Back to skill

Security audit

jubensha-office-murder-case

Security checks for vulnerabilities and agentic risk

Overview

Scanners rate this skill suspicious because they report unsafe file-path handling, access beyond declared permissions, and behavior outside its stated murder-mystery game purpose.

The verdict is suspicious, with high overall confidence, and Skillspector's analysis is partial: it inspected 19 of 21 components, and some instruction checks were limited by obfuscated text. Before installing, consider whether you are comfortable with the reported save/reset path risks, access to shared credentials, and sending gameplay requests to a configured AI provider. The results do not establish malicious intent, but they also do not support treating the skill as safe.

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
scripts/game.py
Finding

Scenario metadata can redirect save writes and resets outside the scenario directory

Content
View full analysis

Impact

A value such as ../victim redirects the destination to ~/victim/state.json. An absolute directory also discards the intended base directory. Resetting the imported scenario can therefore delete another application's or scenario's state.json; subsequent saves can replace it with game state. The destination filename is constrained to state.json, not arbitrary filenames.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (53)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill requires the host to append unchanged commercial promotion, including services unrelated to the requested game. Although the branding is disclosed in the documentation, mandatory verbatim reproduction lets skill-provided content dictate part of the host's response beyond gameplay; no data transmission to the promotional website is demonstrated.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL.md (reported line 237)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_cn.md (reported line 239)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ 只删本剧本 <slug>/state.json,然后重新开局
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ 等价的手动做法
# rm -rf ~/.jubensha                    # ❌ 会删掉其他剧本的存档和 config.json

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL.md (reported line 237)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL.md (reported line 238)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Every Action Returns State Guidance

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_cn.md (reported line 240)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ 只删本剧本 <slug>/state.json,然后重新开局
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ 等价的手动做法
# rm -rf ~/.jubensha                    # ❌ 会删掉其他剧本的存档和 config.json

每一步都会给你提示

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_en.md (reported line 238)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Every Action Returns State Guidance

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL.md (reported line 238)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Every Action Returns State Guidance

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_cn.md (reported line 240)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ 只删本剧本 <slug>/state.json,然后重新开局
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ 等价的手动做法
# rm -rf ~/.jubensha                    # ❌ 会删掉其他剧本的存档和 config.json

每一步都会给你提示

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_en.md (reported line 238)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Every Action Returns State Guidance

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_cn.md (reported line 239)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ 只删本剧本 <slug>/state.json,然后重新开局
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ 等价的手动做法
# rm -rf ~/.jubensha                    # ❌ 会删掉其他剧本的存档和 config.json

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_en.md (reported line 237)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL_en.md (reported line 237)May include surrounding context.

bash
python scripts/game.py --reset          # ✅ Delete only this script's <slug>/state.json and restart
rm -rf ~/.jubensha/jubensha-office-murder-case   # ✅ Equivalent manual reset
# rm -rf ~/.jubensha                    # ❌ Deletes saved games and config.json for every script

Code capability 'file_write' detected in scripts/game.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
97% confidence
Finding

Writing the normal save file is authorized, but this implementation also accepts arbitrary --state paths and constructs the default path from unsanitized scenario metadata rather than the loader's sanitized slug. A malicious scenario can use an absolute or traversal-containing slug to place state writes outside the declared directory, potentially overwriting accessible files with game JSON.

Content

No source excerpt is available for this finding.

Code capability 'network' detected in scripts/llm.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
80% confidence
Finding

The client sends conversation messages and an API credential to a configured network endpoint, but the static report indicates that network access is not declared. Network access is expected for this game's LLM functionality, so this is a permission-transparency gap rather than evidence of malicious exfiltration; the missing declaration cannot be independently verified without the permission manifest.

Content

No source excerpt is available for this finding.

Code capability 'file_read' detected in scripts/load_script.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'file_read' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

Generic investigation requests can activate the game without an explicit gameplay context.

Medium
Category
Trigger Abuse
Confidence
96% confidence
Finding

The activation description includes generic requests to investigate or search for clues without requiring a fictional-game context. A host using this description for routing could divert a legitimate investigation task into gameplay, with irrelevant responses and potentially unnecessary tool or API use.

Content

No source excerpt is available for this finding.

Real-model gameplay requires network requests despite an empty network permission list.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The permission metadata declares no network access, while real-model gameplay explicitly sends requests to a configured API endpoint. This mismatch can mislead users or permission reviewers about whether their game inputs leave the machine, although the documentation does explain remote-model operation.

Content

No source excerpt is available for this finding.

Configuration setup accesses a shared credential file outside the declared filesystem scope.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

Configuration setup reads or writes ~/.jubensha/config.json, outside the sole declared filesystem directory. Because this file contains credentials shared across scenarios, undeclared access affects a more sensitive resource than this game's save file, even though configuration is a legitimate feature.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 96)May include surrounding context.

python scripts/llm.py show # Display the effective configuration with the key masked

text

### Method 2: Write `~/.jubensha/config.json` directly

```json
{

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL_en.md (reported line 96)May include surrounding context.

python scripts/llm.py show # Display the effective configuration with the key masked

text

### Method 2: Write `~/.jubensha/config.json` directly

```json
{

The activation list includes generic investigation and confrontation phrases without restricting them to an active game.

Medium
Category
Trigger Abuse
Confidence
93% confidence
Finding

The activation criteria include generic requests to search for clues, investigate, or confront someone without requiring a game context. This can route unrelated or sensitive real-world requests into a fictional workflow that may persist input and transmit it to an external model.

Content

No source excerpt is available for this finding.

The live-play workflow requires network requests despite declaring no network permissions.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest declares no network permissions, but live play explicitly sends messages to an external LLM endpoint. This mismatch can undermine permission-based review and expose player input to a provider despite an apparent no-network declaration; the supplied documentation does not establish whether the host enforces that declaration.

Content

No source excerpt is available for this finding.

Configuration setup accesses a global credential file outside the permitted filesystem directory.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The documented setup reads and writes ~/.jubensha/config.json, outside the only declared filesystem directory. Because this shared file contains API credentials and affects other scenarios, undeclared access expands the security boundary beyond this game's saves; the documented mode 600 mitigates local exposure but not the permission mismatch.

Content

No source excerpt is available for this finding.

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 · SKILL_cn.md (reported line 96)May include surrounding context.

json
{
  "base_url": "https://api.deepseek.com/v1",
  "api_key": "sk-xxxxxxxx",
  "model": "deepseek-flash"
}