Back to skill

Security audit

Meeting Quality Scorer

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed meeting-transcript scorer that sends the user-selected transcript to a configured LLM endpoint, with privacy and reliability caveats users should review.

Install only if you are comfortable sending meeting transcripts to the LLM endpoint you configure. Prefer a local or approved provider for sensitive meetings, keep API keys out of shared config files where possible, protect any config file permissions, and be aware the current package appears to be missing report templates needed for the advertised Markdown/HTML output.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

other

Warning
Location
scripts/analyzer.py:7
Finding

Untrusted Transcript Content Can Inject Instructions into LLM Analysis

Content
View full analysis
list[dict]: """Return list of {topic, decided, owner, deadline} dicts.""" msgs = [{"role": "user", "content": _DECISION_PROMPT.format(transcript=transcript)}] result = llm.chat(msgs) if isinstance(result, list): return result return [] def analyze_filler(transcript: str, llm) -> list[dict]: """Return list of {text, reason} filler segments.""" msgs = [{"role": "user", "content": _FILLER_PROMPT.format(transcript=transcript)}] result = llm.chat(msgs) if isinstance(result, list): return result return [] ``` ### Technical Analysis The application interpolates the complete, untrusted meeting transcript directly into the same user message that contains the LLM's operational instructions. There is no separate system message, no robust data boundary, and no explicit instruction that content inside the transcript must n ...[truncated 2151 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
requirements.txt:1
Finding

Unpinned and Unhashed Third-Party Dependencies Create Supply-Chain Exposure

Content
View full analysis
=1.30.0 pyyaml>=6.0 jinja2>=3.1.0 pytest>=8.0.0 ``` The documented installation command is: ```bash pip install -r requirements.txt ``` ### Technical Analysis Every dependency uses an open-ended minimum version constraint. The project provides neither an exact-version lock file nor package hashes. As a result, installation may retrieve any future release satisfying the lower bound from the package index configured in the user's environment. This prevents reproducible installation and means that the reviewed source code does not uniquely determine the third-party code that will execute. A compromised package account, malicious eligible release, compromised package index, or unexpected future dependency change could introduce code that was not part of this audit. The runtime requirements also include `pytest`, although it is only needed for development and testing. Installing unnecessary packages increases the dependency surface without supporting the normal CLI execution path. ### Attack Path 1. The user follows the documented `pip install -r requirements.txt` instruction. 2. pip queries the user's configured package index and resolves the newest releases satisfying the open-ended constraints. 3. A compromised, malicious, or otherwise unsafe eligible package release is selected. 4. The package or one of its transitive dependencies executes installation-time or import-time code. 5. That code runs with the permissions of the user performing the installation or invoking the Skill. This path depends on compromise or malicious modification of a dependency or package source; no malicious dependency is embedded directly in the audited repository. ### Impact Assessment Successful supply-chain compromise co ...[truncated 562 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (30)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The code generally aligns with the meeting-quality analysis theme and does use an LLM-compatible chat interface to analyze transcripts. However, the declared description promises three scoring dimensions, while this code only computes two: decision clarity and time efficiency. There is no logic for participation balance. Additionally, the description states that the skill outputs a Markdown report and an HTML visualization, but this code only returns a raw Python dictionary of decisions, filler segments, and two scores. No undeclared harmful capability is present, but the implemented behavior materially under-delivers relative to the declared functionality, so this is a description-behavior mismatch.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The README explicitly instructs users to send meeting transcripts and API credentials to a configurable external LLM endpoint, with a cloud provider example by default. Meeting transcripts can contain sensitive business or personal information, so this creates a real data exfiltration/privacy risk if users point the tool at third-party services without adequate review.

Content

Scanner excerpt · README.md (reported line 38)May include surrounding context.

1. Environment Variables

bash
export MQS_BASE_URL="https://api.openai.com/v1"
export MQS_API_KEY="sk-..."
export MQS_MODEL="gpt-4o-mini"

Session Persistence

Medium
Category
Rogue Agent
Confidence
88% confidence
Finding

The README advises storing an API key in a persistent plaintext config file under the user's home directory. Persistent credential storage increases the chance of accidental disclosure through backups, misconfigured permissions, shared systems, or local compromise.

Content

Scanner excerpt · README.md (reported line 45)May include surrounding context.

2. Config File

Create ~/.config/meeting-quality-scorer/config.yaml:

yaml
base_url: "https://api.openai.com/v1"

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The sample config file stores both the remote API endpoint and API key for a service that will receive transcript contents. This reinforces a workflow where sensitive meeting data is routinely transmitted off-host, and if the config file is mishandled it can also expose credentials.

Content

Scanner excerpt · README.md (reported line 48)May include surrounding context.

Create ~/.config/meeting-quality-scorer/config.yaml:

yaml
base_url: "https://api.openai.com/v1"
api_key: "sk-..."
model: "gpt-4o-mini"
timeout: 60

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README specifies a degraded-mode banner entirely in Chinese, even though the tool otherwise presents itself as processing multiple languages and does not state that outputs are restricted to Chinese. This creates a language-policy concern because users are not given an opt-in or alternative locale for that user-facing message.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill documents capabilities that require environment access, reading input files, and writing reports, but it does not declare any explicit tool scope or permission boundaries. In agent environments, undeclared capabilities increase the risk of overbroad access, accidental misuse of local data, and execution with more privileges than users expect.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Broad trigger phrases like 'meeting score', 'score this meeting', and 'rate the meeting' can cause unintended invocation in ordinary conversation. In an agent ecosystem, accidental activation may lead to unintentional transcript processing, file access, or transmission of sensitive meeting content to a configured LLM endpoint.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
78% confidence
Finding

Storing API credentials in a persistent config file under ~/.config increases the chance of credential leakage through weak filesystem permissions, backups, shell access, or accidental sharing. While common, persistence of secrets is still a security concern when the file may contain long-lived API keys tied to external services.

Content

Scanner excerpt · SKILL.md (reported line 47)May include surrounding context.

export MQS_MODEL=qwen2.5:72b

text

Or create `~/.config/meeting-quality-scorer/config.yaml`:
```yaml
base_url: https://api.openai.com/v1
api_key: sk-...

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The skill is explicitly designed to send meeting transcripts to a configured LLM endpoint, including potentially external services such as api.openai.com. Meeting transcripts commonly contain sensitive business, personal, or confidential information, so external transmission creates a real data exposure risk if users misconfigure the endpoint or do not understand the privacy implications.

Content

Scanner excerpt · SKILL.md (reported line 49)May include surrounding context.

Or create ~/.config/meeting-quality-scorer/config.yaml:

yaml
base_url: https://api.openai.com/v1
api_key: sk-...
model: gpt-4o-mini

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · config.example.yaml (reported line 1)May include surrounding context.

yaml
base_url: "https://api.openai.com/v1"
api_key: "sk-..."
model: "gpt-4o-mini"
timeout: 60

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The function packages the full meeting transcript into a prompt and sends it to llm.chat, which is a network/LLM transmission of potentially sensitive meeting data. While the code has docstrings describing analysis behavior, there is no explicit user-facing warning, confirmation, or disclosure that transcript contents will be transmitted for external analysis.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This call sends the transcript again for filler/off-topic analysis, creating another transmission of potentially sensitive meeting content. The code does not provide any user-facing notice, confirmation, or explicit warning about this data-sharing behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest says the skill objectively scores meeting quality across three dimensions: decision clarity, time efficiency, and participation balance. This file's orchestration returns only decision_clarity and time_efficiency, with no computation or output for participation balance, so the implemented behavior does not match the claimed scope.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This Python file performs an outbound API call using the OpenAI-compatible client and sends the messages payload to a remote endpoint. While the module docstring identifies it as an LLM client, there is no user-facing print/log, confirmation, or inline warning near the transmission despite the operation potentially sending user or system data off-host.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code generates user-facing recommendation text exclusively in Chinese, which imposes a specific language on all users. The policy for this category applies to all file types and flags language constraints unless the skill offers a language choice or clearly documents a justified locale restriction.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script sends full transcript content to an external LLM service via a configurable OpenAI-compatible endpoint without any user-facing notice, consent flow, or clear disclosure at the point of use. Because meeting transcripts commonly contain sensitive business or personal information, this can lead to unintended data exposure to third-party services, especially when users point the tool at non-local providers.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · tests/test_llm_adapter.py (reported line 20)May include surrounding context.

python
def test_resolve_from_env(self, monkeypatch):
        """Test resolving config from MQS_ prefixed env vars."""
        monkeypatch.setenv("MQS_BASE_URL", "https://api.example.com/v1")
        monkeypatch.setenv("MQS_API_KEY", "test-key-123")
        monkeypatch.setenv("MQS_MODEL", "gpt-4")
        monkeypatch.setenv("MQS_TIMEOUT", "90")

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · tests/test_llm_adapter.py (reported line 27)May include surrounding context.

python
def test_resolve_from_env(self, monkeypatch):
        """Test resolving config from MQS_ prefixed env vars."""
        monkeypatch.setenv("MQS_BASE_URL", "https://api.example.com/v1")
        monkeypatch.setenv("MQS_API_KEY", "test-key-123")
        monkeypatch.setenv("MQS_MODEL", "gpt-4")
        monkeypatch.setenv("MQS_TIMEOUT", "90")

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · tests/test_llm_adapter.py (reported line 111)May include surrounding context.

python
def test_resolve_from_env(self, monkeypatch):
        """Test resolving config from MQS_ prefixed env vars."""
        monkeypatch.setenv("MQS_BASE_URL", "https://api.example.com/v1")
        monkeypatch.setenv("MQS_API_KEY", "test-key-123")
        monkeypatch.setenv("MQS_MODEL", "gpt-4")
        monkeypatch.setenv("MQS_TIMEOUT", "90")

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · tests/test_llm_adapter.py (reported line 147)May include surrounding context.

python
def test_resolve_from_env(self, monkeypatch):
        """Test resolving config from MQS_ prefixed env vars."""
        monkeypatch.setenv("MQS_BASE_URL", "https://api.example.com/v1")
        monkeypatch.setenv("MQS_API_KEY", "test-key-123")
        monkeypatch.setenv("MQS_MODEL", "gpt-4")
        monkeypatch.setenv("MQS_TIMEOUT", "90")

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · tests/test_llm_adapter.py (reported line 171)May include surrounding context.

python
def test_resolve_from_env(self, monkeypatch):
        """Test resolving config from MQS_ prefixed env vars."""
        monkeypatch.setenv("MQS_BASE_URL", "https://api.example.com/v1")
        monkeypatch.setenv("MQS_API_KEY", "test-key-123")
        monkeypatch.setenv("MQS_MODEL", "gpt-4")
        monkeypatch.setenv("MQS_TIMEOUT", "90")

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
96% confidence
Finding

The dependency is specified with a lower-bound version only, which allows future unresolved versions to be installed. This harms build reproducibility and can introduce vulnerable or breaking releases through the supply chain without review.

Content

Scanner excerpt · requirements.txt (reported line 1)May include surrounding context.

text
openai>=1.30.0
pyyaml>=6.0
jinja2>=3.1.0
pytest>=8.0.0

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
98% confidence
Finding

PyYAML is not pinned to a specific version, so installs may resolve to different releases over time. Because this package has a history of security advisories, leaving it unpinned increases uncertainty and supply-chain risk.

Content

Scanner excerpt · requirements.txt (reported line 2)May include surrounding context.

text
openai>=1.30.0
pyyaml>=6.0
jinja2>=3.1.0
pytest>=8.0.0

Unverifiable Dependency: pyyaml has 8 known advisory(ies) (CVE-2019-20477 (Deserialization of Untrusted Data in PyYAML); CVE-2020-1747 (Improper Input Validation in PyYAML); CVE-2020-14343 (Improper Input Validation in PyYAML) +5 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
95% confidence
Finding

The manifest does not pin PyYAML, so it is impossible to verify whether deployed environments will receive a version affected by known advisories. Given PyYAML's history of deserialization-related issues, this uncertainty is a legitimate supply-chain security concern.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
97% confidence
Finding

Jinja2 is declared with a minimum version instead of an exact version, so dependency resolution may pull in unexpected releases. In a skill that generates HTML, templating libraries are security-relevant, making version drift a meaningful risk factor.

Content

Scanner excerpt · requirements.txt (reported line 3)May include surrounding context.

text
openai>=1.30.0
pyyaml>=6.0
jinja2>=3.1.0
pytest>=8.0.0