Back to skill

Security audit

sn-ppt-standard

Security checks for vulnerabilities and agentic risk

Overview

This presentation skill needs review because it can terminate unrelated browsers, execute substituted code from temporary folders, and delete existing deck assets.

Prefer a corrected release that restricts browser cleanup to its own processes, removes shared-temporary-directory imports and preserves existing assets. Until corrected, use a disposable, nonprivileged container with separate process visibility, restricted network access, execution limits and only required credentials. Back up the deck and fonts, redact confidential image queries, and verify requested PPTX files independently. This was a static review; vendor code was not exhaustively audited.

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
scripts/render.py:142
Finding

Dependency fallback executes Python packages from predictable shared temporary directories

Content
View full analysis

Impact

Attacker-supplied Python runs with the renderer's OS privileges and can access files and credentials available to that process. This is a cross-user code-execution path, not merely temporary-file corruption.

Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (45)

Tainted flow: 'key' from os.environ.get (line 24, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/serper_images.py (reported line 27)May include surrounding context.

python
key = os.environ.get("SERPER_API_KEY", "").strip()
    if not key:
        raise SystemExit("SERPER_API_KEY is not configured")
    response = requests.post(
        SEARCH_ENDPOINT,
        headers={"X-API-KEY": key, "Content-Type": "application/json"},
        json={"q": query, "num": min(max(limit, 1), 10)},

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The reported cleanup behavior can kill unrelated Chromium sessions based only on age or orphan status, disrupting other workloads and potentially losing unsaved work. Browser rendering is legitimate here, but it does not justify terminating processes outside the current job; the implementation was not included, so this assessment relies on the static finding's description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied script’s primary function is attachment preprocessing, materially different from the declared presentation-generation workflow. Although material preparation could support a broader presentation system, this chunk independently exposes a broad ingestion pipeline that the description does not disclose. It neither checks the DECK_DIR/outline.md prerequisites nor implements static_postprocess or the mandatory presentation deliverables. Its PNG outputs are derived attachment views, not rendered HTML slides. This assessment is limited to the supplied code; it does not establish malicious behavior or what other skill files implement.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The code substantially supports the described presentation workflow and explicitly checks for present.html during audit. However, build and audit can succeed without any PPTX because neither examines static_postprocess nor checks or generates that artifact. This omits a central, explicitly mandatory delivery condition. The supplied chunk also acts primarily as an assembler and validator for pre-existing slide HTML rather than the full generator described. These findings apply to the supplied code; other unprovided skill components could implement additional workflow steps. No unrelated data exfiltration or malicious capability is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The chunk implements external image acquisition rather than the declared presentation-generation workflow. Asset storage and registration could support slide generation, but credentialed third-party searches and caller-directed network downloads introduce capabilities and resource access not disclosed by the description. The absence of presentation outputs in this helper alone is not treated as evidence that the entire skill fails to deliver them.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 82)May include surrounding context.

md
`<DECK_DIR>/<DECK_ID>.pptx` ← 必交付产物(`static_postprocess` 含 `pptx` 时):只能由 `scripts/export_pptx/html_to_pptx.mjs` 生成

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 331)May include surrounding context.

md
| `render.py` | 保留:单页诊断;`--batch` 复用同一 Chromium 渲染整册或指定页 |

Exporter setup failure silently downgrades mandatory PPTX delivery to HTML-only delivery.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

On setup failure, the exporter returns normally and declares HTML to be the final deliverable even when PPTX is required. Although the JSON explicitly reports a skipped export, the successful process termination and permissive message can cause an agent or automation to accept incomplete delivery, violating the skill's mandatory-output contract.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: image-size==1.2.1 — 2 advisory(ies): CVE-2025-71329 (image-size: JXL and HEIF parsers allow denial of service through infinite loops); CVE-2025-71330 (image-size: ICNS parser allows denial of service through an infinite loop)

High
Category
Supply Chain
Confidence
87% confidence
Finding

The lockfile includes image-size 1.2.1 through pptxgenjs, matching the version identified by the supplied infinite-loop advisories. If attacker-controlled images reach the affected parsers, PPTX export can hang and consume worker capacity; actual exposure depends on accepted formats and exporter call paths.

Content

No source excerpt is available for this finding.

Chromium cleanup contradicts its promise to preserve other active renderers and kill only leaked processes.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

_reap_leaked_chromium scans all visible processes and sends SIGKILL to matching Chromium/headless/NaCl names when their parent is PID 1 or their age exceeds 300 seconds, without verifying ownership by this rendering job. Because it is registered as an exit handler, an ordinary render invocation can terminate unrelated browser sessions or other active rendering jobs that the executing account has permission to kill; shared rendering infrastructure makes this especially dangerous.

Content

No source excerpt is available for this finding.

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

Image search and generation workflows lack a privacy disclosure for task-derived queries and prompts.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The workflow proactively derives image queries and generation prompts from task material without specifying a confidentiality check or authorization boundary for external transmission. Presentation inputs may contain unpublished products, internal plans, or personal information, which can consequently reach external providers even though image acquisition itself is legitimate.

Content

No source excerpt is available for this finding.

The delivery workflow permits HTML-only delivery after a required PPTX export fails.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest at L003 makes PPTX mandatory when static_postprocess contains pptx and prohibits delivery if it is missing; HTML-only delivery is allowed only when explicitly requested. However, L250–L252 instruct the agent to mark an export failure as partial and deliver the HTML version, creating a failure-path exception that contradicts the manifest's mandatory delivery gate.

Content

No source excerpt is available for this finding.

The Serper search workflow omits disclosure that search queries are sent to a third-party service.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

This contract directs the parent agent to submit image-search queries through Serper without requiring a check that task-derived content is authorized for external disclosure. Protecting the API key from logs does not protect confidential names, project details, or personal data included in the query.

Content

No source excerpt is available for this finding.

External image searches and generation are instructed without warning that brief-derived content may leave the workspace.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The instructions encourage sending topic-derived search queries and image-generation prompts to external services without a confidentiality check. In a presentation skill handling internal business or research briefs, these requests could disclose sensitive names, plans, or unpublished information; downloading the resulting assets locally does not prevent that disclosure.

Content

No source excerpt is available for this finding.

The checklist allows a partial HTML outcome when a required PPTX export fails.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The checklist permits a partial HTML delivery after a requested PPTX export fails, contradicting the skill's explicit requirement that a missing mandatory PPTX is a technical failure that must block delivery. This creates a workflow-integrity weakness: an agent following the checklist can bypass a required deliverable gate, although retaining the error and marking the result partial reduces the risk of concealed failure.

Content

No source excerpt is available for this finding.

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/deck.py (reported line 473)May include surrounding context.

python
def _validate_player_runtime(root: Path) -> None:
    render = Path(__file__).resolve().parent / "render.py"
    proc = subprocess.run(
        [sys.executable, str(render), "--audit-player", str(root)],
        cwd=root,
        capture_output=True,

The delivery audit can report PASS without checking the conditionally mandatory PPTX.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

At L1288–L1297, the audit validates HTML-related artifacts and requires present.html, then prints delivery-audit:PASS. It never checks static_postprocess or PPTX existence and provenance, so it can certify a delivery that violates the manifest when PPTX is requested.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx playwright'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

The converter fetches arbitrary HTTP(S) image URLs from input HTML.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Input HTML controls HTTP(S) destinations fetched by the exporter without host or address restrictions, enabling server-side requests to loopback, private-network services, or cloud metadata endpoints. Responses are saved as assets without image validation, while missing timeouts and response-size limits also enable hangs or memory exhaustion.

Content

No source excerpt is available for this finding.

scripts/export_pptx/lib/postprocess_pptx.mjs documents collision-free gradient placeholders, but generates ordinary RGB colors that can also replace unregistered fills.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

L070–L071 claim that gradient tokens never collide with real six-digit hexadecimal colors, but L078–L088 generate valid colors such as AB0001. The replacement logic at L151–L164 matches these values throughout each slide without distinguishing registered placeholders from genuine colors, contradicting the safety assurance at L146–L149 and potentially converting unrelated solid fills into gradients.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: echarts==5.6.0 — 1 advisory(ies): CVE-2026-45249 (Apache ECharts has a cross-site scripting (XSS) vulnerability)

Medium
Category
Supply Chain
Confidence
80% confidence
Finding

The lockfile confirms ECharts 5.6.0, which the supplied advisory identifies as affected by XSS. Untrusted chart content could execute script if it reaches an affected browser-rendered feature; the lockfile alone does not establish that feature's reachability or independently verify the advisory.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: echarts==5.6.0 — 1 advisory(ies): CVE-2026-45249 (Apache ECharts has a cross-site scripting (XSS) vulnerability)

Medium
Category
Supply Chain
Confidence
70% confidence
Finding

OSV returned vulnerability advisories matching the dependency's resolved version. Review the matched advisories to assess their impact.

Content

No source excerpt is available for this finding.

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/font_bundle.py (reported line 566)May include surrounding context.

python
weight: str,
    style: str,
) -> None:
    result = subprocess.run(
        [
            tool,
            str(source),

Font bundling silently deletes existing WOFF2 files not included in the newly generated bundle.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

Cleanup deletes every WOFF2 file in assets/fonts that is absent from the newly generated records, without establishing that the bundler created it. Although stale generated-font removal is appropriate, the custom-font loader permits sources anywhere inside the workspace, including assets/fonts, so an authorized uploaded source or an unrelated font can be silently deleted during ordinary bundling.

Content

No source excerpt is available for this finding.