Back to skill

Security audit

OpenClaw Editing

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this video-editing skill suspicious because scanners found risky deletion guidance and a confirmed path for plan data to alter FFmpeg filter instructions; analysis was also incomplete for one referenced image.

Consider the verdict’s key risks before installing: following the cleanup guidance could delete project data, and a crafted plan value could alter FFmpeg filter behavior. The scan was partial: 20 of 21 components were scanned, and the referenced image could not be fully interpreted, so the evidence does not establish that all content is safe.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (32)

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 97)May include surrounding context.

md
- The look comes from the openclaw.ai Foundation social image (`assets/style-reference.png`): dark fresco, a red claw reaching toward an ember spark. Use the st

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
91% confidence
Finding

The surrounding recommendation escalates a glob-handling issue into rm -rf on the containing directory, substantially broadening the destructive operation. This is unsafe operational guidance rather than evidence of malicious tool manipulation, but following it against the wrong project directory can cause irreversible data loss.

Content

Scanner excerpt · references/pipeline.md (reported line 8)May include surrounding context.

md
- ffmpeg/ffprobe: from PATH (`brew install ffmpeg`) or `OPENCLAW_FFMPEG`. The build must include prores_ks and libx264; `render.py` finds it automatically.
- If the shell is **zsh**:
  - `for a in "x 1" …; do set -- $a` does not word-split. Feed `name time` pairs through a bash heredoc or `while read` (see examples).
  - A glob with no matches (`rm dir/*` on an empty dir) aborts the whole command line. Use `rm -rf dir && mkdir dir`.
- Scratchpads can be wiped between sessions. Keep plans and scripts in the project's `Motion Graphics/source/`.

## Frame-accurate cutting (edit.py)

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

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 · examples/enterprise_interview_composite.py (reported line 42)May include surrounding context.

python
"-c:v:0", "libx264", "-preset", "medium", "-crf", "16", "-profile:v:0", "high", "-r:v:0", "30",
         "-colorspace", "bt709", "-color_primaries", "bt709", "-color_trc", "bt709", "-color_range", "tv",
         "-c:a", "aac", "-b:a", "192k", "-c:v:1", "mjpeg", "-disposition:v:1", "attached_pic", "-movflags", "+faststart", sys.argv[1]]
subprocess.run(args, check=True)

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 · examples/long_update_recut_composite.py (reported line 47)May include surrounding context.

python
"-c:a", "aac", "-b:a", "192k", "-ar", "48000", "-movflags", "+faststart",
         "-metadata", "title=OpenClaw update and other features you should know", sys.argv[1]]
json.dump({"total_frames": TOTAL, "placement_frames": place}, open("timeline_v3.json", "w"), indent=1)
subprocess.run(args, check=True)

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 · examples/enterprise_interview_graphics.py (reported line 502)May include surrounding context.

python
"-vf", "scale=out_color_matrix=bt709:out_range=tv,format=yuva444p10le",
           "-c:v", "prores_ks", "-profile:v", "4444", "-alpha_bits", "16", "-vendor", "apl0", "-qscale:v", "6",
           "-colorspace", "bt709", "-color_primaries", "bt709", "-color_trc", "bt709", out]
    p = subprocess.Popen(cmd, stdin=subprocess.PIPE)
    for i in range(nframes(clip.dur)):
        p.stdin.write(clip.frame(i).tobytes())
    p.stdin.close()

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 · examples/short_update_edit.py (reported line 50)May include surrounding context.

python
def run(args):
    subprocess.run([FF, "-v", "error", "-y"] + args, check=True)


# ---- 1. cut ----

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 · examples/short_update_edit.py (reported line 72)May include surrounding context.

python
# ---- 2. audio ----
def rd(p):
    """mono float32 at the native rate, decoded by ffmpeg"""
    b = subprocess.run([FF, "-v", "error", "-i", p, "-ac", "1", "-f", "f32le", "-"], capture_output=True, check=True).stdout
    return np.frombuffer(b, np.float32)

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 · examples/short_update_edit.py (reported line 91)May include surrounding context.

python
L = lag(rd(raw), rd(f"{WORK}/den.wav"))
print("denoiser latency (samples):", L)
comp = f"atrim=start_sample={L},asetpts=PTS-STARTPTS," if L > 0 else ""
m = subprocess.run([FF, "-hide_banner", "-i", f"{WORK}/den.wav", "-af", comp + "loudnorm=I=-16:TP=-2:LRA=11:print_format=json",
                    "-f", "null", "-"], capture_output=True, text=True).stderr
mj = json.loads(m[m.rindex("{"):m.rindex("}") + 1])
ln = (f"loudnorm=I=-16:TP=-2:LRA=11:measured_I={mj['input_i']}:measured_TP={mj['input_tp']}:"

Recommends recursive directory deletion without warning about data loss.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The suggested replacement changes deletion of a directory's non-hidden entries into unconditional recursive deletion of the entire directory, including hidden files and subdirectories. An agent applying this general recipe to a project directory could destroy footage, plans, or deliverables without confirmation, making it particularly unsafe in a workflow that keeps valuable assets beside generated files.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/preferences.md (reported line 7)May include surrounding context.

md
## Visual identity
- **Style source: the OpenClaw Foundation announcement image** (`assets/style-reference.png`): a Michelangelo-style "Creation of Adam" with a lobster claw reaching toward a human finger, an ember spark between them, a near-black cracked-fresco background, lobster red and bone. His words: the claw-and-hand "doesn't have to be that exact image, it just has to be that style of imagery." Reuse the motifs (claw, spark, fresco, engraved bone linework, warm embers), never a copy of the image.
- **Drawn in code, not generated.** When offered AI image generation that spends his Figma credits, he chose procedural. Don't spend paid credits without asking. The procedural look has been accepted on every project since.
- **His old house style** (Avenir/Helvetica white block type on black, cartoon red lobster mascot, plain white name tags) is what gets replaced. When a video arrives with those burned in, swap them for the new style. Look for a clean source first.

## Title card

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/composite.py (reported line 59)May include surrounding context.

python
"-c:v:0", "libx264", "-preset", "medium", "-crf", "16", "-profile:v:0", "high", "-r:v:0", str(FR),
         "-colorspace:v:0", "bt709", "-color_primaries:v:0", "bt709", "-color_trc:v:0", "bt709", "-color_range:v:0", "tv"] + acodec + \
        ["-movflags", "+faststart", p["output"]]
subprocess.run(args, check=True)
print("wrote", p["output"])

Tainted flow: 'args' from open (line 28, file read) → subprocess.run (code execution)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/composite.py (reported line 59)May include surrounding context.

python
"-c:v:0", "libx264", "-preset", "medium", "-crf", "16", "-profile:v:0", "high", "-r:v:0", str(FR),
         "-colorspace:v:0", "bt709", "-color_primaries:v:0", "bt709", "-color_trc:v:0", "bt709", "-color_range:v:0", "tv"] + acodec + \
        ["-movflags", "+faststart", p["output"]]
subprocess.run(args, check=True)
print("wrote", p["output"])

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/contact_sheet.py (reported line 18)May include surrounding context.

python
if a.times:
    ts = [float(x) for x in a.times.split(",")]
else:
    dur = float(subprocess.run([FFPROBE, "-v", "error", "-show_entries", "format=duration", "-of", "csv=p=0", a.video],
                               capture_output=True, text=True).stdout)
    ts = [i * a.every for i in range(int(dur // a.every) + 1)]
tmp = tempfile.mkdtemp()

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/contact_sheet.py (reported line 25)May include surrounding context.

python
ims = []
for k, t in enumerate(ts):
    p = os.path.join(tmp, f"{k}.jpg")
    subprocess.run([FFMPEG, "-v", "error", "-y", "-ss", str(t), "-i", a.video, "-frames:v", "1", "-vf", f"scale={a.width}:-2", p])
    if os.path.exists(p):
        im = Image.open(p).convert("RGB"); d = ImageDraw.Draw(im)
        d.text((6, 4), f"{int(t // 60)}:{t % 60:05.2f}", fill=(255, 255, 0)); ims.append(im)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
88% confidence
Finding

The subprocess uses an argument list, so this is not shell-command injection. However, the audio-cleanup caller interpolates the unvalidated plan value audio.lufs into an FFmpeg filter graph passed through this wrapper; a malicious string can introduce additional filters, potentially accessing local files or network resources supported by the installed FFmpeg.

Content

Scanner excerpt · scripts/edit.py (reported line 42)May include surrounding context.

python
def run(args):
    subprocess.run([FFMPEG, "-v", "error", "-y"] + args, check=True)


segs, pos = [], LEAD

Tainted flow: 'args' from open (line 147, file read) → subprocess.run (code execution)

Medium
Category
Data Flow
Confidence
87% confidence
Finding

Passing plan-derived arguments to FFmpeg is generally intentional, and the reported file-read provenance is imprecise. Nevertheless, this sink receives a filter graph containing the unvalidated audio.lufs value, creating a concrete filter-language injection path from a malicious JSON plan, rather than shell execution.

Content

Scanner excerpt · scripts/edit.py (reported line 42)May include surrounding context.

python
def run(args):
    subprocess.run([FFMPEG, "-v", "error", "-y"] + args, check=True)


segs, pos = [], LEAD

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/edit.py (reported line 89)May include surrounding context.

python
# ---- 2. audio cleanup ----
def rd(p):
    b = subprocess.run([FFMPEG, "-v", "error", "-i", p, "-ac", "1", "-f", "f32le", "-"], capture_output=True, check=True).stdout
    return np.frombuffer(b, np.float32)

Tainted flow: 'p' from open (line 107, file read) → subprocess.run (code execution)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/edit.py (reported line 89)May include surrounding context.

python
# ---- 2. audio cleanup ----
def rd(p):
    b = subprocess.run([FFMPEG, "-v", "error", "-i", p, "-ac", "1", "-f", "f32le", "-"], capture_output=True, check=True).stdout
    return np.frombuffer(b, np.float32)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
93% confidence
Finding

audio.lufs is loaded from JSON without type or range validation and inserted directly into the -af filter expression. An attacker-supplied string can change the filter graph rather than merely set loudness, including introducing file- or network-reading filters where supported; shell=False does not prevent this interpreter-level injection.

Content

Scanner excerpt · scripts/edit.py (reported line 110)May include surrounding context.

python
L = lag(rd(raw), rd(f"{WORK}/den.wav"))
    src_for_ln = f"{WORK}/den.wav"
comp = f"atrim=start_sample={L},asetpts=PTS-STARTPTS," if L > 0 else ""
m = subprocess.run([FFMPEG, "-hide_banner", "-i", src_for_ln, "-af", comp + f"loudnorm=I={lufs}:TP=-2:LRA=11:print_format=json",
                    "-f", "null", "-"], capture_output=True, text=True).stderr
mj = json.loads(m[m.rindex("{"):m.rindex("}") + 1])
ln = (f"loudnorm=I={lufs}:TP=-2:LRA=11:measured_I={mj['input_i']}:measured_TP={mj['input_tp']}:"

Tainted flow: 'src_for_ln' from open (line 108, file read) → subprocess.run (code execution)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/edit.py (reported line 110)May include surrounding context.

python
L = lag(rd(raw), rd(f"{WORK}/den.wav"))
    src_for_ln = f"{WORK}/den.wav"
comp = f"atrim=start_sample={L},asetpts=PTS-STARTPTS," if L > 0 else ""
m = subprocess.run([FFMPEG, "-hide_banner", "-i", src_for_ln, "-af", comp + f"loudnorm=I={lufs}:TP=-2:LRA=11:print_format=json",
                    "-f", "null", "-"], capture_output=True, text=True).stderr
mj = json.loads(m[m.rindex("{"):m.rindex("}") + 1])
ln = (f"loudnorm=I={lufs}:TP=-2:LRA=11:measured_I={mj['input_i']}:measured_TP={mj['input_tp']}:"

Tainted flow: 'WORK' from open (line 32, file read) → open (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/edit.py (reported line 153)May include surrounding context.

python
"-c:a", "aac", "-b:a", "192k", "-movflags", "+faststart"] + extra + [OUT])
json.dump({"fps": str(FR), "lead_frames": LEAD, "thumbnail_frames": TH, "total_frames": TOTAL + TH,
           "segments": segs, "placement_frames": place, "denoiser_latency_samples": L, "residual_lag_samples": L2},
          open(f"{WORK}/edl.json", "w"), indent=1)
print("wrote", OUT, f"({(TOTAL + TH) / FPS:.3f} s)")

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/envelope.py (reported line 18)May include surrounding context.

python
segs = json.load(open(sys.argv[3])); t0, t1 = float(sys.argv[4]), float(sys.argv[5])
    print(" ".join(f"{w['start']:.2f}:{w['word'].strip()}" for s in segs for w in s.get("words", []) if t0 <= w["start"] <= t1))
    sys.exit()
pcm = subprocess.run([FFMPEG, "-v", "error", "-i", wav, "-ac", "1", "-ar", "16000", "-f", "f32le", "-"], capture_output=True, check=True).stdout
a = np.frombuffer(pcm, np.float32)
t0, t1 = float(sys.argv[2]), float(sys.argv[3]); st = float(sys.argv[4]) if len(sys.argv) > 4 else 0.02
out = []

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/qa.py (reported line 22)May include surrounding context.

python
a = ap.parse_args()
os.makedirs(a.qa_dir, exist_ok=True)

print(subprocess.run([FFPROBE, "-v", "error", "-count_packets", "-show_entries",
                      "stream=index,codec_type,codec_name,nb_read_packets,duration,r_frame_rate:stream_disposition=attached_pic",
                      "-of", "compact", a.video], capture_output=True, text=True).stdout)
lo = subprocess.run([FFMPEG, "-hide_banner", "-nostats", "-i", a.video, "-map", "0:a", "-af", "ebur128=peak=true", "-f", "null", "-"],

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/qa.py (reported line 25)May include surrounding context.

python
print(subprocess.run([FFPROBE, "-v", "error", "-count_packets", "-show_entries",
                      "stream=index,codec_type,codec_name,nb_read_packets,duration,r_frame_rate:stream_disposition=attached_pic",
                      "-of", "compact", a.video], capture_output=True, text=True).stdout)
lo = subprocess.run([FFMPEG, "-hide_banner", "-nostats", "-i", a.video, "-map", "0:a", "-af", "ebur128=peak=true", "-f", "null", "-"],
                    capture_output=True, text=True).stderr
print("loudness:", " | ".join(l.strip() for l in lo.splitlines() if l.strip().startswith(("I:", "Peak:"))))