Back to skill

Security audit

Research Logger

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent research-note purpose, but its script handles web, template, and local file data in ways that create meaningful user-safety risks.

Do not install this version for routine use until it escapes template replacements safely, validates fetched URLs, constrains template paths, uses mktemp or private temp storage, fixes the Bear fallback control flow, pins grizzly, and declares all required tools.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (5)

T09 · Insecure Skill Coding Practices

Error
Location
research_logger.sh:92
Finding

Command Injection Through Dynamically Constructed sed Expressions

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
research_logger.sh:34
Finding

Server-Side Request Forgery Through Automatic Fetching of Search Results

Content
View full analysis
/dev/null | head -c 1500 || echo "Summary unavailable.") fi ``` ### Technical Analysis The script automatically passes the first URL returned by `web_search` to `web_fetch`. It performs no validation of: - The URL scheme. - The destination hostname. - The resolved IP address. - Redirect targets. - Loopback, private, link-local, or reserved address ranges. - Cloud instance metadata endpoints. Search results are external data and cannot be assumed trustworthy. A manipulated, poisoned, or attacker-controlled search result could point at a service that is reachable from the Agent environment but not from the attacker. The behavior exceeds minimum privilege because fetching arbitrary search-result destinations is not necessary without validating that the target is a normal public web resource. ### Attack Path 1. The attacker influences the top result for a chosen search query or causes a malicious URL to appear in the search provider's response. 2. The URL targets a loopback service, private-network host, link-local metadata endpoint, or public host that redirects to such a destination. 3. The Skill invokes `web_fetch` automatically, without user confirmation or destination validation. 4. The request originates from the Agent environment and may reach resources unavailable to the attacker. 5. The response is placed into `SUMMARY` and processed as note content. 6. In combination with the unsafe `sed` construction, malicious response text may also reach the command-injection path. ### Impact Assessment The flaw may expose internal HTTP services, local administrative interfaces, private-network resources, or cloud metadata. The exact accessible scope depends on the network permissi ...[truncated 161 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
research_logger.sh:63
Finding

Workspace Escape and Arbitrary Local File Reading Through Template Path Traversal

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
research_logger.sh:107
Finding

Predictable Temporary-File Write Enables Symlink Clobbering and Unintended Data Persistence

Content
View full analysis
/dev/null && \ echo "✅ Research note created in Bear!" || \ echo "❌ Failed to create Bear note. Content saved to /tmp/research_${TOPIC// /_}.md" && \ echo "$NOTE_CONTENT" > "/tmp/research_${TOPIC// /_}.md" ``` ### Technical Analysis The fallback filename is predictable and derived directly from the user-supplied topic. The redirection uses ordinary shell file creation, which follows symbolic links and truncates an existing destination. In addition, the mixed `&&` and `||` expression does not scope the fallback write as intended. These operators have equal precedence and are evaluated left-to-right. Therefore, if `grizzly create` and the success message both succeed, the final `&& echo ... > file` still executes. The note is consequently written to `/tmp` even when Bear creation succeeds. Replacing spaces does not make the topic a safe filename. Slashes and other path-significant characters remain, allowing the topic to affect the resulting path below `/tmp`. ### Attack Path 1. A local attacker predicts the filename associated with a chosen topic. 2. The attacker creates a symbolic link at that `/tmp` path pointing to a file writable by the Skill user. 3. The victim invokes the Skill using the corresponding topic. 4. Due to the left-associative control flow, the final write can occur even if Bear note creation succeeds. 5. Shell redirection follows the symbolic link and truncates or overwrites its target with `NOTE_CONTENT`. A separate privacy path exists without a symlink: generated notes are unnecessarily left in a globally shared temporary directory under predictable names. ### Impact Assessment An attacker may overwrite or corrupt files writable ...[truncated 387 chars]
Remediation
View remediation
"$output_file" fi ``` - Use `mktemp` to obtain an unpredictable, atomically created file. - Prefer a private temporary directory with mode `0700`. - Set a restrictive umask, such as `077`, before creating files containing note content. - Do not derive filesystem paths directly from the topic. - If a human-readable component is required, strictly allowlist safe characters and retain a random suffix. - Avoid persisting the note at all when Bear creation succeeds. - Delete temporary files as soon as they are no longer needed. ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL.md:10
Finding

Unpinned Installation of an Executable Third-Party Dependency

Content
View full analysis
Remediation
View remediation
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 (2)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script performs an additional network/content-retrieval step with web_fetch on the top search result and uses that content to generate a summary, but the declared skill behavior only mentions web search, GIF matching, and note logging. This hidden expansion of data access increases the skill’s effective permissions and can expose users to unanticipated remote content ingestion, privacy issues, or unsafe downstream processing assumptions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill documentation describes automated web search, content fetching, template filling, and Bear note creation, but it does not clearly warn users that untrusted web content and template data will be written into a local Bear note. This can lead to surprising side effects, accidental persistence of misleading or sensitive content, and reduced informed consent for users running the skill.

Content

No source excerpt is available for this finding.