Security audit
Security checks for vulnerabilities and agentic risk
The skill is mostly a coherent transcript workflow, but it under-discloses external transcript sharing and asks for broad persistent memory logging.
Review before installing. Use this only for transcripts you are comfortable sending to external transcription and translation services, remove or ignore the durable memory-log instruction, and prefer explicit user approval before running the Google translation helper or writing persistent audit state.
scripts/translate_transcript_google.py:16Transcript Content Is Disclosed to a Third-Party Translation Service Through URL Query Parameters
SKILL.md:53Broad Instruction Permits Unnecessary Modification of Persistent Agent Memory
The skill claims to perform transcript generation, correction, translation, diarization handling, and packaging, but the detected implementation behavior reportedly only scaffolds folders and writes a manifest. This mismatch is dangerous because operators may trust the skill to have completed sensitive processing steps when it has not, leading to incorrect outputs, unsafe workflow assumptions, and possible chaining with additional tools under false pretenses.
The skill describes actions that inherently require network access, file reads, and file writes, but it does not declare any explicit tool scope or permissions boundary. This is dangerous because consumers or agents may execute the workflow with broader-than-expected privileges, reducing visibility and increasing the chance of unintended data access, external downloads, or filesystem modification.
The script sends transcript content to Google's public translation endpoint over the network without any explicit warning, consent prompt, or privacy notice. Because interview transcripts may contain sensitive personal, business, or confidential information, this can cause unintended third-party data disclosure in a workflow that appears to be a local processing utility.
Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.
else:
out_lines.append(f"{prefix}{translated}")
out_path.write_text('\n'.join(out_lines) + ('\n' if lines and not lines[-1].endswith('\n') else ''), encoding='utf-8')
if __name__ == '__main__':
main()
The code performs a file write to the provided output path, but the script contains no confirmation, logging, or descriptive comment/docstring that makes this side effect explicit. Under the rule, file writes in code should have some visible disclosure unless already clearly documented elsewhere.