Back to skill

Security audit

Album Pipeline

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real album-production automation skill, but it needs Review because it launches broad background workflows, sends creative content to external MiniMax tools, and includes unsafe shell command templates using user-derived names.

Install only if you are comfortable with a large automated workflow that creates many files, runs background sub-agents, and sends prompts, lyrics, images, audio, and video-generation requests to MiniMax tools. Before use, require filename slug validation/quoting for album and track names, review external-service privacy and billing terms, and add explicit confirmations before CLI execution and overwrites.

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

Error
Location
phase6-packager/SKILL.md:37
Finding

Shell Command Injection Through an Unquoted User-Controlled Album Name

Content
View full analysis
zip -r -promotional-materials.zip \ docs/album-overview.md \ docs/promotional-materials.md \ docs/cover-concept.md \ docs/artist-story-cn.md \ docs/artist-story-en.md \ docs/artist-story-short.md \ docs/artist-story-quotes.md \ docs/platform-check.txt \ generate/covers/album-cover-p1.png \ generate/covers/album-cover-p2.png \ generate/covers/album-cover-p3.png \ generate/covers/tracks/ \ docs/promo-video.mp4 \ docs/promo-video-15s.mp4 ``` The original command uses the user-derived album-name placeholder directly as an unquoted shell argument. The album name is collected during the concept phase and subsequently used to construct project paths and output filenames: ```text 1. Create the project directory workspace/projects/album-{album-name}/ 2. Copy templates/album-overview.md to docs/album-overview.md 3. Create song files according to the requested track count. ``` ### Technical Analysis The album name originates from user-supplied concept information. The packaging command incorporates that value into the ZIP filename without quoting, escaping, slug generation, or character validation. If an agent implements the documented command by directly replacing the placeholder and invoking a shell, shell metacharacters in the album name—such as semicolons, command substitutions, pipes, redirection operators, or newlines—will be interpreted as shell syntax rather than as part of the filename. For example, an album name structurally equivalent to: ```text release; attacker-command; # ``` could produce a command with this structure: ```bash zip -r release; attacker-command; #-promotional-materials.zip ... ``` The shell would treat `attacker-command` as a sepa ...[truncated 1942 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (52)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown contract is written as a mandatory instruction set and uses Chinese-only operational language from the outset, stating that all skills must strictly comply. Because the file does not offer a language choice or document a justified locale restriction, it imposes a specific language/locale policy on downstream agents.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The README advertises full automation from concept to release and describes extensive file production, but it does not warn users about the scope of generated artifacts or the fact that packaging can create many deliverables automatically. This weakens informed consent and can surprise users with substantial workspace changes or downstream publication-ready outputs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README presents bilingual Chinese-and-English output as a fixed behavior rather than a user-selectable option, which can cause unnecessary data generation, higher external-service usage, and possible disclosure of content through translation or duplicated processing. The issue is less severe than direct code execution, but in this pipeline it expands output scope and cost without explicit user consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill describes invoking external AI services for music, image, and video generation without clearly warning users that prompts and content may be sent to third-party systems. In context, this is significant because creative inputs, lyrics, and generated assets may contain sensitive or proprietary material, creating privacy, policy, and cost risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger phrases are broad enough to activate on ordinary conversation about making an album or AI music, which can cause unintended execution of a complex multi-phase workflow. In this skill, accidental activation is more concerning because the pipeline can generate many artifacts and invoke downstream external services, increasing the blast radius of a mistaken trigger.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger condition "用户表达想做概念专辑" is broad enough to activate the skill during general music discussion, ideation, or educational conversations that are not explicit requests to run a multi-phase production pipeline. Because this skill orchestrates many downstream agents and file-producing phases, ambiguous activation increases the chance of unintended execution and unnecessary artifact generation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill describes generation, formatting, CLI execution, transcoding, and packaging, but does not clearly disclose to the user that these phases will create and modify files across the workspace. In a skill that enforces strict file contracts and automated sub-agent execution, lack of explicit write-side transparency can lead to unexpected state changes, unwanted outputs, and reduced user ability to give informed consent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill is written primarily in Chinese and operational prompts are Chinese-only, while the flow does not require checking or honoring the user's language preference before proceeding. This can cause consent and comprehension failures, especially around confirmation, timing, and workflow side effects, making users more likely to approve actions they do not fully understand.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger examples include a very generic phrase, "album concept", which can match ordinary conversation and cause unintended invocation of this skill. Over-broad activation can route users into the wrong workflow, create unexpected file operations and agent spawning, and increase the chance of context confusion or unauthorized actions in a multi-skill environment.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrase "Phase 6" is very generic and can match routine conversation, status references, or documentation mentions rather than an explicit request to run this skill. Because this skill orchestrates multiple downstream agents and writes files/creates archives, accidental invocation could start a broad workflow with unintended side effects.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown skill is almost entirely written in Chinese and does not offer an alternative language or indicate that Chinese is optional. Under the policy rule, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file defines the skill interfaces and trigger conditions using only Chinese example invocations such as “我想做一张关于 XX 的专辑”, “写第 N 首歌”, and “生成音乐”. Because the file is presented as the design basis for the skill, this effectively biases activation and usage toward a single language without documenting user choice, opt-in, or a justified locale restriction.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrase "提取歌词" is generic and likely to appear in normal conversation or unrelated requests, which increases the chance of accidental skill activation. Because this skill performs orchestration and writes output files, unintended triggering could lead to unauthorized or premature processing of project data.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Using the standalone trigger phrase "Phase 3" is overly broad and can match ordinary discussion about workflow stages rather than a deliberate invocation of this skill. In an agent system, that ambiguity can cause unintended execution of file-processing actions and spawning of sub-agents when the user did not explicitly request this operation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
72% confidence
Finding

The file is written almost entirely in Chinese and specifies Chinese-language workflow elements such as Chinese standard lyric files and Chinese trigger phrases, without stating that users may choose another language. Under SQP-3, forcing a specific language without opt-in can be a natural-language policy issue unless the locale constraint is explicitly justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The prompt review stage explicitly overwrites existing prompt files, which can destroy prior artifacts, user edits, or audit history without confirmation. In an agentic workflow this is dangerous because automated re-runs may silently replace curated prompts, making recovery difficult and undermining traceability of what was actually sent to downstream generation tools.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

These instructions send prompt text and lyric files to the MiniMax CLI/service without any warning that project content is being transmitted to an external provider. This creates a real privacy and confidentiality risk if lyrics, drafts, or embedded metadata are sensitive, especially because the workflow encourages bulk parallel submission of all songs.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill explicitly instructs sending prompt and lyrics content to an external service via MiniMax CLI, but it does not warn that local content will leave the environment or require a privacy review before transmission. If prompts or lyrics contain unpublished, licensed, or sensitive material, this can cause unauthorized disclosure to a third-party provider and possible retention by that provider.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill description and instructions are written as Chinese-only operational guidance, and the inputs/checklist are framed around Chinese standard lyrics by default. There is no statement that the user may choose language/locale, nor a documented reason that the skill must operate only in Chinese, which can violate language/locale policy expectations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The '修改范围' section says the skill may write only generate/prompts/T{N}-曲名-promptN.txt. Later, the document explicitly instructs the skill to also generate generate/prompts/index.json, which contradicts the earlier documented write scope rather than merely adding detail.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The checklist states that Chinese albums must generate Chinese prompts, which imposes a language requirement as a default behavior. Under the policy, forcing a specific language without user choice or opt-in is a natural-language policy violation unless the locale constraint is clearly documented as justified or optional.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The core description is written as an instruction that the agent operates in Chinese ("独立审查 Agent。对所有 Prompt 进行 6 项检查并优化...") with no indication that users may choose another language or locale. This creates a natural-language locale policy concern because it implicitly forces a specific language without opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file’s operational instructions, triggers, inputs, execution steps, and checklist are all presented in Chinese, which effectively forces a specific language for users and operators. Under the policy, locale or language constraints should either be optional for the user or explicitly justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The natural-language instructions, headings, and checklist are all presented in Chinese, which effectively forces a specific language for users of the skill. The file does not offer bilingual usage, opt-in language selection, or state that the skill is limited to a Chinese-only regional context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file’s operational instructions, triggers, inputs, and checklist are all presented in Chinese, which effectively forces a specific language for users or downstream agents. Under the policy, a fixed language is only acceptable when the skill offers opt-in/choice or clearly documents a justified regional constraint, neither of which appears here.

Content

No source excerpt is available for this finding.