Back to skill

Security audit

短剧创作

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, mainly because it reports that project settings and packaging can bypass documented approval checks; some scanner coverage was partial.

Consider holding off on installation until you can review how authority changes and packaging are validated, and what file, network, and shell access the skill needs. Coverage was partial: 24 of 26 components were scanned, two had limited static analysis, and three local references were unresolved, so the results are 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Rogue AgentSelf-Modification, Session Persistence
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (12)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · assets/dashboard/app.js (reported line 999)May include surrounding context.

js
if (!ask) return "";
  return `<button class="btn sm ${primary ? "primary" : ""}" type="button" data-copy="${esc(ask)}" data-copy-label="已复制,去对话里发送">${icon("copy")}${label}</button>`;
}
/** Distance from a target value, stated without judging it: the profile gives the target, not a tolerance. */
function gapText(value, goal) {
  const d = Math.round((value - goal) * 10) / 10;
  return d === 0 ? "与目标相同" : `${d > 0 ? "多" : "少"} ${num(Math.abs(d))} 秒`;

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · scripts/creator_markdown_check.py (reported line 307)May include surrounding context.

python
VOICE_RECORD_PATH_RE = re.compile(
    r"([^(;\n]+?\.(?:" + "|".join(AUDIO_SUFFIXES) + r"))(?=(|$)", re.IGNORECASE
)
# The screenplay's dialogue grammar, mirrored from the write skill's
# screenplay_index.py (DIALOGUE_RE, and TAG_RE + VOICE_TAG_BODY_RE for [VO] and
# [OS]). Each is matched against a whole paragraph, so a line that wraps onto
# the next physical line is still one line of dialogue.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/creator_markdown_check.py (reported line 616)May include surrounding context.

python
if not lines or any(not line.startswith(">") for line in lines):
        return None
    prompt = "\n".join(line[1:].lstrip() for line in lines).strip()
    return prompt or None


def _copyable_prompt_cause(

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/creator_views.py (reported line 497)May include surrounding context.

python
if re.search(r"^#{2,4}[ \t\u3000]*IMG(?:-|\b)", text, re.MULTILINE):
            raise ValueError("没有找到 `## IMG-… · 标题` 这样的图片提示词条目")
        return []
    return prompts


def parse_video_prompts(text: str) -> list[dict[str, Any]]:

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/creator_views.py (reported line 519)May include surrounding context.

python
})
    if not prompts:
        raise ValueError("没有找到 `## MOTION-… · 标题` 这样的视频提示词条目")
    return prompts


def parse_video_groups(text: str) -> list[dict[str, Any]]:

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
75% 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 · assets/dashboard/app.js (reported line 1577)May include surrounding context.

js
extItem ? `<button class="btn sm primary" type="button" data-copy="${esc(nextItem.text)}" data-key="${esc(nextItem.key)}" data-next="1">${icon("copy")}复制下一条 · <span class="mono">${esc(nextItem.id.replace(/^(MOTION|SHOT|IMG|GROUP)-(EP\d+-)?/, ""))}</span></button>` : `<span class="status ok">${icon("check")}全部复制过</span>`}
      <button class="btn sm ghost" type="button" data-act="clear-copied" data-tab="${tab}">清除标记</button><span class="muted lbl-long">标记只存在这台电脑的浏览器里</span></div>
    <div class="plist">${list.map((item) => `<article class="card pcard ${S.copied.has(item.key) ? "copied" : ""}" id="p-${esc(item.id)}" data-key="${esc(item.key)}">
      <div><div class="pid">${esc(item.id)}</div><div class="pt">${esc(item.title)}</div><div class="pm">${item.meta.filter(Boolean).map((line) => `<span>${line}</span>`).join("")}</div></div>
      <div class="ptext ${open === item.id ? "" : "clamp"}" data-act="expand">${esc(item.text)}</div>
      <div class="pact">${copyButton(item.text, item.k

Session Persistence

Medium
Category
Rogue Agent
Confidence
75% 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 · assets/dashboard/styles.css (reported line 542)May include surrounding context.

css
.copybar { position: sticky; top: calc(var(--zs-topbar-height) + var(--zs-tabbar-height) + var(--zs-space-2h)); z-index: 5; display: flex; align-items: center; gap: var(--zs-space-3h); padding: var(--zs-space-2h) var(--zs-space-3h); margin-bottom: var(--zs-space-3h); box-shadow: var(--zs-shadow-sm); flex-wrap: wrap; }
.copybar .bar { width: 160px; margin: 0; }
.copybar .lbl-long { font-size: var(--zs-text-12); margin-left: auto; }
.plist { display: grid; gap: var(--zs-space-2h); }
.pcard { display: grid; grid-template-columns: 200px minmax(0, 1fr) auto; gap: var(--zs-space-4h); padding: var(--zs-space-3h) var(--zs-space-4); align-items: start; scroll-margin-top: 180px; }
.pcard > div { min-width: 0; }
.pcard.copied { background: linear-gradient(90deg, var(--zs-ok-wash), transparent 40%), var(--zs-surface); }

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/dashboard_server.py (reported line 1879)May include surrounding context.

python
else:
        options["start_new_session"] = True
    try:
        return subprocess.Popen(
            command,
            stdin=subprocess.DEVNULL,
            stdout=log,

The documented decision-backed authority boundary is bypassed by the configure command.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

L1130 states that set-authority is the only sanctioned way to modify short-drama.json, but configure_project directly writes authority settings, marks blocks accepted, and records decision="direct:configure" without requiring an accepted decision artifact (L2283–L2337). This contradicts the documented provenance boundary: a reported bound setting need not originate from the accepted-decision validation performed by set-authority.

Content

No source excerpt is available for this finding.

Documentation presents package as an approved-delivery command although approval checks are optional.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The export documentation identifies package as the command representing approved delivery, but package defaults to strict_ledger=False and therefore permits untracked, unaccepted, or unreviewed files. An agent relying on that distinction can deliver content that never passed creator acceptance or review; changing rejected bytes also avoids the default mode's digest-specific rejection checks without requiring renewed approval. This is a creative-workflow integrity issue rather than code execution or privilege escalation.

Content

No source excerpt is available for this finding.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/dashboard_server.py (reported line 492)May include surrounding context.

python
missing = [
            name
            for name in self.backend.contract
            if not callable(getattr(project_tool, name, None))
        ]
        if missing:
            raise RuntimeError(