🩺 memory-health-check

Performs a 6-dimension memory health check assessing integrity, freshness, bloat, orphans, duplicates, and coverage with optional auto-repair.

Security Scan

The skill's code largely matches its stated purpose (scanning and scoring local OpenClaw memory), but it also reads other agents' memory paths, offers automatic repair that can remove files, and references other skills/paths (e.g., dreaming-optimizer) without declaring those dependencies — these mismatches warrant caution before installing or running with auto-repair enabled.

Purpose & Capability

The name/description match the code: scripts scan SQLite and Markdown memory files, compute bloat/freshness/dedup/coverage, and generate reports. However the code is broader than a single-agent scan: it enumerates ~/.openclaw/workspace/*/memory/ (other workspace agents) and ~/.openclaw/memory/*.sqlite, which means it will inspect memory belonging to other agent workspaces on the same host. The health_score code also emits CLI recommendations that point to another skill's path (~/.openclaw/workspace/skills/dreaming-optimizer/...), but the skill does not declare a dependency on that other skill — this is a cross-skill reference that is unexpected and could lead to non-functional or surprising recommendations.

Instruction Scope

Runtime instructions and code will read file contents and SQLite DBs under ~/.openclaw and workspace subdirs (including other agents). The skill includes an auto-repair feature (scripts/auto_repair.py) which is invoked by the main entrypoint when run with --auto-repair; that script can remove orphaned files and temp files. The package will write reports to ~/.openclaw/workspace/memory/health-reports/*.json. These file reads/writes and deletion capabilities are consistent with the stated purpose but expand scope to other agents' memories and include potentially destructive actions — the skill's SKILL.md says auto-repair is 'on user approval' but the CLI will run auto_repair.py automatically when the flag is passed, and an autonomous agent invocation could choose that flag unless additional safeguards exist.

Install Mechanism

There is no external install spec or network download; this is an instruction+code package with all scripts included. That lowers supply-chain risk: code will run from the skill bundle itself. No brew/npm/URL installs are present in the manifest.

Credentials

The skill declares no required environment variables or credentials, and the code does not appear to attempt to read secrets. That's proportionate. However it accesses multiple filesystem locations under ~/.openclaw and workspace directories (including other agents), which is broader-than-expected filesystem access for some users; it also references other skill paths/CLI commands (dreaming-optimizer) in recommendations without declaring them.

Persistence & Privilege

The skill is not marked always:true and uses normal autonomous invocation defaults, which is expected. The noteworthy privilege is the ability to remove files via scripts/auto_repair.py. While auto-repair only runs when explicitly requested (--auto-repair), an autonomous agent could invoke the skill with that flag; combined with the ability to scan other agents' memory and delete files, this raises operational risk if the skill is allowed to run autonomously without human confirmation.

What to consider before installing

This skill appears to implement the described memory health checks, but exercise caution before running it on a real profile: - Review scripts/auto_repair.py before using --auto-repair. The auto-repair feature can delete files (orphans, temp files). Run it in dry-run mode first and back up your memory directory. - Note the skill scans ~/.openclaw/workspace/*/memory/ and ~/.openclaw/memory/*.sqlite — it may read other agents' memory on the same host. If you don't want cross-workspace scanning, run the scripts with an explicit base-dir argument or inspect/modify find_memory_dbs/find paths. - The health_score recommendations reference external skill paths (dreaming-optimizer) that are not declared; those recommended commands may not exist and could be confusing or misleading. - There are no network downloads or obvious exfil endpoints in the provided files, but the package writes reports to ~/.openclaw/workspace/memory/health-reports/ (inspect generate_report.py to confirm it does not upload reports anywhere). If you need higher assurance, open and review generate_report.py and scripts/auto_repair.py for any network or deletion behaviour not covered here. - Recommended actions: run the tool first in a safe/testing environment (or with --dims to limit what runs), review auto-repair's deletion rules, run dry-run, and keep backups. If you want the skill to only scan a single agent's memory, pass a specific --base-dir or edit the search logic to narrow paths. What would change this assessment: seeing the full auto_repair.py and generate_report.py contents (to confirm exact deletion and any network/upload behavior) would increase confidence; explicit declaration of any cross-skill dependencies (dreaming-optimizer) or limiting scans to a single agent workspace would reduce the 'suspicious' flags.

Version: 1.0.0
Skill ID: memory-health-check
Status: Ready for Development

Quick Reference

项目	内容
核心功能	6维内存健康诊断（完整性/新鲜度/膨胀率/孤儿/去重/覆盖率）
主入口	bin/health_check.sh
依赖	OpenClaw v2026.4.9+, sqlite3

Features (MVP)

1. Integrity scan — DB corruption / checksum checks
2. Bloat detection — DB size, file count, growth rate
3. Orphan detection — Entries with zero inbound references
4. Freshness report — Entry age distribution
5. Dedup scanner — Duplicate / near-duplicate entries
6. Health score — Aggregate 0–100 score across all dimensions
7. Auto-repair — Orphan cleanup on user approval

File Structure

memory-health-check/
├── SKILL.md                   # This file
├── bin/
│   ├── health_check.sh        # Main entry point
│   ├── integrity_scan.py      # DB corruption checks
│   ├── bloat_detector.py      # Size analysis
│   ├── orphan_finder.py       # Reference graph orphan detection
│   ├── dedup_scanner.py       # Duplicate detection
│   ├── freshness_report.py     # Entry age distribution
│   └── health_score.py         # Aggregate scoring
├── scripts/
│   ├── generate_report.py      # Report generator
│   └── auto_repair.py         # Cleanup script
├── config/
│   └── thresholds.yaml
├── reports/
│   └── .gitkeep
└── README.md

Usage

# Full health check
memory-health-check/bin/health_check.sh

# With auto-repair
memory-health-check/bin/health_check.sh --auto-repair

# Specific dimensions only
memory-health-check/bin/health_check.sh --dims integrity,bloat

Health Score Dimensions

Dimension	Healthy	Warning	Critical
Integrity	✅	⚠️	🔴
Freshness (>70%)	>70%	40–70%	<40%
Bloat	<500MB	500MB–2GB	>2GB
Orphans	0%	1–5%	>5%
Dedup	<2%	2–10%	>10%
Coverage	>80%	50–80%	<50%

Pricing

Tier	Price	Features
Free	$0	1 scan/week, text report
Pro	$9.90/mo	Unlimited scans, 6-dim diagnostics, auto-repair, 90-day history
Bundle	$17.90/mo	dreaming-optimizer Pro + memory-health-check Pro

Development

See: ``