Back to skill

Security audit

planning-and-task-br

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a task-planning helper, but it asks for file reading and command execution while also giving broad operational guidance that is not well scoped.

Review this before installing. Use it only in a host that constrains file reads and command execution, requires approval for commands, and keeps credentials out of prompts and task files. Do not run the agent as administrator for this skill.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:16
Finding

Overprivileged File-Read and Command-Execution Capabilities

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 16-18; related elevated-execution guidance at lines 214-216 and 237-239
Vulnerability Type: Excessive tool permissions and unsafe privilege guidance
Risk Level: Medium

Vulnerable Code

yaml
tools:
- read
- exec

The troubleshooting guidance also includes:

markdown
| 权限不足 | 当前用户无读写权限 | 检查文件权限,以管理员身份运行 |

This guidance is duplicated later in the document.

Technical Analysis

The Skill describes itself as a planning and task-breakdown assistant. That purpose can ordinarily be fulfilled through language-model reasoning and does not inherently require unrestricted local file reading or command execution. Declaring both read and exec therefore violates the principle of least privilege unless the host independently constrains those tools.

The document defines no file-scope restrictions, command allowlist, argument validation, execution sandbox, confirmation requirement, or prohibition against processing instructions embedded in untrusted task content. In addition, its troubleshooting guidance recommends running with administrator privileges when access is denied. Elevation is not an appropriate generic response to a permission failure because it can turn an unsafe or incorrectly generated command into a system-wide operation.

This package contains no executable scripts, and the reviewed text does not explicitly direct the agent to run a particular malicious command. The risk arises from the unnecessary capabilities requested by the Skill and the possibility that untrusted planning content could influence their use in a permissive Agent host.

Attack Path

  1. A user installs or loads the Skill on an Agent platform that grants the tools declared in SKILL.md.
  2. The platform grants local file-read and command-execution access to the Skill.
  3. An attacker supplies task content, a specification, or a project file containing i ...[truncated 1518 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove read and exec from the tool declaration when the Skill is used only for planning and task decomposition.
  2. If file access is essential, limit reads to paths explicitly selected by the user and confined to the active project workspace. Deny access to home directories, credential stores, SSH material, environment files, and operating-system paths by default.
  3. If command execution is essential, replace unrestricted execution with narrowly scoped operations or an explicit command allowlist. Validate executable names and arguments independently rather than constructing shell command strings from user-controlled content.
  4. Run commands in a sandbox with minimal filesystem and network access, a non-privileged account, resource limits, and strict timeouts.
  5. Require explicit user confirmation that displays the exact command, arguments, working directory, and expected side effects before execution.
  6. Treat specifications, project files, and task content as untrusted data. Explicitly prohibit following embedded instructions that request unrelated tool use, sensitive-file access, credential disclosure, or privilege changes.
  7. Remove the recommendation to run as an administrator. Replace it with guidance to diagnose ownership and permissions, grant only the minimum required access, or ask the user to approve a narrowly scoped change.
  8. Document the legitimate reason for every requested tool and define enforceable boundaries for paths, commands, network access, and execution privileges.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (5)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The activation language is extremely broad, covering generic efficiency improvement, automation, batch processing, and workflow optimization. Overly broad triggers can cause the agent to select this skill in unrelated contexts, effectively granting a loosely scoped skill access in situations where a narrower, safer tool should have been used.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill advertises GitHub, Jira, and API integration even though its stated function is task breakdown. This broadens the skill's effective access to external systems and data, creating opportunities for unintended data exposure, over-collection, or actions in third-party services without clear necessity or user understanding.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill presents itself as a planning/task-breakdown tool, but its documentation also advertises command execution, API calls, file handling, and multi-interface data aggregation. This mismatch expands the apparent authority of the skill beyond its stated purpose, increasing the chance that an agent or user will invoke it in higher-risk operational contexts without appropriate scrutiny or consent.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Documenting command execution in a task-breakdown skill is dangerous because it normalizes system-impacting behavior where users would expect only analysis and planning. In an agent environment, this can lead to unanticipated shell access or execution of harmful commands under the guise of a low-risk planning tool.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The markdown instructs users to configure API keys, call interfaces, prepare files, and execute commands, but it does not provide clear warnings about credential handling, system modification, network effects, or other operational risks. In practice, this can mislead users into treating a high-impact skill as harmless documentation-driven automation, increasing the likelihood of unsafe execution and secret exposure.

Content

No source excerpt is available for this finding.