Skip to content

Security: ReS0421/agent-config-bridge

Security

SECURITY.md

Security Policy

Agent Config Bridge projects executable agent configuration and unattended Schedule prompts into product and host-managed locations. Security reports are taken seriously.

Supported versions

Until the project reaches a stable release, security fixes are provided on the latest version of the main branch. Older snapshots may not receive fixes.

Reporting a vulnerability

Please report vulnerabilities through a private GitHub security advisory. Do not open a public issue for an undisclosed vulnerability.

Include only the information needed to reproduce and assess the problem:

  • Affected version or commit
  • Operating system, product, and surface
  • Minimal reproduction steps
  • Expected and observed behavior
  • Security impact and any known mitigations

Remove tokens, credentials, personal prompts, filesystem inventories, and other sensitive data. You should receive an acknowledgment through GitHub. The maintainers will coordinate validation, remediation, and disclosure there.

Security boundaries

Useful reports include unintended writes outside managed paths, path traversal, unsafe symlink handling, overwrite of unmanaged content or Settings leaves, command injection, secret leakage, generated-marketplace or Schedule-snapshot integrity failures, scheduler ownership confusion, and cases where a read-only plan operation changes state.

Catalog hooks are executable code. Running a hook that a user deliberately installed from an untrusted catalog is not, by itself, a vulnerability in Agent Config Bridge. Executing a hook that was not selected, bypassing a conflict or trust check, or altering the hook during synchronization may be a vulnerability.

Agent Config Bridge is not intended to synchronize authentication, session databases, caches, logs, trust decisions, private Desktop scheduler state, or complete Codex or Claude Code configuration homes.

The generated state_dir is non-secret operational state by design, but rendered packages, retained Skill backups, Settings digests, and Schedule snapshots reproduce or describe canonical catalog content. A catalog containing embedded secrets can therefore copy them into generated state. Do not attach catalog or state files to a report without reviewing and redacting them.

The current alpha does not claim an all-actions atomic transaction, target lock, automatic rollback, recovery log, product capability/version probe, arbitrary vendor artifact validation, or full installed-state validation. A report should distinguish an advertised ownership/integrity guarantee from one of these known limitations.

Raw filesystem permission bits are excluded from portable content identity; the bridge does not guarantee executable-mode preservation. Target names also key ownership state. Rename/delete targets only after reconciling the old identity with components = [], apply, and register; otherwise diagnostics report an orphan record and state-changing commands stop until it is restored and reconciled. Orphan state is never adopted or removed automatically.

There aren't any published security advisories