Agent Economy ReportAgent Economy ReportCAUTION version 1.0.0 · audited 2026-09-09
This is a short documentation-only skill: two markdown files that tell an AI agent how to inspect, wipe, and re-seed Clawdbot's local LanceDB vector memory. It contains no scripts, no network calls, no credential access and no installer. The main real-world risk is a destructive shell command it recommends, `rm -rf ~/.clawdbot/memory/lancedb/`, plus a suggested monthly cron job that repeats that wipe automatically without human review. It also suggests changing the memory plugin config to turn off auto-capture. Nothing in the package sends data anywhere or tries to override the agent's other rules, but the delete-and-schedule advice deserves a human check before use.
| Declared purpose | Audit, clean, and optimize Clawdbot's LanceDB vector memory to reduce junk memories and token waste, including maintenance automation. |
| Observed behavior | Provides markdown instructions only: a memory_recall audit query, a recursive delete of the local LanceDB directory, a gateway restart, memory_store reseeding examples, a JSON config patch disabling autoCapture, and a cron job template for monthly wipe-and-reseed. No executable files, no network requests, no credential or environment variable access. |
| Verdict, rules only | CAUTION · uses sudo, rm -rf or dd |
| Verdict, AI | CAUTION |
| Final verdict | CAUTION (never better than either pass) |
| capability | vs purpose | detail |
|---|---|---|
| shell | needed | Instructs the agent to run `rm -rf ~/.clawdbot/memory/lancedb/` and `clawdbot gateway restart` (SKILL.md lines 20, 22). Destructive but consistent with the declared 'wipe memory' purpose. |
| filesystem_write | needed | Deletes the local memory database directory and writes new memory entries via memory_store (SKILL.md lines 20, 26). Scope is limited to ~/.clawdbot/memory/. |
| persistence | needed | Suggests adding a recurring cron job named 'memory-maintenance' that wipes and reseeds memory monthly (SKILL.md lines 69-73). Declared in the skill's own docs, but it is a scheduled unattended destructive task. |
| network | exceeds purpose | No network calls in code. Only two informational URLs appear: a GitHub homepage (SKILL.md line 4) and the ClawHub publisher page (skill-card.md line 7). |
| severity | finding |
|---|---|
| medium | Recursive delete of the memory database directory The skill instructs the agent to permanently delete the LanceDB memory directory with no backup step, no confirmation prompt and no dry run. This matches the declared purpose (wiping memory) but is irreversible; if the path is wrong or expanded differently, data loss is unrecoverable. SKILL.md:20 rm -rf ~/.clawdbot/memory/lancedb/ |
| medium | Scheduled unattended destructive job Recommends registering a cron job that repeats the memory wipe every month at 04:00 without human review. This creates ongoing automated data deletion; any facts not captured in MEMORY.md are lost each cycle. SKILL.md:72 "text": "Monthly memory maintenance: 1) Wipe ~/.clawdbot/memory/lancedb/ 2) Parse MEMORY.md 3) Store key facts to fresh LanceDB 4) Report completion" |
| low | Modifies agent plugin configuration Instructs the agent to patch its own gateway configuration to set autoCapture false for the memory-lancedb plugin. This is within the declared purpose but changes persistent agent behavior beyond a one-off cleanup. SKILL.md:48 Use `gateway action=config.patch` to apply. |
| low | Advises storing locations of credentials in vector memory Encourages writing where credentials live into the long-term memory store. Nothing here transmits that data anywhere, but recording credential locations in a searchable memory index increases exposure if memory is ever leaked or recalled into a prompt. SKILL.md:55 - Important facts (accounts, credentials locations, contacts) |
| info | Publisher discloses the destructive behavior The skill card openly documents the data-loss and unattended-reset risks and suggests backups, which is consistent with the code and supports a CAUTION rather than UNSAFE reading. skill-card.md:19 Risk: Wipe instructions can permanently delete Clawdbot vector memory. <br> |
0 AI finding(s) were dropped because their file, line or quote did not match the package.
| domain | service rating | where |
|---|---|---|
| clawhub.ai | not in the service index | skill-card.md:7 |
| github.com | not in the service index | SKILL.md:4 |
| Binaries invoked | bash, rm -rf |
| Environment variables read | none |
| Hard-coded wallets | none |
| Pipes a download to a shell | 0 |
| eval / exec / subprocess | 0 |
| base64 blobs | 0 |
| File writes | 1 |
| Persistence | 0 |
| file | lines | sha256 |
|---|---|---|
SKILL.md | 82 | 58d9e56da3f96b41… |
_meta.json | 6 | 0c7d8c2b161c0a28… |
skill-card.md | 42 | 66bd4c75ec333b8c… |
For agents
JSON: https://agenteconomy.report/k/memory-hygiene.audit.json · badge: https://agenteconomy.report/k/memory-hygiene.audited.svg ·
skill rating: /k/memory-hygiene · commission an audit of any skill: US$ 29 per version.
The complete published package of this exact version was downloaded from the registry and read statically; nothing was executed. A deterministic pass extracts network destinations, binaries, environment variables, writes, install commands, obfuscation markers, subprocesses and wallets, each with file and line. An AI then reads the whole package with those facts and writes the summary, the capabilities and the findings under a strict schema; every finding it produces must cite a file, a line and the exact text, or it is dropped. The final verdict is the worst of the two passes. The audit does not change the skill's trust tier (policy); the author may respond through the dispute channel and the response is published here. Commissioned by: the Agent Economy Report (free program: the 150 most downloaded skills, September 2026).