Back to skill

Security audit

Incident Response Network

Security checks for vulnerabilities and agentic risk

Overview

This network forensics skill has a valid incident-response purpose, but it is labeled read-only while directing users to run packet captures and write sensitive evidence files on network devices.

Install only for authorized network incident response use. Treat packet captures, support bundles, and saved configurations as sensitive evidence: get approval before collection, use narrow filters and time or packet limits, monitor device health, transfer files to approved encrypted evidence storage, and remove temporary on-device copies according to your retention policy.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:13
Finding

Misleading Read-Only Safety Classification for Stateful Evidence Collection

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:13-16, 29-32, 103-116, 137-147; references/cli-reference.md:20-34, 138-149
Vulnerability Type: Safety metadata and operational behavior mismatch
Risk Level: Medium

The Skill declares itself read-only:

yaml
metadata:
  safety: read-only
  author: network-security-skills-suite
  version: "1.0.0"
  openclaw: '{"emoji":"🚨","safetyTier":"read-only","requires":{"bins":["ssh"],"env":[]},"tags":["incident","forensics","network"],"mcpDependencies":[],"egressEndpoints":[]}'

It also makes the following behavioral claim:

text
All commands are read-only. Containment verification confirms that previously applied
controls are effective — it does not execute containment actions.

However, the prescribed evidence-collection procedure starts capture processes and writes files to device storage:

text
**1b. Active packet captures** (real-time — exists only while traffic flows):

If the incident is active and the investigation requires payload-level
evidence, initiate packet captures on relevant interfaces immediately.

- **[Cisco]** — `monitor capture CAP1 interface <intf> both` then
  `monitor capture CAP1 start` — export with
  `monitor capture CAP1 export flash:evidence.pcap`
- **[JunOS]** — `monitor traffic interface <intf> write-file /var/tmp/capture.pcap`
- **[EOS]** — `bash tcpdump -i <intf> -w /mnt/flash/evidence.pcap -c 10000`

The procedure additionally creates technical-support and configuration artifacts:

text
**1e. Device configuration and comprehensive state:**

Save the running configuration and full technical support output for each
device in scope:

- **[Cisco]** — `show tech-support | redirect flash:tech-<hostname>-<date>.txt`
- **[JunOS]** — `request support information | save /var/tmp/tech-<hostname>-<date>.txt`
- **[EOS]** — `show tech-support | redirect flash:tech-<hostname>-<date>.txt`

T ...[truncated 3949 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the read-only safety tier with a classification that explicitly permits stateful forensic collection and device-local file creation.
  2. Separate the procedure into two clearly labeled modes:
    • Passive inspection using commands that only display existing state.
    • Authorized evidence collection that starts captures or writes files.
  3. Require explicit operator approval before starting packet capture, invoking a device shell, or writing evidence to device storage.
  4. Default to narrowly scoped capture filters instead of match any. Require an approved interface, host or subnet filter, protocol filter, packet limit, duration limit, and maximum file size.
  5. Add preflight checks for available storage, CPU utilization, device role, expected traffic rate, and capture-platform limitations.
  6. Define an emergency stop procedure and objective resource thresholds at which collection must be terminated.
  7. Require restrictive access controls for evidence files and transfer them through an approved encrypted channel to controlled evidence storage.
  8. Document retention and secure-cleanup requirements for copies left on device flash or temporary filesystems after hash verification and transfer.
  9. Warn that packet captures, running configurations, and support bundles may contain credentials, payload data, topology details, and security policy.
  10. Update the statement “All commands are read-only” to accurately distinguish containment verification from evidence-collection side effects.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (10)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The front-matter labels the skill as 'read-only' even though later commands redirect output to flash, save support bundles, and write packet capture files. Safety metadata is often consumed programmatically, so inaccurate labeling can bypass safeguards or cause an agent to approve operations that create files or impact device resources.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The document-wide assertion that all commands are read-only is contradicted by later instructions that initiate captures and persist evidence to flash or temporary storage. In an incident-response context, this inconsistency is dangerous because responders may rely on the claim when operating under tight change-control or forensic-preservation constraints.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly states that all commands are read-only, but the documented workflow includes starting packet captures and exporting artifacts to local device storage. That mismatch can cause operators or downstream agents to execute state-changing actions under false safety assumptions, including disk writes and capture activation on production devices.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The packet capture guidance writes PCAPs locally and may collect full payloads, but it does not prominently warn that the artifacts can contain sensitive communications, credentials, secrets, or regulated data. Without explicit disclosure and handling guidance, operators may over-collect, store evidence insecurely, or violate data-minimization and legal requirements.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Support bundles and evidence captures can include sensitive configuration details, topology, SNMP strings, user/account data, and other credential-adjacent material, yet the skill does not warn about this exposure. During incidents, these artifacts are often widely shared; failing to flag their sensitivity increases the chance of secondary data leakage.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document states commands are read-only unless explicitly noted, but multiple commands later in the file change session settings, create captures, configure monitor sessions, or save files without being consistently flagged as non-read-only. This mismatch can mislead responders into running state-changing commands under the assumption they are safe, increasing the chance of unintended operational impact or evidence-handling mistakes during a live incident.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Packet capture commands can collect full payloads that may include credentials, cookies, personal data, business content, or regulated information, yet the reference lacks an explicit warning about this sensitivity. In an incident-response context, omission of this warning raises the risk of overcollection, mishandling, and legal or policy violations even if the commands are technically valid.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The reference includes active packet-capture and mirroring commands that are operationally invasive and not purely read-only, despite the skill description emphasizing read-only containment verification and network artifacts collection. These commands can increase CPU load, alter device state, and collect full packet contents containing credentials, tokens, or user data, creating both service-risk and privacy exposure if used without explicit guardrails.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The evidence preservation section instructs writing technical support bundles and configuration snapshots to persistent device storage, which exceeds a read-only verification scope and modifies device state. These files may contain secrets, topology details, keys, SNMP communities, and other sensitive information, and writing them locally can create residual exposure or consume constrained storage during an incident.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Saving tech-support bundles and configuration snapshots to persistent storage can create files containing highly sensitive operational and security data without any explicit warning to the user. In incident response, such artifacts often include credentials, community strings, internal IP space, device identifiers, and historical logs, so the lack of handling guidance materially increases disclosure risk.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.