ClawBridge - Find Connections

Runs nightly searches to identify and rank relevant candidates matching your offer and ask, delivering evidence-backed connection briefs for human review.

Security Scan

This is an instruction-only OpenClaw skill that is internally coherent: it describes web-based scouting, uses only the declared web tools, asks for no credentials, and contains sensible safeguards (human-in-loop, rate limits, avoid lists).

Purpose & Capability

The skill's stated purpose (nightly scouting to find connection opportunities and draft outreach) lines up with what it requires and describes: web_search, web_fetch and browser for JS-rendered pages, output JSON/Markdown, and local run/run budget. There are no unrelated credentials, binaries, or config paths requested. Minor note: package.json version (1.0.0) differs from registry version (1.0.1) and the top-level registry description is missing — these are quality issues but not functional mismatches.

Instruction Scope

SKILL.md and the prompt files explicitly confine actions to web discovery, fetching public pages, filtering, ranking, and drafting. The instructions call out privacy and safety (respect robots.txt, avoid closed/ private communities, keep secrets out of prompts, honor avoid_list, human approval required). It does collect and persist public contact evidence (including emails if publicly available) which is consistent with its purpose — users should be aware of privacy and anti-spam/legal constraints before enabling runs.

Install Mechanism

This is instruction-only (no install spec); install guidance references standard CLIs (clawhub/openclaw) and cloning from GitHub with placeholders. No downloads from arbitrary URLs or archive extraction are present. Installing recommended CLIs (npm global installs) is normal for this ecosystem.

Credentials

The skill declares no required environment variables, no primary credential, and no config paths. The README shows optional env knobs (CLAWBRIDGE_TOP_K, CLAWBRIDGE_RECENCY_DAYS) which are proportionate. There are no requests for unrelated secrets or multiple external credentials.

Persistence & Privilege

always:false and default model invocation settings are appropriate. The skill does write output to local files (run.json/run.md) which is expected. It does not request persistent elevated privileges or modifications to other skills/configs.

Assessment

This skill is internally consistent and appears to do what it claims: scour public web venues, validate evidence, rank candidates, and produce draft messages for human review. Before installing, consider: 1) Confirm you want an agent that will fetch public pages and potentially extract publicly listed contact info — ensure this aligns with your privacy policy and anti-spam/legal requirements in your jurisdictions. 2) Configure run_budget, avoid_list, and constraints to limit scraping and rate usage; enable the platform tool allowlist so only web_search/web_fetch/browser are available to this skill. 3) Keep the human-in-the-loop requirement (do not enable auto-send actions) and review outputs before outreach. 4) Note small quality issues (version mismatch in package.json and missing registry description/homepage) — these don't indicate maliciousness but you may want to verify the skill source (the GitHub placeholders) before trusting scheduled runs.

claw-clawbridge

The Intelligent Connection Bridge: A high-signal scouting agent that runs nightly to bridge you with the right people.

Overview

Clawbridge transforms a simple human prompt into a persistent, nightly scouting operation. It doesn't just find leads; it builds a bridge between your goals and the people who can help you achieve them.

1. Human Intent: You define what you offer and who you're looking for once.
2. Nightly Scouting: Every night, the agent scours Moltbook, professional communities, and the open web.
3. Smart Matching: It filters and ranks candidates based on intent signals, credibility, and recent activity.
4. Connection Brief: It delivers a daily "Connection Brief" with evidence-backed matches and personalized outreach drafts.
5. Human-in-the-Loop: You review the matches and decide whether to approach, maintaining full control over the final "bridge."

Installation

Via ClawHub (Recommended)

# Install the ClawHub CLI
npm install -g clawhub

# Install this skill
clawhub install claw-clawbridge

Via Legacy clawdbot CLI

# From registry
clawdbot skills install claw-clawbridge

# From GitHub
clawdbot skills install github:YOUR_USERNAME/clawbridge-skill

Manual

Clone and copy to your OpenClaw workspace:

git clone https://github.com/YOUR_USERNAME/clawbridge-skill.git ~/.openclaw/workspace/skills/claw-clawbridge
openclaw gateway restart

Inputs

The skill requires the following inputs:

1. Project Profile (required)

offer: "What your agency/company offers"
ask: "What you want (partners, clients, co-marketing, advisors)"
ideal_persona: "Exact target persona(s)"
verticals:
  - "keyword1"
  - "keyword2"
  - "keyword3"
geo_timezone: "optional - geographic/timezone preferences"
disallowed:
  - "do not contact constraints"
tone: "Short style guidance for draft messages"

2. Constraints (optional)

no_spam_rules:
  - "No cold outreach to competitors"
  - "Respect unsubscribe requests"
regions:
  - "US"
  - "EU"
avoid_list:
  - "[e-mail]"
  - "@spam_account"

3. Targets (optional)

venues:
  - "moltbook"
  - "web"
  - "communities"
query_templates:
  - "{vertical} + hiring + partner"
  - "{vertical} + looking for + {ask}"

4. Run Budget (optional)

max_searches: 20
max_fetches: 50
max_minutes: 10

Tools Used

This skill uses the following OpenClaw tools:

Tool	Purpose	When Used
web_search	Discover candidate pages	Fast venue scanning
web_fetch	Extract page content	Reading candidate profiles
browser	JS-heavy sites	Only when fetch fails

Security Requirements

⚠️ MUST follow these security defaults:

1. Keep secrets out of prompts - Pass via env/config only
2. Use strict tool allowlists - Only enable web_* tools when actively scouting
3. Human-in-the-loop - NEVER auto-send outreach in MVP
4. Rate limiting - Respect run budget constraints
5. Avoid list enforcement - Never contact entries in avoid_list

Execution Flow

┌─────────────────────────────────────────────────────────────────┐
│                     DISCOVERY PHASE                             │
│  ┌──────────┐    ┌──────────┐    ┌──────────┐                  │
│  │web_search│───▶│ Filter   │───▶│ Dedupe   │                  │
│  │ (venues) │    │ Results  │    │ & Queue  │                  │
│  └──────────┘    └──────────┘    └──────────┘                  │
└─────────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────────┐
│                     ENRICHMENT PHASE                            │
│  ┌──────────┐    ┌──────────┐    ┌──────────┐                  │
│  │web_fetch │───▶│ Extract  │───▶│ Validate │                  │
│  │ (pages)  │    │ Signals  │    │ Evidence │                  │
│  └──────────┘    └──────────┘    └──────────┘                  │
└─────────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────────┐
│                     RANKING PHASE                               │
│  ┌──────────┐    ┌──────────┐    ┌──────────┐                  │
│  │  Score   │───▶│  Rank    │───▶│  Top K   │                  │
│  │ Heuristic│    │  Sort    │    │ Selection│                  │
│  └──────────┘    └──────────┘    └──────────┘                  │
└─────────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────────┐
│                     DRAFTING PHASE                              │
│  ┌──────────┐    ┌──────────┐    ┌──────────┐                  │
│  │  Draft   │───▶│  Review  │───▶│  Output  │                  │
│  │ Messages │    │  Tone    │    │  Brief   │                  │
│  └──────────┘    └──────────┘    └──────────┘                  │
└─────────────────────────────────────────────────────────────────┘

Output

The skill outputs a Connection Brief in two formats:

1. Structured JSON (run.json)

See schema/connection_brief.json for the full schema.

2. Human-Readable Markdown (run.md)

See examples/sample_run.md for a sample report.

Candidate Selection Rules

Hard Requirements (discard if missing)

• ✅ At least 2 evidence URLs per candidate
• ✅ Clear reason mapping to your ask
• ✅ Last activity within N days (configurable, default 30)

Risk Flags

Candidates are flagged if they exhibit:

• 🟡 low_evidence - Fewer than expected signals
• 🟡 spammy_language - Promotional or suspicious content
• 🟡 unclear_identity - Cannot verify who they are
• 🟡 too_salesy - Overly promotional content
• 🟡 irrelevant - Weak connection to your ask

Ranking Heuristic (v1)

Each candidate is scored on:

Factor	Weight	Description
Relevance	30%	Match to keywords + ask
Intent	25%	Actively building/hiring/seeking
Credibility	20%	Consistent footprint across sources
Recency	15%	Recent activity signals
Engagement	10%	Mutual interests/communities

Output: Top K candidates (default K=3, configurable 5-10)

Examples

See the examples/ directory for:

• sample_run.json - Full JSON output example
• sample_run.md - Human-readable report example

Prompts

The skill uses modular prompts located in prompts/:

• discovery.md - How to search for candidates
• filtering.md - How to apply hard requirements
• ranking.md - How to score and rank candidates
• drafting.md - How to write outreach messages

Venues

Venue-specific search strategies are in venues/:

• moltbook.md - Moltbook platform scouting
• web.md - General web search strategies
• communities.md - Community/forum discovery

Configuration

Environment Variables

# Optional: Override defaults
CLAWBRIDGE_TOP_K=5                    # Number of candidates to return
CLAWBRIDGE_RECENCY_DAYS=30           # Activity recency threshold
CLAWBRIDGE_MAX_SEARCHES=20           # Max search queries per run
CLAWBRIDGE_MAX_FETCHES=50            # Max page fetches per run

Workspace Configuration

The skill reads workspace config from the runner or vault:

workspace_id: "ws_abc123"
workspace_token: "tok_..."  # For vault uploads
delivery_target: "discord"  # or "slack" or "email"

License

MIT License - See LICENSE file for details.

Contributing

Contributions welcome! Please read the prompts carefully and ensure any changes maintain:

1. Deterministic output schema
2. No secrets in prompts
3. Human-in-the-loop requirement
4. Evidence-based candidate selection