Back to skill

Security audit

tech-dynamics

Security checks for vulnerabilities and agentic risk

Overview

The skill claims to provide current tech-company briefings, but it outputs unsupported hard-coded news claims while making outbound web requests and using an unsafe temporary file.

Review before installing. Do not rely on this skill for business, financial, regulatory, or reputational decisions unless it is changed to derive every claim from current sources with citations. Treat its current output as an unreliable mock briefing, and be aware that search terms are sent to TechCrunch when a query is provided.

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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
execute.sh:32
Finding

Predictable Temporary File Permits Symlink-Based File Truncation

Content
View full analysis
/tmp/tech_dynamics.tmp echo "### Topic Overview Latest news for $1 --- ### Key Updates - Trends from past 7 days - Key decisions from global tech companies --- ### Potential Impact/Risks - May influence industry direction - Possible regulatory scrutiny --- ### Information Sources - TechCrunch (https://techcrunch.com) --- " rm /tmp/tech_dynamics.tmp ``` ### Technical Analysis The script writes network response data to the fixed, globally predictable path `/tmp/tech_dynamics.tmp`. Shell output redirection opens this path without securely creating it, checking its ownership, or preventing symbolic-link traversal. On systems where applicable symbolic-link protections are absent, disabled, or bypassed by the execution context, another local user can pre-create that path as a symbolic link. When the script executes, the shell follows the link and opens the target with truncation semantics before `curl` runs. Any linked regular file writable by the account running the skill can consequently be emptied or overwritten with filtered response content. The subsequent `rm` does not repair damage to the linked target. Depending on sticky-directory protections and file ownership, removal may fail, but target truncation may already have occurred. Concurrent executions can also interfere with each other because every process uses the same pathname. The temporary file is never read, so its creation provides no functional benefit. ### Attack Path 1. A local attacker predicts the fixed pathname `/tmp/tech_dynamics.tmp`. 2. Before the victim invokes the skill, the attacker creates that pathname as a symbolic link to a file writable by the victim account. 3. The victim runs `execute.sh` with a query argume ...[truncated 883 chars]
Remediation
View remediation
"$tmp_file" ``` ]]>

other

Warning
Location
execute.sh:5
Finding

Fetched News Is Discarded While Unsupported Claims Are Presented as Current Results

Content
View full analysis
/tmp/tech_dynamics.tmp echo "### Topic Overview Latest news for $1 --- ### Key Updates - Trends from past 7 days - Key decisions from global tech companies --- ### Potential Impact/Risks - May influence industry direction - Possible regulatory scrutiny --- ### Information Sources - TechCrunch (https://techcrunch.com) --- " rm /tmp/tech_dynamics.tmp fi ``` `SKILL.md` documents the same pattern: data is fetched and supplied to a loop, but the loop variable is not used and fixed claims are printed instead. ### Technical Analysis The skill describes itself as retrieving and summarizing recent technology-company news. Its implementation does not perform that operation: - Without an argument, it prints specific hard-coded claims about acquisitions, leadership changes, regulatory matters, and financial exposure without retrieving or valida ...[truncated 1937 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • 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 (8)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared behavior says the skill retrieves and summarizes latest public web updates, but the implementation discards fetched data and emits hard-coded claims such as acquisitions, leadership changes, and regulatory actions. This is dangerous because consumers may trust fabricated business intelligence as current fact, leading to misinformation, flawed decisions, and reputational or compliance harm.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The top-level description promises retrieval and summarization of latest updates, but the implementation only superficially fetches pages and returns predetermined briefings. In context, this mismatch is dangerous because the skill is positioned as a current-events intelligence source, so users are likely to rely on false, stale, or invented claims as real reporting.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The 'Structured Briefing' example performs a network request but ignores the returned content, then prints a fixed briefing containing specific purported events. This creates authoritative-looking but unverified output, which is especially risky in a news-monitoring skill where users expect timeliness and factual accuracy from external sources.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The company-specific example claims to generate an Apple M&A briefing from search results, but it always outputs the same acquisition narrative regardless of the fetched page. That can directly mislead users into believing a transaction exists or is imminent, amplifying false information in a high-trust workflow.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
95% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · execute.sh (reported line 48)May include surrounding context.

sh
- TechCrunch (https://techcrunch.com)
---
"
  rm /tmp/tech_dynamics.tmp
fi

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill invokes shell commands (curl, grep, head, while read) but does not declare any tool scope such as permissions or allowed-tools. This weakens execution governance and can cause the agent to run network-capable shell operations without explicit user-visible authorization boundaries, increasing the chance of unintended external requests or abuse if the skill is reused or extended.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This code silently fetches data from an external website using curl, writes output to /tmp/tech_dynamics.tmp, and then deletes that file. Aside from a brief code comment, there is no user-facing prompt, warning, or runtime disclosure that the script will contact a remote service and modify local filesystem state.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The examples send search queries to external websites but do not clearly disclose that user-provided terms will be transmitted off-platform. While lower severity than the fabrication issues, this still poses a privacy and transparency problem, especially if users submit sensitive company names, deal hypotheses, or investigative topics.

Content

No source excerpt is available for this finding.