Back to skill

Security audit

Poker Clip

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a local poker-video clipping tool with disclosed, purpose-aligned file generation and no evidence of exfiltration, hidden persistence, or destructive behavior.

Install and run this only in a workspace where generated clips and JSON artifacts can be overwritten. Use a virtual environment, consider pinning openai-whisper, and review generated transcripts/subtitles before publishing videos from untrusted media sources.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T08 · Insecure Dependencies

Warning
Location
README.md:15
Finding

Unpinned OpenAI Whisper Dependency

Content
View full analysis
Remediation
View remediation
\ --hash=sha256: ``` 3. Install with hash enforcement: ```bash python -m pip install --require-hashes -r requirements.txt ``` 4. Generate and retain a complete dependency lock so that transitive package versions cannot change silently. 5. Periodically scan the locked dependency set for known vulnerabilities. 6. Install dependencies inside an isolated virtual environment using a non-privileged account. 7. Document the trusted package index explicitly and avoid unreviewed alternative indexes. ]]>

T09 · Insecure Skill Coding Practices

Note
Location
scripts/poker_clipper.py:267
Finding

ASS Subtitle Instruction Injection Through Unescaped Transcript and Hook Text

Content
View full analysis
s0+dur: continue ts = max(0, seg['start']-s0); te = min(dur, seg['end']-s0) if te <= ts: continue sp = '{\\pos(' + str(CW//2) + ',' + str(SUB_Y) + ')}' events.append(f"Dialogue: 0,{ass_time(ts)},{ass_time(te)},Sub,,0,0,0,,{sp}{wrap(seg['text'])}") ap.write_text(hdr + "\n".join(events) + "\n", encoding='utf-8') ``` The same issue appears in `scripts/patch_hooks.py:66-80`: ```python hook_wrapped = wrap(hook_text, mx=22) hook_pos = '{\\pos(' + str(CW // 2) + ',' + str(HOOK_Y) + ')}' events.append(f"Dialogue: 1,{ass_time(0)},{ass_time(min(3.0, dur))},Hook,,0,0,0,,{hook_pos}{hook_wrapped}") for seg in segs: if seg['end'] < s0 or seg['start'] > s0 + dur: continue ts = max(0, seg['start'] - s0) te = min(dur, seg['end'] - s0) if te <= ts: continue wrapped = wrap(seg['text'], mx=18) sub_pos = '{\\pos(' + str(CW // 2) + ',' + str(SUB_Y) + ')}' events.append(f"Dialogue: 0,{ass_time(ts)},{ass_time(te)},Sub,,0,0,0,,{sub_pos}{wrapped}") ap.write_text(hdr + "\n".join(events) + "\n", encoding='utf-8') ``` ### Technical Analysis ASS subtitle dialogue text supports override instructions enclosed in braces, such as position, color, size, animation, clipping, and transformation controls. Backslashes and line breaks also have syntactic meaning. The `wrap()` functions only divide text by spaces and add `\N` line s ...[truncated 2330 chars]
Remediation
View remediation
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (11)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README presents the skill description, usage instructions, and example user prompts entirely in Chinese, including explicit interaction text to tell the AI assistant. Under SQP-3, forcing a specific language without opt-in is a natural-language policy concern unless the locale restriction is clearly documented and justified.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The invocation examples include very broad natural-language triggers like “帮我处理这个扑克视频” and “切割新视频”, which are generic enough to match unrelated user requests about video processing. In an agent environment, this can cause accidental skill activation and unintended file/video processing, especially if routing is based on fuzzy intent matching rather than explicit commands.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This code performs a file write that updates an existing JSON file in place. Although the script prints progress messages, it does not clearly disclose beforehand that it will overwrite clips_analysis.json or offer any confirmation, which is a safety-relevant data modification for a code file.

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/patch_hooks.py (reported line 126)May include surrounding context.

python
'-c:a', 'aac', '-b:a', '128k', str(mp4)]

        print(f"  [{n}] {hook[:60]}")
        r = subprocess.run(cmd, capture_output=True, text=True)
        if r.returncode != 0:
            print(f'      [WARN] {r.stderr[-200:]}')
        else:

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/poker_clipper.py (reported line 300)May include surrounding context.

python
'-c:a', 'aac', '-b:a', '128k', str(mp4)]

        print(f"  [{n}] {hook[:60]}")
        r = subprocess.run(cmd, capture_output=True, text=True)
        if r.returncode != 0:
            print(f'      [WARN] {r.stderr[-200:]}')
        else:

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The transcription call hard-codes language="en", which imposes an English-only locale choice on all inputs. Under the stated policy, forcing a specific language without user opt-in or documented justification is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file documents a workflow that processes an input video and emits generated clips, hooks, and patched videos, but it does not include any user-facing warning about file creation or potential overwriting in clips/ or related paths. For markdown files, SQP-2 applies when behavior affecting user data or system state is described without disclosure.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The file presents the skill’s operational instructions entirely in Chinese, which can amount to a language/locale constraint if users are not given an explicit choice or opt-in. Under the policy, language forcing is acceptable only when the locale restriction is documented and justified or when users can choose their language.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The docstring states a guarantee that every clip gets a unique hook. However, after exhausting formula-based hooks and the finite fallback pool, the code returns a generic last-resort string template at L139 without checking it against used_hooks, so the uniqueness guarantee is not actually enforced by code in all cases.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

This code writes a full transcript JSON file to the clips output directory, which persists derived content from the user's video input. While the module has brief internal comments and console progress messages, there is no explicit user-facing warning in the code comments/docstring that the skill saves transcripts and analysis artifacts to disk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The ffmpeg command uses the '-y' flag, which automatically overwrites existing output files, and the function also creates new MP4 clips on disk. Although filenames are printed during processing, there is no explicit disclosure or confirmation that prior outputs may be replaced.

Content

No source excerpt is available for this finding.