Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed YouTube digest workflow, but users should review the configured recipients, publishing destinations, transcript retention, and optional automation before running it.
Before installing or running, change the example recipient and sharing settings, decide which outputs should be enabled, review the first generated digest before sending, and set a retention policy for raw transcripts and scheduled runs.
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
Do not attach the transcript unless someone explicitly asks for it—email only the digest body linked above.
GOG_KEYRING_PASSWORD=... gog gmail send \
--to "{{config.outputs.email.to}}" \
--subject "Nate B Jones Digest — {{DATE}}" \
--body-file summary.txt \
The manifest says this skill monitors Nate B Jones's YouTube channel, retrieves transcripts, summarizes videos, and distributes digests. This file instead contains a transcript about Claude's Chrome extension performing general browser automation, email handling, Google Drive organization, testing websites, and multi-tab workflows, which does not describe or implement the declared digest pipeline.
The skill explicitly instructs sending generated digests through email, chat, and document platforms, which causes externally sourced content to leave the local environment without any user-facing warning or approval checkpoint. Even though the source is a public YouTube channel, transcripts and summaries may still include personal data, copyrighted material, or unexpected sensitive content, and broad publication increases privacy and compliance risk.
The skill directs operators to retain raw transcripts in logs for traceability, but provides no retention limits, access controls, or warning that transcripts can contain sensitive or regulated content. Persistent storage of full transcripts increases the blast radius of accidental disclosure and creates unnecessary long-term data retention risk.
The transcript promotes an agent model where 'anything that is in the browser that you can do, Claude will see' and encourages broad automation across websites, email, CRM data, and other logged-in contexts. In a skill ecosystem, this normalizes overbroad delegation and could lead downstream implementations to grant excessive permissions or operate on sensitive sites without strong scoping, increasing the risk of data exposure or unauthorized actions.
The config sets prefer_languages to en, which imposes an English-language preference. Under the policy rules, forcing a specific language without user opt-in or documented justification is a natural-language policy concern.
The document declares lang="en", which enforces an English-language locale in the natural-language content without any visible opt-in or explanation. Under the policy, locale constraints should either be optional for the user or clearly justified as region-specific.
Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.
surface. One summary might focus on a
tangential update that might not be
super relevant. Essentially when you
start to expand the scope of the task in
a single workflow right now claude in
Chrome does not do a perfect job of
recognizing salience of recognizing
The phrase 'don't be dumb' is a policy-relevant natural-language issue because it uses demeaning language in user-facing guidance. Even though it appears in a transcript, this wording is not appropriate for organizational skill content and may violate standards for respectful communication.
The prefer_languages: ["en"] setting hard-codes a language preference in a natural-language-facing configuration. This can violate the language/locale policy when users are not given an opt-in choice or a documented reason for restricting transcript language to English.
No suspicious patterns detected.