Back to skill

Security audit

story-cover:小说封面生成

Security checks for vulnerabilities and agentic risk

Overview

The verdict is benign because the reported activity fits a book-cover image-generation skill, though privacy concerns and incomplete reference checking leave some uncertainty.

The verdict is benign, but consider whether you are comfortable sending unpublished book details or reference images to the configured image provider. The scan was partial because a local path referenced in SKILL.md was missing, so the evidence is incomplete.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (10)

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · references/api-fallback.md (reported line 90)May include surrounding context.

md
OUT="$BOOK_DIR/封面/封面_v${i}.png"
RESP=$(mktemp)
REF_TMP=""
trap '[ -n "$REF_TMP" ] && rm -f "$REF_TMP"; rm -f "$RESP"' EXIT

# URL 先下载到临时文件,本地路径直接用。用裸 mktemp 以保证 macOS/Linux 行为一致。
case "$REF_IMAGE" in

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · references/api-fallback.md (reported line 90)May include surrounding context.

md
OUT="$BOOK_DIR/封面/封面_v${i}.png"
RESP=$(mktemp)
REF_TMP=""
trap '[ -n "$REF_TMP" ] && rm -f "$REF_TMP"; rm -f "$RESP"' EXIT

# URL 先下载到临时文件,本地路径直接用。用裸 mktemp 以保证 macOS/Linux 行为一致。
case "$REF_IMAGE" in

The image-generation workflow lacks a privacy disclosure for submitted book details and reference images.

Medium
Category
Output Handling
Confidence
93% confidence
Finding

The workflow can derive prompts from unpublished local book information and submit reference images to a hosted generation service without explaining which information leaves the workspace. Requesting a cover establishes the intended generation task, but does not necessarily establish informed approval to disclose confidential story details or sensitive reference images.

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 · references/api-fallback.md (reported line 10)May include surrounding context.

md
| 变量 | 必填 | 默认 | 说明 |
|:-----|:----:|:-----|:-----|
| `GPT_IMAGE_API_KEY` | 是 | — | OpenAI 或兼容代理的 API Key |
| `GPT_IMAGE_BASE_URL` | | `https://api.openai.com/v1` | 用兼容代理时改这个 |
| `GPT_IMAGE_MODEL` | | `gpt-image-2` | 只在测试新模型时覆盖 |
| `GPT_IMAGE_SIZE` | | `1024x1536` | 官方接口竖版用 `1024x1536`(支持 `1024x1024`/`1024x1536`/`1536x1024`/`auto`,勿传非标 `768x1024` 免报 400);3:4 上传图由入口 Step 5 裁切导出 |
| `BOOK_DIR` | 是 | — | `封面/` 的上一层,按入口「收集信息」定的保存位置 |

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 · references/api-fallback.md (reported line 23)May include surrounding context.

md
set -euo pipefail
: "${GPT_IMAGE_API_KEY:?请设置 export GPT_IMAGE_API_KEY=你的key}"
: "${PROMPT:?请先 export PROMPT=构建提示词步骤拼好的完整提示词}"
BASE_URL="${GPT_IMAGE_BASE_URL:-https://api.openai.com/v1}"
MODEL="${GPT_IMAGE_MODEL:-gpt-image-2}"
SIZE="${GPT_IMAGE_SIZE:-1024x1536}"
BOOK_DIR="${BOOK_DIR:?请先 export BOOK_DIR=书目录或 ./covers/<书名>}"

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 · references/api-fallback.md (reported line 76)May include surrounding context.

md
set -euo pipefail
: "${GPT_IMAGE_API_KEY:?请设置 export GPT_IMAGE_API_KEY=你的key}"
: "${PROMPT:?请先 export PROMPT=构建提示词步骤拼好的完整提示词}"
BASE_URL="${GPT_IMAGE_BASE_URL:-https://api.openai.com/v1}"
MODEL="${GPT_IMAGE_MODEL:-gpt-image-2}"
SIZE="${GPT_IMAGE_SIZE:-1024x1536}"
BOOK_DIR="${BOOK_DIR:?请先 export BOOK_DIR=书目录或 ./covers/<书名>}"

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 · references/api-fallback.md (reported line 44)May include surrounding context.

md
--arg s "$SIZE" \
  '{model:$m, prompt:$p, size:$s}')

curl -fsS --max-time 180 --retry 2 --retry-delay 5 \
  "$BASE_URL/images/generations" \
  -H "Authorization: Bearer $GPT_IMAGE_API_KEY" \
  -H "Content-Type: application/json" \

The API workflow lacks a privacy warning about sending prompts and reference images to an external provider or configured proxy.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The API workflow uploads the full prompt and reference image to the configured provider or proxy without a privacy notice or sensitive-content confirmation. Unpublished story information or private image content could therefore reach a third party whose handling practices the author has not considered.

Content

No source excerpt is available for this finding.

The skill requires English design instructions without offering a language choice.

Low
Category
Output Handling
Confidence
99% confidence
Finding

The instruction explicitly requires design instructions to be written in English while preserving only the title and pen name in their original language. This also governs prompts delivered directly to users in prompt-only mode, but no user opt-in or justification for the language restriction is provided.

Content

No source excerpt is available for this finding.

Design instructions are required to be in English without user opt-in or a documented justification.

Low
Category
Output Handling
Confidence
98% confidence
Finding

The instruction “设计指令用英文” explicitly requires English for the generated design instructions. Although book titles and pen names retain their original language, the file offers no language choice or justification for imposing English on the design instructions.

Content

No source excerpt is available for this finding.