cinematic-scroll

The fragment is a choreography/code generator, not malware. It has a meaningful injection risk because document-controlled values are embedded into generated JavaScript and HTML, especially raw text, attributes, inline SVG, URLs, and CSS. This can become stored XSS or generated-code injection when the JSON is untrusted and the output is opened or executed. No evidence of data theft, persistence, system damage, reverse shells, cryptomining, or credential harvesting is present. The apparent syntax defect may reflect truncation or an actual implementation error.

The fragment appears to be a legitimate browser-based review tool and contains no clear malware, credential theft, exfiltration, or sabotage behavior. It has a meaningful injection risk because manifest-controlled fields are interpolated into innerHTML and URL attributes without comprehensive validation or escaping. Risk is dependent on the integrity of manifest.json and the hosting origin; if the manifest is trusted and immutable, practical risk is lower. Recommended mitigations are DOM construction or context-appropriate escaping, strict URL allowlisting, iframe sandboxing, rel="noopener noreferrer", validation of IDs used in selectors and paths, and robust localStorage error handling.

The code is a local stylesheet extractor with no evident malicious behavior. Its primary security concern is path traversal/local file disclosure: attacker-controlled HTML can supply relative paths such as `../../sensitive-file` and cause the process to read files outside the expected stylesheet directory. The risk depends on whether `raw` and `file` are trusted and whether returned contents are exposed. Mitigate by resolving the target and enforcing that it remains within an approved directory, or by rejecting traversal components and validating the resolved path.

The code is a build-orchestration wrapper around the Claude CLI, not apparent malware. Its primary security concern is operational: an unvalidated directory selector and a privileged, permission-bypassed Claude session with Bash access. This could enable command execution or filesystem changes if prompt files, working-directory contents, or the invoked CLI are compromised. Restrict `c` to an allowlist, verify the target directory, avoid bypassPermissions where possible, and run the process in a genuinely isolated environment.