The public repository contains software, blank templates, and fictional examples. Personal context is created separately, with no remote by default. The local demo is ignored under .local/; a .gitignore rule cannot remove data already committed to Git history. If creating another public distribution, export only reviewed software/templates and begin a fresh history.

The dashboard binds to 127.0.0.1, rejects writes, and reads committed canonical records from one selected Git repository. It excludes restricted records and session exports. private records are visible on this local dashboard; this label is not encryption or access control. Do not expose the server through a public host, tunnel, proxy, or hosted deployment with personal data.

The dashboard does not make external AI requests. If you use an AI assistant to read context, that assistant's own data handling settings apply. An approved context proposal is a workflow requirement; the application does not provide a generic apply-and-sync engine. The separate capture CLI enforces a narrower path: selected facts queue outside canonical context by default; an explicitly enabled, committed policy permits one new private journal entry and a local commit. It never rewrites existing records, enables its own policy, executes Git hooks, or pushes. Source and secret checks are defense in depth, not a complete personal-data detector.

The global skill uses an explicit local binding, never the current project or demo as a silent fallback. Its local queue holds selected private facts with restrictive filesystem permissions; queued facts are not canonical records and do not appear in the dashboard. Capture checks the selected payload, bounded policy/Git metadata, and canonical headers in selected link scopes. It excludes ignored file contents and transcript stores. A skill is not a sandbox: the local assistant's filesystem permissions and the owner's context policy still apply.

Never put credentials, private keys, government identifiers, complete account records, or third-party confidential material in context at any privacy level. Do not publish suspected private data in a public issue. Use the repository's private vulnerability reporting channel if enabled, or describe the issue without including sensitive content and request a private reporting route.