Skip to content

PRIVACY.md

Latest commit

 

History

History
80 lines (52 loc) · 5.73 KB

PRIVACY.md

File metadata and controls

80 lines (52 loc) · 5.73 KB
Raw

Privacy

Clean Your Data is designed for local, metadata-first analysis. The bundled scanner and GUI server do not make network requests and do not upload reports. An agent may download this public repository, but the scan itself runs on the user's machine. The optional GUI/TUI cleanup loop can move an exact user-confirmed path to the local system Trash; that action and its undo record stay on the machine.

Installing the Agent Skill or Python package requires network access and writes to the Agent's Skill directory, a package-tool environment, or a package runner cache. The Skill pins its fallback package command to a release tag and asks before that download. This installation traffic is separate from local scanning.

Local Browser GUI

  • The GUI binds only to 127.0.0.1 on a random port by default.
  • It validates the loopback Host; its JSON API requires a random per-run token and rejects cross-origin browser requests.
  • Public API responses omit absolute local paths and internal _local_path fields.
  • The browser receives redacted paths, bounded metadata, and only the preview the user explicitly selects.
  • Closing the server ends the session. Agent conversations in the browser are not persisted by Clean Your Data.

Default Redaction

By default:

  • The home directory is rendered as ~.
  • Git origin URLs are omitted.
  • Reports do not include file contents.
  • Reports do not include chat, mail, browser, or document contents.

Avoid sharing reports created with:

--no-redact
--include-git-origins

Data The Scanner May Collect

  • Directory paths, usually home-relative.
  • Allocated byte counts from du or a standard-library fallback.
  • Modification times.
  • Known category labels such as workspace, app-state, cloud-sync, cache, and inbox.
  • Optional Git dirty-entry counts from git status --porcelain.

One-Path Explanations

cyd why PATH creates a bounded, read-only evidence packet for one selected path. It may inspect names, directory shape, allocated size, timestamps, project-marker existence, Git index membership for a selected file, and local prospective trace records. It does not read the selected file body, parse project manifests, enumerate Git dirty state, or follow a selected symbolic-link target.

Relationship traversal stays within the selected directory and reports its scope and counts. Ancestor lookup checks up to ten levels for project-marker names; it does not parse those marker files. Cleanup review may query the Git index for one selected path so tracked content can be blocked without reading working-tree file bodies.

JSON and text output redact the configured home prefix to ~ and reduce paths outside that home to <external>/NAME. Project names and home-relative folder names can still be identifying. The command never moves a file and always returns action_gate.authorizes_move as false.

Data The Scanner Should Not Collect

  • Message text.
  • Browser history.
  • Email bodies.
  • Document text.
  • Secrets, tokens, keychains, or credentials.
  • Source file contents.

Optional AI Commands

AI is disabled or configured by the user. Clean Your Data can call an already authenticated Codex CLI or a trusted custom stdin/stdout command. It stores the selected mode and custom-command arguments verbatim in ~/.clean-your-data/ai-config.json, with user-only file permissions where supported. The GUI reports that a command is configured but does not echo its arguments through the local browser API; cyd config ai --show is the explicit terminal view. There is no API-key field: never put credentials in command arguments, and keep them in the provider's environment or credential store.

The opt-in tracer also stores the traced command and its arguments verbatim in the local ~/.clean-your-data/provenance.sqlite3 database. Do not put credentials directly in a traced command line. The state directory and database use user-only permissions where the operating system supports them.

The stdin prompt constructed by Clean Your Data contains a redacted path, name, kind, size, modified time, category, measurement status, project-marker names, bounded relationship summaries, and any matching local prospective trace association. A trace association exposes the captured executable's basename, event type, and timestamps, but not its full command arguments. Clean Your Data does not place the selected file preview or file contents in that prompt. The built-in Codex mode uses a read-only, ephemeral sandbox. A custom command still has its own operating-system permissions and may use its own network connection or credential store; configure only a trusted command and review that provider's privacy behavior separately.

Sharing Reports

Before sharing a report publicly:

  • Prefer Markdown over JSON.
  • Review path names for sensitive project names.
  • Do not include Git origin URLs.
  • Do not include absolute home paths.
  • Do not include full raw reports from work machines without review.

JSON snapshots are intended for local before/after comparison. They can still contain home-relative project and folder names, so treat them as private until reviewed.

Cleanup History

The TUI stores the original and Trash paths needed for undo in ~/.clean-your-data/cleanup-history.json. This file is local-only and is not included in JSON/Markdown reports or sent to the coding agent. Review or remove it according to your local retention needs.

Bookmarks, recent paths, local tags, and the last workspace selection are stored locally in ~/.clean-your-data/workspace-state.json. Tabs exist only for the current TUI session. Deep path analysis uses names, extensions, sizes, project markers, and directory shape; it does not read file contents or calculate hashes. These state files and analysis results are not included in reports or uploaded.