ClawBridge - Find Connections
PassAudited by VirusTotal on May 12, 2026.
Overview
Type: OpenClaw Skill Name: claw-clawbridge Version: 1.0.1 The OpenClaw skill 'claw-clawbridge' is designed for lead generation and networking, requiring web access tools (`web_search`, `web_fetch`, `browser`) to scout for potential connections. While these tools inherently carry risk, the skill's `SKILL.md` and `README.md` explicitly instruct the AI agent to adhere to strict security best practices, including 'Keep secrets out of prompts', 'Use strict tool allowlists', and crucially, 'Human-in-the-loop - NEVER auto-send outreach'. The various prompt files (`prompts/`, `venues/`) reinforce these by detailing rate limiting, privacy considerations ('Only use publicly visible information'), and avoiding spam. There is no evidence of intentional malicious behavior, obfuscation, or prompt injection with a harmful objective; instead, the instructions actively guide the agent towards secure and transparent operation aligned with its stated purpose.
Findings (0)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The agent may perform multiple public searches and page fetches to build candidate profiles.
The skill directs the agent to use web tools across public sites. This is central to the stated scouting purpose and is paired with budgets and human review, but users should understand the browsing scope.
`web_search` | Discover candidate pages ... `web_fetch` | Extract page content ... `browser` | JS-heavy sites
Set conservative run budgets, confirm target venues, and keep browser access limited to scouting tasks.
Following the examples blindly could install tools or clone from an unintended source.
The installation documentation includes user-run global CLI/daemon commands and placeholder GitHub URLs. These are setup examples rather than hidden execution, but provenance should be verified.
npm install -g clawhub ... git clone https://github.com/YOUR_USERNAME/clawbridge-skill.git ... openclaw onboard --install-daemon
Install only from the official ClawHub/OpenClaw sources and do not use placeholder GitHub URLs without replacing them with a trusted repository.
Connection reports may contain personal/professional contact information and business context.
The output schema can store candidate contact details, evidence links, and generated outreach drafts. This is expected for connection briefs, but the stored information can affect later decisions if over-trusted or retained too broadly.
"email": { "format": "email", "description": "Email address if publicly available" } ... "evidence_urls" ... "suggested_intro"Store reports in an appropriate workspace, limit sharing, delete stale briefs, and verify evidence before acting on recommendations.
A human could send AI-assisted outreach that does not disclose how the candidate was found.
The skill drafts personalized outreach and advises not to mention AI discovery, while also requiring human approval. This is not hidden auto-sending, but it affects how recipients understand the interaction.
❌ Mentioning that an AI found them ... These messages are drafts for human review. Do not send without approval.
Review and edit all drafts, avoid misleading personalization, and follow applicable outreach, privacy, and platform rules.
If paired with a scheduler, the agent may repeatedly run web searches and generate new briefs without a fresh prompt each time.
The skill describes ongoing nightly operation. This persistence is disclosed and aligned with the product description, but any scheduled automation should be explicitly controlled by the user.
persistent, nightly scouting operation ... Every night, the agent scours Moltbook, professional communities, and the open web
Enable nightly runs only intentionally, define stop conditions and budgets, and periodically review whether the scouting profile is still appropriate.