The Agent Skills Directory

Subprocess Execution for Project Resolution: The skill uses subprocess.check_output in scripts/gitlab.py to resolve the target GitLab project from the local Git configuration. This is implemented using an argument list with shell=False, preventing shell injection. The output is further validated against path traversal patterns before being used in API requests.
Hardened OAuth Implementation: The skill implements OAuth 2.0 using both Authorization Code (PKCE) and Device Authorization flows. The loopback listener for callbacks validates the Host header and uses secrets.compare_digest for anti-forgery state checks. It also includes validation logic to prevent UI-based misdirection during device login.
Credential Redaction and Sanitization: A comprehensive redaction mechanism is implemented in scripts/gitlab.py using multiple regex patterns to scrub access tokens, private tokens, and other secrets from stdout, stderr, and audit logs. Structured JSON output is also recursively sanitized to remove sensitive fields.
Secure Credential Persistence: In scripts/_gitlab_credentials.py, OAuth profiles are stored in a dedicated directory with POSIX mode 0700 and files with mode 0600. The skill uses os.open with O_NOFOLLOW and verifies file ownership and permissions before reading or writing credentials. It also implements a fail-closed policy on Windows systems where these granular permissions cannot be enforced.
Transport Security: The skill utilizes a custom urllib.request opener that includes a _NoRedirect handler. This prevents short-lived tokens from being leaked to third-party hosts via HTTP 30x redirects. All API traffic is enforced over TLS except for explicitly allowed loopback development environments.
Fail-Closed Audit Logging: An optional audit logging feature is provided which follows a fail-closed, write-ahead pattern. The skill attempts to record the operation before execution; if the audit sink is unavailable, the operation is aborted.

gitlab