Why this verdict
- suspicious: finding skill:templates/plate-playground-template/.agents/skills/browser-debug-setup:templates/plate-playground-template/.agents/skills/browser-debug-setup/SKILL.md:103:CRED
- suspicious: finding skill:templates/plate-playground-template/.agents/skills/browser-debug-setup:templates/plate-playground-template/.agents/skills/browser-debug-setup/SKILL.md:30:CRED
- suspicious: finding skill:templates/plate-playground-template/.agents/skills/browser-debug-setup:templates/plate-playground-template/.agents/skills/browser-debug-setup/SKILL.md:91:CRED
Security findings 3
Credential access
highContextual riskhigh confidenceThe skill's setup commands instruct the agent to read Chrome's 'Local State' account metadata and to copy the entire signed-in 'Profile 1' directory plus the Local State file into '~/.config/google-chrome-debug-profile', duplicating the user's Chrome credential store on disk outside its original location.
$HOME/Library/Application Support/Google/Chrome/Profile 1/
vector:code
Counter-argument considered
Refutation fails: the skill body openly instructs the agent to run rsync over the signed-in 'Profile 1' directory and cp Local State into ~/.config/google-chrome-debug-profile, duplicating credential-bearing browser data outside its original store; neither the declared description nor claims D3-D5 (all limited to debug-browser launch/connectivity on 9222) declares profile cloning or Local State reads, and no harm carve-out or mitigation covers duplicating a signed-in profile.
Credential access
highContextual riskmedium confidenceThe skill body instructs the agent to clone a signed-in Chrome profile into the dedicated debug --user-data-dir, copying credential-bearing browser data (cookies, saved logins) out of its original profile store for reuse by the debug browser.
Clone that separate signed-in Chrome profile into the dedicated debug
vector:instruction
Counter-argument considered
Refutation fails: the Rules section contains a concrete imperative to clone a signed-in Chrome profile (with its cookies/session auth) into the dedicated debug --user-data-dir; the declared description and claims D3-D5 cover only persistent debug-browser setup and dev-browser connectivity, so this credential-duplicating instruction is undeclared and fits no carve-out, even though the rule confines the source to a non-Default profile.
Credential access
highContextual riskmedium confidenceThe skill instructs the agent to run a python3 heredoc that reads Chrome's Local State to enumerate signed-in profiles (with Google account names), then rsync the entire signed-in Profile 1 directory (cookies, session and form data) plus Local State into ~/.config/google-chrome-debug-profile and launch Chrome there with CDP remote debugging on 127.0.0.1:9222, handing agent automation a browser running with the user's real signed-in identity — a profile-clone capability absent from the declared description.
p = pathlib.Path('~/Library/Application Support/Google/Chrome/Local State').expanduser()vector:instruction
Counter-argument considered
Refutation fails: the quoted heredoc reads Local State to enumerate profiles with Google account names, and the same block rsyncs 'Profile 1' plus Local State into the debug profile and launches CDP Chrome with the user's signed-in identity ('That keeps the signed-in identity while still satisfying Chrome's dedicated --user-data-dir requirement.'); the declared description and D3-D5 declare only debug-browser launch/connectivity, so this profile-clone capability is undeclared and not covered by any harm carve-out.
Open questions 1
What the analysis could not settle from the code alone.
[file request templates/plate-playground-template/.agents/rules/browser-debug-setup.mdc] The SKILL.md frontmatter cites metadata.skiller.source: .agents/rules/browser-debug-setup.mdc as its origin; does that rules file contain instructions steering the agent beyond the declared debug-browser setup task (hijack, tool poisoning, manipulation) or additional credential/exfiltration steps not shown in SKILL.md?
Agent hijacking
Coverage gaps 4
Parts of the artifact the static analysis could not fully examine.
context:unanalyzable_binary binary content is only byte-scanned
apps/www/public/favicon.ico
context:unanalyzable_binary binary content is only byte-scanned
templates/plate-playground-template/src/app/favicon.ico
context:unanalyzable_binary binary content is only byte-scanned
templates/plate-template/src/app/favicon.ico
context:manifest_unparseable yaml: YAML node count exceeds 20000
pnpm-lock.yaml