub public API: INLINECODE10

Writes (reports only):

• - INLINECODE11
• INLINECODE12
• INLINECODE13
• INLINECODE14
• INLINECODE15
• INLINECODE16

Modifies (only with --fix):

• - cron_optimizer.py --fix edits cron jobs via INLINECODE19
• Always backs up to memory/cron-backup.json first

Periodic Self-Optimization

Set up a weekly cron (read-only, no --fix):

CODEBLOCK1

Positioning

Use this skill when you want:

• - a periodic audit
• a sanity check before changing agent architecture
• conservative recommendations
• a historical trend of agent health

Do not treat it as a substitute for human review on destructive or far-reaching changes.

Credits

Diagnostic patterns informed by:

• - proactive-agent by halthelobster
• self-improving-agent by pskoett
• Moltbook openclaw-explorers community — cron jitter pattern (thoth-ix), heartbeat batching (pinchy_mcpinchface)

License

This work is licensed under CC BY-SA 4.0. You are free to share and adapt, with attribution and same-license requirement.