ClawHub
Back to skill

Security audit

Reconnect

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed professional networking workflow with a local tracking helper and explicit-authorization LinkedIn guidance; I found no hidden exfiltration, destructive behavior, or unsafe persistence.

Install only if you are comfortable letting your agent research professional contacts, use browser tools for explicitly approved LinkedIn actions, and keep local, unencrypted project ledgers containing profile URLs, evidence, relationship states, and approval references. Keep those ledgers outside the skill directory and avoid importing private contact exports unless you have permission.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is about networking and LinkedIn-style outreach operations. The actual code does none of that: it does not search for peers, collaborators, mentors, communities, or profiles, nor does it track invitations or reconnect with contacts. Instead, it packages a fixed list of repository files into a deterministic release ZIP and generates checksums after validating file contents. This is a materially different primary purpose and introduces undeclared capabilities related to software release engineering and secret scanning, so the description does not accurately represent the code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents an end-user professional networking skill focused on finding contacts, communities, mentors, and managing outreach. The supplied code does not implement networking discovery, profile resolution, LinkedIn invitation tracking, or outreach logic directly. Instead, it tests release engineering and packaging safeguards for a software project named reconnect. While some demo commands reference candidate import and batch preparation, the code itself is concerned with build reproducibility, installation, archive contents, and privacy-related packaging checks. That is a materially different primary purpose from the declared networking functionality, so this is a clear mismatch.

Content

No source excerpt is available for this finding.

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · references/tracking.md (reported line 30)May include surrounding context.

md
records for historical reconciliation; reservation does not make them actionable.

An explicit `release --batch ID --url URL --reason TEXT` releases an unattempted or
resolved reservation. It refuses unresolved attempts. Release does not erase history
or undo a connection request. `observe` can also record `eligibility` (eligible,
review, excluded), `group` (invited, joined, unknown), or relationship states
(not_contacted, requested, pending, connected, uncertain, withheld, do_not_contact).

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
77% confidence
Finding

The skill documentation directs the agent to read and write project files, use browser/network capabilities, and invoke a helper script, but it does not declare any explicit tool scope or allowed-tools boundaries. This creates unnecessary ambient authority: a host may grant broader file, shell, or network access than the workflow actually needs, increasing the blast radius if the skill is misused or its helper/scripts are modified.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_release.py (reported line 36)May include surrounding context.

python
z.extractall(d / 'unpacked')
            source = d / 'unpacked/reconnect'
            target = d / 'skills/reconnect'
            subprocess.run([sys.executable, str(source / 'scripts/install.py'), '--target', str(target)], check=True, capture_output=True)
            def command(*args):
                p = subprocess.run([sys.executable, str(target / 'scripts/reconnect.py'), '--db', str(d / 'demo.sqlite3'), *args], check=True, capture_output=True, text=True)
                return json.loads(p.stdout)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_release.py (reported line 38)May include surrounding context.

python
target = d / 'skills/reconnect'
            subprocess.run([sys.executable, str(source / 'scripts/install.py'), '--target', str(target)], check=True, capture_output=True)
            def command(*args):
                p = subprocess.run([sys.executable, str(target / 'scripts/reconnect.py'), '--db', str(d / 'demo.sqlite3'), *args], check=True, capture_output=True, text=True)
                return json.loads(p.stdout)
            self.assertEqual(command('import', '--input', str(source / 'examples/candidates.json'))['imported_rows'], 2)
            self.assertEqual(command('prepare', '--batch', 'demo', '--size', '2')['count'], 2)

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The top-level docstring says "No network or invitation sending," which implies the skill does not handle invitation-related actions. However, the CLI exposes an authorize command that updates batch authorization and logs an authorization event, so the documentation understates that the tool manages invitation-authorization workflow metadata.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.