subtitle-extractor
Analysis
The skill is mostly coherent for subtitle extraction, but it uses exported browser cookies for logged-in video access and makes an overly broad privacy claim about those cookies.
Findings (5)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Checks for instructions or behavior that redirect the agent, misuse tools, execute unexpected code, cascade across systems, exploit user trust, or continue outside the intended task.
Cookie files are read locally only, never transmitted externally. ... Bilibili requires a cookie file for all requests.
The wording can mislead users: cookies are used to authenticate requests to external video platforms, even if the skill is not shown sending them to an unrelated telemetry endpoint.
snapshot_download(f'pkufool/faster-whisper-{model_size}', local_dir=str(target_dir), ...)The transcription path can download Whisper model files from an external ModelScope repository without a pinned checksum. This is expected for transcription but is still a supply-chain dependency.
EXECUTION ORDER — NON-NEGOTIABLE ... Steps 1–4 in this skill MUST be completed in full before addressing any user request.
The skill uses strong ordering language that forces extraction and saving before the agent handles the user's final requested task. This is coherent with the subtitle-extraction purpose but should be visible to users.
Checks whether tool use, credentials, dependencies, identity, account access, or inter-agent boundaries are broader than the stated purpose.
for f in sorted(skill_dir.glob('*.txt')): ... if 'bilibili' in f.name.lower(): ... return str(f.resolve()) ... if cookie_file: cmd += ['--cookies', cookie_file]The code automatically discovers Bilibili cookie files in the skill directory and passes cookie files to yt-dlp. Platform cookies are account session credentials, while the registry metadata states there is no primary credential.
Checks for exposed credentials, poisoned memory or context, unclear communication boundaries, or sensitive data that could leave the user's control.
Agent saves subtitle to `outputs/` ... The subtitle file MUST be saved to disk
The skill intentionally persists extracted subtitle text locally before the agent summarizes, translates, or analyzes it.