Back to skill

Security audit

Geo Deep Imitate

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent for GEO content generation, but it needs review because it persists a GEO API key in plaintext and uses undeclared shell/API operations.

Review before installing. Use this only if you intend the agent to contact the GEO SaaS service with your API key, submit source material and task metadata there, and store exports locally. Prefer a platform secret manager or manually protect ~/.qclaw/geo-api-key with owner-only permissions, and rotate the key if it may have been exposed.

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:23
Finding

API Key Stored Without Restrictive File Permissions

Content
View full analysis
" > ~/.qclaw/geo-api-key ``` ### Technical Analysis The Skill instructs the agent to persist a user-provided GEO API key as plaintext in `~/.qclaw/geo-api-key`. It does not securely create the parent directory, set an explicit restrictive umask, or enforce file permissions such as mode `0600`. The resulting permissions therefore depend on the existing directory permissions and the process umask. In an environment with a permissive umask or improperly protected home directory, other local users or processes may be able to read the credential. Persistent storage is necessary for the declared workflow, but leaving access control to ambient filesystem settings exceeds the minimum safe privilege model. The credential should be accessible only to its owner. ### Attack Path 1. The Skill asks the user for a GEO API key. 2. The key is written in plaintext to `~/.qclaw/geo-api-key`. 3. The file inherits permissions determined by the current umask, without an explicit owner-only permission check. 4. A local user or process with filesystem access reads the key from that file. 5. The attacker submits the stolen key as a Bearer token to the GEO service API. 6. The attacker gains whatever account and workflow capabilities are authorized by that key, until it is revoked or expires. This attack requires local filesystem access and a permission configuration that exposes the credential. ### Impact Assessment Successful exploitation could disclose the GEO API key and allow unauthorized use of the associated service account. The accessible scope is limited to the permissions granted to that key, but may include reading task metadata, generating articles, retrieving exports, or submitting workflow status reports. No evidence was found ...[truncated 354 chars]
Remediation
View remediation
"$HOME/.qclaw/geo-api-key" chmod 600 "$HOME/.qclaw/geo-api-key" ``` Additional hardening should include: - Prefer an operating-system credential store or secret manager instead of a plaintext file. - Validate that the credential file is a regular file owned by the current user before reading it. - Reject symbolic links to prevent writing to or reading from an unintended path. - Avoid placing the key directly in shell command history or process arguments. - Never print the key in logs, diagnostics, or error messages. - Document key revocation and rotation procedures. - Restrict `web_fetch` source URLs to HTTP or HTTPS and reject loopback, private, link-local, and cloud metadata addresses to reduce server-side request forgery risk from API-supplied URLs. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest declares only the web_fetch tool, but the documented workflow performs direct authenticated API calls, key verification, export downloads, and local filesystem writes. This mismatch obscures the skill's true capabilities and trust boundary, making it easier for operators or downstream agents to invoke external actions and data transfers they did not expect.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill instructs reading API credentials from user home-directory files and, if absent, asks the user for a key and persists it to disk. Storing long-lived secrets in plaintext under predictable paths increases the chance of credential disclosure through other tools, backups, logs, or local compromise, and this behavior is broader than the core content-generation task itself.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill tells the user to enter an API key and save it directly into a local file without any warning that the secret will persist on disk. This can cause users to expose credentials unintentionally, especially on shared systems or in environments where home directories are synced, backed up, or readable by other processes.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
78% confidence
Finding

The verify-key call is an external transmission of a bearer token to a remote service. While such verification can be legitimate, it still expands the attack surface and can expose credential usage patterns or leak secrets if the endpoint, transport, or surrounding environment is untrusted.

Content

Scanner excerpt · SKILL.md (reported line 31)May include surrounding context.

验证密钥(信源/成稿/导出前必做)

bash
curl -s -X POST "$BASE/api/geo/verify-key" \
  -H "Authorization: Bearer $GEO_KEY"

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The workflow sends authenticated requests and later transmits task IDs, source URLs, reference content, outlines, and export/reporting metadata to an external service without explicit user-facing notice of the data flow. In a content-production context, this may disclose proprietary source material, operational identifiers, or generated drafts to a third party beyond what a user expects from a web-fetch-only skill.

Content

No source excerpt is available for this finding.