Batch Redaction

To redact an existing directory tree, drive the HTTP API from a client that walks the tree and POSTs each file. Philter Router stays a stateless gateway: it routes and forwards, the client handles the filesystem. A reference client, scripts/redact-tree.sh, ships with the repository and needs only curl and find.

scripts/redact-tree.sh --in /data/src --out /data/redacted --url https://localhost:8080 --insecure

It walks --in recursively and writes each redacted file to --out under the same relative path, so the output mirrors the input tree. The router selects the policy and engine per file.

Option	Purpose
--in DIR	Input directory to walk (required).
--out DIR	Output directory; the input tree is mirrored here (required).
--url URL	Router base URL (default https://localhost:8080).
--jobs N	Files to process at once (default 4).
--api-key KEY	Philter API key, sent as Authorization and forwarded to Philter.
--policy NAME	Force a policy, overriding routing.
--context NAME	Philter context.
--force	Re-redact even when the output already exists.
--insecure	Skip TLS verification, for the router's self-signed certificate.

Resumable

A file whose output already exists is skipped, so an interrupted run can be re-run to finish the remainder. --force re-redacts everything. There is no separate state to manage: the presence of the output file is the record of what is done. Output is written atomically (a temporary file is moved into place only on success), so a failed or interrupted file is never mistaken for completed.

Failures

A file that the router rejects is logged to stderr with its HTTP status and left without an output file; the run continues. The script prints a redacted / skipped / failed summary and exits non-zero if any file failed, so it works in a pipeline or CI job.

Other clients

The script is only a convenience. Any tool that can POST bytes works. For example, with GNU parallel:

find /data/src -type f -print0 \
  | parallel -0 -j8 curl -sS -k -X POST --data-binary @{} \
      'https://localhost:8080/api/filter?filename={/}' -o /data/redacted/{/}

Directory-based routing needs the source directory: the script sends it in the X-Source-Directory header (an uploaded file has no path of its own). A hand-rolled client that relies on directory routes should do the same.