Back to skill

Security audit

Industry News Summary

Security checks for vulnerabilities and agentic risk

Overview

This skill fetches public AI news and saves a dated local summary, with no evidence of deception, exfiltration, or destructive behavior.

Install only if you are comfortable with the skill querying a web search service and saving a dated markdown summary locally. Treat saved summaries as untrusted news data, not agent instructions, and consider clearing or relocating the memory cache if you do not want daily news history retained.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • 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
  • 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 (2)

T02 · Agent Memory Poisoning

Warning
Location
run.sh:11
Finding

Untrusted Web Content Is Persisted in Agent Memory

Content
View full analysis
"$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Topic Overview" >> "$SUMMARY_FILE" echo "$TOPIC_OVERVIEW" >> "$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Key Events" >> "$SUMMARY_FILE" echo "$KEY_EVENTS" >> "$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Potential Impact" >> "$SUMMARY_FILE" echo "$IMPACT" >> "$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Information Sources" >> "$SUMMARY_FILE" echo "$SOURCES" >> "$SUMMARY_FILE" ``` The corresponding persistence behavior is documented in `SKILL.md:48`: ```markdown - Caches summary to `memory/YYYY-MM-DD-ai-summary.md` for future reference ``` ### Technical Analysis The result titles and URLs returned by `web_search` are externally controlled or attacker-influenceable data. The script extracts these fields and writes them directly into a file under the Agent's `memory/` directory without content validation, provenance enforcement, length restrictions, or explicit untrusted-data boundaries. Using `jq` provides JSON encoding and reduces the likelihood of direct shell interpretation, but it does not neutralize natu ...[truncated 1766 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Note
Location
run.sh:6
Finding

Predictable Output Path Allows Symlink-Based File Overwrite

Content
View full analysis
"$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Topic Overview" >> "$SUMMARY_FILE" echo "$TOPIC_OVERVIEW" >> "$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Key Events" >> "$SUMMARY_FILE" echo "$KEY_EVENTS" >> "$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Potential Impact" >> "$SUMMARY_FILE" echo "$IMPACT" >> "$SUMMARY_FILE" echo "" >> "$SUMMARY_FILE" echo "## Information Sources" >> "$SUMMARY_FILE" echo "$SOURCES" >> "$SUMMARY_FILE" ``` ### Technical Analysis The output filename is predictable because it is derived only from the current date. It is also relative to the process's current working directory. Before opening the file with shell redirection, the script does not verify that the `memory` directory and target file are trusted regular filesystem objects rather than symbolic links. The initial `>` redirection opens the resolved target with truncation. On systems where an attacker can modify the working directory or the `memory` directory, the attacker can pre-create `memory/YYYY-MM-DD-ai-summary.md` as a symbolic link. Shell redirection will normally follow that link and truncate or overwrite its destination. The script also lacks restrictive directory creation, ownership checks, atomic file replacement, and no-follow file-opening semantics. ### Attack Path 1. An attacker obtains write access to the directory from which the skill runs, or specifically to its `memory` subdirectory. 2. The attacker predicts the filename from the current date. 3. The attacker creates `memory/YYYY-MM-DD-ai-summary.md` as a symbolic link to another file writable by the skill' ...[truncated 769 chars]
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 (6)

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The skill claims to fetch and summarize public news using only web_search, but it also documents persistent local writes to memory/YYYY-MM-DD-ai-summary.md. Undeclared persistence creates a hidden side effect, can retain user-request-derived data without consent, and breaks the principle of least surprise for a skill that appears read-only.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The declared requirements list only web_search, but the skill also introduces file persistence, which is effectively an additional capability with security and privacy implications. Hidden capability expansion is dangerous because policy engines, sandboxing decisions, or user trust may rely on the declared tool set being complete.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation states the skill only uses web_search, yet the implementation notes add local caching to memory files. This inconsistency can mislead reviewers and users about the skill's actual capabilities, allowing unanticipated data retention in an otherwise low-risk news summarization workflow.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill mentions caching summaries locally but does not warn the user that invoking a simple news summary may create persistent files. Even if the cached content is not highly sensitive, silent persistence can accumulate history, leak usage patterns, and violate user expectations in environments where local writes are restricted or monitored.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The script calls web_search to fetch current news, which is a network operation that may transmit user or system-originated request data. The operation is only described in an internal comment and lacks a user-facing notice or warning in this file.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The script creates and appends to a file under memory/ using shell redirection, which modifies local state. Although comments describe the steps, there is no user-facing prompt, print, or warning that the skill will write a dated markdown file.

Content

No source excerpt is available for this finding.