gitbackup-github-desktop

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEMETADATA_POISONING
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes repository and owner identifiers retrieved from the GitHub API and uses them to construct local filesystem paths.
  • Ingestion points: The repo.owner.login and repo.name fields are ingested in electron/services/backup-orchestrator.ts and passed to path.join for directory creation.
  • Boundary markers: None. The code snippets do not include logic to enforce boundaries or ignore potential instructions embedded in the metadata of the repositories being processed.
  • Capability inventory: The skill has the capability to create directories (fs.mkdirSync), write archives (tar.create), clone repositories (git.clone), and upload files to cloud storage (cloudService.uploadArchive).
  • Sanitization: Absent. The logic does not explicitly sanitize these identifiers to prevent path traversal characters (e.g., ..), relying instead on upstream API restrictions which could be bypassed via homoglyph or encoding techniques.
  • [CREDENTIALS_UNSAFE]: The repository cloning implementation in electron/services/git.service.ts temporarily embeds the plaintext GitHub Personal Access Token into the remote URL to authenticate the operation.
  • Evidence: const authenticatedUrl = repoUrl.replace("https://", "https://x-access-token:${token}@"); followed by a git clone command.
  • Risk: If the cloning process is interrupted or fails before the subsequent remote set-url command can clean up the URL, the plaintext token remains stored in the local .git/config file within the backup directory.
  • [METADATA_POISONING]: There is a discrepancy between the author identified in the platform context ('reason-machines'), the author credited in the markdown body ('ara.so'), and the owner of the referenced GitHub repository ('hiteshchoudhary'). This lack of clear authorship provenance can mislead users regarding the origin and trustworthiness of the software binaries.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 10:36 AM