中文 EN

Scan Report

Scan New Files

⚠ Low Risk — Risk Score 20/100
Last scan:20 hr ago Rescan
20 /100
Sulcus (openclaw-sulcus-skill)
Reactive, thermodynamic memory management for AI agents with heat-based decay, semantic search, reactive triggers, and cross-agent sync
Sulcus is a legitimate thermodynamic memory management skill. No malicious code, shell execution, credential harvesting, or obfuscation present. Minor doc-to-meta mismatches and webhook trigger risk are the only concerns, both documented and user-controlled.
Skill NameSulcus (openclaw-sulcus-skill)
Duration48.8s
Enginepi
✓
Safe to install
Approve for use. No action required. If concerned about webhook exfiltration, ensure users review active triggers via list_triggers before enabling trigger features.

Findings 3 items

Severity Finding Location
Low
Trigger event count mismatch between SKILL.md and _meta.json Doc Mismatch
SKILL.md defines 6 trigger events (on_store, on_recall, on_decay, on_boost, on_relate, on_threshold) and 6 actions (notify, boost, pin, tag, deprecate, webhook), while _meta.json lists only 4 events (on_create, on_recall, on_decay, on_boost) and 4 actions (notify, boost, tag, deprecate). This is an informational discrepancy — SKILL.md is the authoritative user-facing doc.
SKILL.md line 52: 'on_store | New memory created' vs _meta.json: 'on_create'
→ Align _meta.json trigger events and actions with SKILL.md. Consider adding 'webhook' action to _meta.json.
 SKILL.md, _meta.json:52
Low
Webhook trigger can POST memory content to arbitrary URLs Data Exfil
The webhook trigger action POSTs full memory content (including potentially sensitive stored information) to any user-configured URL. This is a theoretical exfiltration vector if a user creates a malicious trigger pointing to an attacker-controlled endpoint.
webhook | HTTP callback to external URL
→ This is documented with a ⚠️ warning. No action needed from a code standpoint. Users should be informed to only create webhooks pointing to URLs they control, and to audit triggers via list_triggers before enabling in production.
 SKILL.md:52
Info
captureToolResults and captureLlmInsights are documented as capturing sensitive tool outputs Doc Mismatch
When captureToolResults and captureLlmInsights are enabled (auto-capture), the plugin captures and transmits tool results and LLM decisions/preferences to the Sulcus server. While opt-in, this could include sensitive information depending on tool usage.
captureToolResults: Capture significant tool results as memories (default: false)
→ This is already marked opt-in (default: false) and documented. Consider adding a note in the Security section about what types of tool results might be captured when enabled.
 SKILL.md:130
ResourceDeclaredInferredStatusEvidence
Filesystem NONE NONE — No file I/O operations found. This is a pure-document skill with no executable c…
Network READ WRITE ✓ Aligned SKILL.md declares network:READ via webhook triggers and API calls to api.sulcus.…
Shell NONE NONE — No subprocess, exec, or shell command execution found in any file.
Environment NONE NONE — No environment variable access observed.
Skill Invoke NONE NONE — No skill-to-skill invocation patterns detected.
Clipboard NONE NONE — No clipboard access.
Browser NONE NONE — No browser automation.
Database NONE NONE — No direct database access. Remote server-side storage on api.sulcus.ca is docume…
15 findings
🔗
Medium External URL 外部 URL
https://clawhub.ai/mcdoolz/openclaw-sulcus-skill
SKILL.md:10
🔗
Medium External URL 外部 URL
https://clawhub.ai/packages/mcdoolz/@digitalforgestudios/openclaw-sulcus
SKILL.md:10
🔗
Medium External URL 外部 URL
https://www.npmjs.com/package/@digitalforgestudios/openclaw-sulcus
SKILL.md:10
🔗
Medium External URL 外部 URL
https://sulcus.ca
SKILL.md:10
🔗
Medium External URL 外部 URL
https://api.sulcus.ca
SKILL.md:260
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/agent/nodes
references/api.md:22
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/agent/search
references/api.md:35
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/agent/nodes/UUID
references/api.md:44
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/mcp
references/mcp-setup.md:55
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/settings/thermo
references/thermodynamics.md:73
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/feedback
references/thermodynamics.md:123
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/triggers
references/triggers.md:37
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/triggers/TRIGGER_ID/history
references/triggers.md:89
🔗
Medium External URL 外部 URL
https://api.sulcus.ca/api/v1/triggers/TRIGGER_ID
references/triggers.md:92
📧
Info Email 邮箱地址
[email protected]
SKILL.md:4

File Tree

7 files · 37.3 KB · 1065 lines
Markdown 6f · 939L JSON 1f · 126L
├─ ▾ 📁 references
│ ├─ 📝 api.md Markdown 110L · 3.4 KB
│ ├─ 📝 mcp-setup.md Markdown 63L · 1.6 KB
│ ├─ 📝 openclaw-setup.md Markdown 114L · 3.5 KB
│ ├─ 📝 thermodynamics.md Markdown 133L · 3.5 KB
│ └─ 📝 triggers.md Markdown 130L · 3.4 KB
├─ 📋 openclaw.plugin.json JSON 126L · 5.6 KB
└─ 📝 SKILL.md Markdown 389L · 16.3 KB

Security Positives

✓ No executable code — skill is pure documentation and config, eliminating most attack surface
✓ No shell execution, subprocess, or any form of code execution found
✓ No credential harvesting or environment variable scanning
✓ No obfuscation (base64, eval, atob) or anti-analysis techniques
✓ No downloads, no remote script execution (curl|bash, wget|sh)
✓ No sensitive file/path access (~/.ssh, ~/.aws, .env)
✓ autoRecall and autoCapture are disabled by default (explicit opt-in model)
✓ API key is user-provided and stored in user-controlled config, not hardcoded
✓ Strong documentation with security warnings for webhook triggers
✓ Self-hosted option (sulcus-local) available for fully offline operation
✓ Open source with verifiable GitHub repository
✓ Tenant isolation and server-side ACL for cross-namespace access