agency-docs-updater

SUSPICIOUS. The overall workflow matches the stated purpose of publishing lab meetings, and most capabilities are proportionate, but the skill performs high-impact autonomous publishing actions and forwards credentials/work product through a separate youtube-uploader dependency that could not be verified from the provided evidence. That unverifiable tool provenance is the main reason for the elevated risk, not confirmed malicious intent.

The code appears to be a legitimate Zoom-to-documentation publishing automation script and contains no clear evidence of malware or intentional data theft. The main security risk is the unattended invocation of Claude with --dangerously-skip-permissions, combined with configurable executable and repository paths. This grants a headless workflow broad filesystem and command authority, so the referenced Python script, Claude installation, and agency-docs-updater workflow should be independently trusted and integrity-checked.

No clear malicious behavior is present; the code is consistent with YouTube content-management automation. The main security risk is command/code injection caused by direct interpolation of shell variables into python3 -c source, plus the sensitivity of token.pickle. Use safe parameter passing and protect the OAuth token before using these examples with untrusted input.