CodeQL still traced taint through the previous helpers: the endpoint
regex fallback returned source-derived text and the payload dict keys
flowed verbatim into the log. Resolve both by sourcing every logged
token from literal lookup tables — endpoints map to constant path
labels and payload keys map through a field-name allow-list, with
unknown keys/paths collapsing to <field>/<other>. No source-derived
string (key, value or id) reaches the logger.
Rebuild logged endpoints from constant path templates and reduce every
payload leaf to a JSON type token, so user/squad UUIDs and PII (email,
telegramId) can never reach the dry-run log as clear text. Resolves
CodeQL py/clear-text-logging-sensitive-data findings.
CodeQL flagged clear-text logging of sensitive information in the panel
dry-run logger: it dumped the full request payload (which can include
proxy credentials like trojanPassword/ssPassword and PII such as email
and telegramId) and the raw endpoint (embedding user UUIDs).
Recursively redact values under sensitive keys before building the
payload preview, and mask opaque id-like segments in logged endpoints.
Add a once-a-day anonymous heartbeat (PostHog) so maintainers can see
active installs and version/OS breakdowns. Self-hosted friendly: opt out
via TELEMETRY_ENABLED in .env or the Admin -> System toggle (applied
without a restart), or by clearing the endpoint/key.
- Share version resolution in bot/utils/app_version.py so the admin
sidebar and the beacon report the same build version
- TelemetryWorker sends an opaque install id plus coarse facts only
(version, OS/arch, python, locale, enabled providers, user-count
range); never tokens, domains or user data
- Register the worker in main_worker.py behind a Redis single-flight lock
- Expose TELEMETRY_* settings and an Admin -> System manifest toggle
- Document the payload and opt-out in docs/configuration/telemetry.md
- Cover bucketing, payload shape and anonymity with tests