Back to skill

Security audit

Skill Discoverer

Security checks for vulnerabilities and agentic risk

Overview

The skill’s recommendation behavior is coherent, but it asks for file-write and command-execution powers that its code does not need.

Review this before installing because the skill currently requests command execution and write access even though its code only needs to search and recommend skills. Prefer installing only after those permissions are reduced, and update the dev dependency lockfile if you plan to run its tests or CI.

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:11
Finding

Unnecessary Command Execution and Filesystem Write Permissions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:11-15
Vulnerability Type: Excessive agent tool permissions
Risk Level: Medium

yaml
tools:
  - Read
  - Write
  - Bash
  - Exec

Technical Analysis

The Skill requests filesystem write access and two command-execution tools, although its implementation only performs in-memory keyword matching, recommendation, workflow analysis, scoring, and guide generation. No implemented function in src/index.js requires Write, Bash, or Exec.

This violates the principle of least privilege. If tool availability is derived from this manifest, an injected instruction, malicious task, or future compromise of the Skill context could invoke these capabilities even though they are unrelated to the declared discovery and recommendation functionality.

The audit did not identify instructions or implementation code that currently abuses these permissions. The risk results from unnecessarily exposing powerful tools that could be abused if the agent's decision-making context is manipulated.

Attack Path

  1. The platform loads the Skill and grants the tools declared in SKILL.md.
  2. The Skill consequently receives Write, Bash, and Exec capabilities despite not needing them for its implemented behavior.
  3. An attacker supplies a malicious task or introduces injected content into data processed by the agent.
  4. The manipulated agent invokes Bash or Exec to run attacker-selected commands, or uses Write to alter files.
  5. Commands execute and files are modified within the permissions and sandbox boundaries of the hosting agent process.

Impact Assessment

Successful exploitation could provide arbitrary command execution with the operating-system privileges assigned to the agent runtime. It could also permit creation, modification, or deletion of files accessible through the Write tool. The precise scope depends on platform sandboxing, tool confir ...[truncated 285 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove Write, Bash, and Exec from the declared tool list.
  2. Remove Read as well if the Skill continues to receive its skill catalog as an in-memory argument. If catalog loading later requires file access, restrict reading to a dedicated catalog directory.
  3. Declare only the minimum tools required by implemented functionality.
  4. Enforce platform-level command allowlists, filesystem path restrictions, sandboxing, and explicit user approval for any future command-execution capability.
  5. Add a security test that compares declared tools against capabilities actually used by the implementation.
  6. Require a separate security review before introducing filesystem mutation or process-execution functionality.
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (12)

Known Vulnerable Dependency: brace-expansion==2.0.3 — 3 advisory(ies): CVE-2026-13149 (brace-expansion: DoS via exponential-time expansion of consecutive non-expanding); CVE-2026-14257 (brace-expansion: DoS via unbounded expansion length causing an out-of-memory pro); CVE-2026-69152 (brace-expansion: DoS via unbounded intermediate arrays, bypassing the CVE-2026-1)

High
Category
Supply Chain
Confidence
91% confidence
Finding

brace-expansion 2.0.3 has multiple DoS advisories involving pathological expansion inputs that can cause excessive CPU or memory consumption. Although present as a dev dependency, this package is commonly exercised by globbing and pattern-matching code in tooling, so malicious filenames, patterns, or config inputs in CI/build contexts could trigger resource exhaustion.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: browserslist==4.28.2 — 2 advisory(ies): CVE-2026-73088 (Browserslist: Uncaught crash / prototype write via untrusted browserslist-stats.); CVE-2026-73089 (Browserslist: Unbounded memory growth (no cache eviction) via distinct query res)

High
Category
Supply Chain
Confidence
84% confidence
Finding

browserslist 4.28.2 is reported vulnerable to crash/prototype-write behavior and unbounded memory growth when given untrusted stats or crafted query inputs. In this repository it is only part of the development dependency graph, which reduces exposure, but build or test processes that ingest attacker-controlled configuration or stats data could still be destabilized.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: js-yaml==3.14.2 — 4 advisory(ies): CVE-2026-84375 (js-yaml: maxTotalMergeKeys does not limit CPU use for empty merge sources); CVE-2026-59869 (js-yaml: YAML merge-key chains can force quadratic CPU consumption); GHSA-5p4m-2wfm-xmqj (JS-YAML: Quadratic CPU consumption in !!omap resolution (3.x and 4.x) — CVE-2026) +1 more

High
Category
Supply Chain
Confidence
93% confidence
Finding

js-yaml 3.14.2 has several CPU-exhaustion issues on crafted YAML documents, especially around merge keys and complex structures. Even though it is a transitive dev dependency, YAML parsing often occurs in config-loading paths, so any tooling that processes untrusted YAML during development or CI could be forced into denial of service.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The examples use generic natural-language prompts like '请帮我发现适合以下任务的技能' and '请推荐技能组合来完成以下任务', which are broad enough to be triggered unintentionally or to route arbitrary user goals into downstream skills without clear scoping. In a meta-skill whose purpose is discovery and orchestration, overly permissive trigger phrasing increases the chance of unsafe delegation, prompt-injection propagation, or activation for sensitive tasks the author did not intend to support.

Content

No source excerpt is available for this finding.

External Model or Provider Selection

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · SKILL.md (reported line 16)May include surrounding context.

md
- Write
  - Bash
  - Exec
model: sonnet
---

# Skill Discoverer - 技能发现器

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The natural-language content of the skill is overwhelmingly Chinese, including the title, usage instructions, FAQ, and guidance examples, but the document does not indicate that the user can choose another language or that the skill is intentionally limited to a Chinese-speaking context. This can violate language/locale policy because it effectively forces a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This JavaScript file contains user-facing natural-language output in Chinese, such as the recommendation reason at L061, and similar Chinese-only strings appear in other returned guide fields. Because the file does not provide a user opt-in, language selection, or justification for a Chinese-only locale, it violates the language/locale policy criterion.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: @babel/core==7.29.0 — 1 advisory(ies): CVE-2026-49356 (@babel/core: Arbitrary File Read via sourceMappingURL Comment)

Low
Category
Supply Chain
Confidence
77% confidence
Finding

The lockfile pins @babel/core 7.29.0, which is reported to have an arbitrary file read issue via sourceMappingURL processing. In this skill, the package appears only as a dev/test dependency transitively used by Jest, so exposure is narrower than a runtime dependency, but it can still matter in CI or local development workflows that process untrusted source files.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: baseline-browser-mapping==2.10.16 — 1 advisory(ies): CVE-2026-45819 (baseline-browser-mapping process termination on invalid input causes denial of s)

Low
Category
Supply Chain
Confidence
66% confidence
Finding

baseline-browser-mapping 2.10.16 is flagged for process termination on invalid input, which is a denial-of-service class issue. Here it is a transitive dev dependency under browserslist tooling, so the practical risk is limited to build/test or tooling paths rather than the deployed skill behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The package description is written only in Chinese ("L0 入口层 - 技能发现与引导引擎") and provides no indication that language choice is optional or that the skill is intended for a Chinese-only audience. This can create a locale-policy issue if users are expected to discover or use the skill without having opted into that language.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 29)May include surrounding context.

json
]
  },
  "devDependencies": {
    "jest": "^30.3.0"
  }
}

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

This JavaScript test file contains natural-language strings exclusively in Chinese for skill names and descriptions. Under the policy rule for language/locale violations, forcing a specific language without opt-in or justification is a potential issue even when it appears in code literals, because it can encode an implicit locale assumption into expected behavior.

Content

No source excerpt is available for this finding.