Sloth-ninja stops Clio logins from sending solicitors to localhost

A production safeguard now blocks a broken Clio connection flow before a solicitor completes consent and lands on an unreachable address.

integrationsecurity

After a real deployment incident, Sloth-ninja has tightened the configuration behind Clio, the legal practice-management platform. Previously, a missing public backend address could quietly produce a sign-in link that sent a user back to their own computer rather than the live service.

The fork now treats that production setup as unavailable and returns a clear not-configured response instead of issuing a faulty authorisation link. It also records a redacted startup warning and documents the diagnostic pattern: provider consent succeeds, then the browser cannot reach the callback address. Crucially, the guidance calls for testing a full sign-in and return journey from a non-development machine, not just checking whether the service is online.

So what Teams connecting practice systems to legal-AI tools should care because a healthy-looking deployment can still fail at the exact moment a user grants access.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

1 commit from Sloth-ninja/JessicaOSS, oldest first. Source extracted verbatim from the harvested git log.

SHA Subject Author Date

6fc6afa6 Harden OAuth callback base: fail closed in production sloth-ninja 2026-08-04 ↗ GitHub

SHA	Subject	Author	Date
`6fc6afa6`	Harden OAuth callback base: fail closed in production	sloth-ninja	2026-08-04	↗ GitHub
commit body Incident 03/08/2026: the deployed backend minted Clio authorize URLs with redirect_uri=http://127.0.0.1:3001/... because neither API_PUBLIC_URL nor BACKEND_URL was set on Fly, so clioBackendBaseUrl() fell through to its dev literal. A pilot solicitor's consent succeeded at Clio then bounced to 127.0.0.1 (ERR_CONNECTION_REFUSED). Fixed operationally by setting API_PUBLIC_URL; this makes the failure loud and early instead of silent. - clio/config.ts: new productionCallbackBaseMissing() - in NODE_ENV=production when the resolved base is the 127.0.0.1 fallback, clioConfigured() returns false (same state as missing credentials), so the start route returns the fixed "not configured" error and never mints a localhost redirect. A redacted safeErrorLog fires once per boot naming API_PUBLIC_URL. Dev/test unchanged. - MCP checked: the interactive authorize redirect derives from the Host header (routes/user.ts backendPublicUrl) - different fallback class, unaffected on Fly, left unchanged. The refresh-path helper mcpOAuthCallbackUrl() shares the localhost fallback but is covered by API_PUBLIC_URL being required in prod. - .env.example + CLAUDE.md env registry: API_PUBLIC_URL/BACKEND_URL documented as REQUIRED in production for OAuth callback flows (Clio + MCP). - DURABLE_LESSONS: env-derived callback bases degrade silently; guard fails closed; OAuth deploy verification needs a real round-trip from a non-dev machine. Includes the debugging signature. - Tests: 4 guard cases (prod+fallback -> not configured/no URL; prod+base -> normal; dev+fallback -> unchanged). Backend vitest 604/604, tsc + prettier clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

commit body

Incident 03/08/2026: the deployed backend minted Clio authorize URLs with
redirect_uri=http://127.0.0.1:3001/... because neither API_PUBLIC_URL nor
BACKEND_URL was set on Fly, so clioBackendBaseUrl() fell through to its dev
literal. A pilot solicitor's consent succeeded at Clio then bounced to
127.0.0.1 (ERR_CONNECTION_REFUSED). Fixed operationally by setting
API_PUBLIC_URL; this makes the failure loud and early instead of silent.

- clio/config.ts: new productionCallbackBaseMissing() - in NODE_ENV=production
  when the resolved base is the 127.0.0.1 fallback, clioConfigured() returns
  false (same state as missing credentials), so the start route returns the
  fixed "not configured" error and never mints a localhost redirect. A redacted
  safeErrorLog fires once per boot naming API_PUBLIC_URL. Dev/test unchanged.
- MCP checked: the interactive authorize redirect derives from the Host header
  (routes/user.ts backendPublicUrl) - different fallback class, unaffected on
  Fly, left unchanged. The refresh-path helper mcpOAuthCallbackUrl() shares the
  localhost fallback but is covered by API_PUBLIC_URL being required in prod.
- .env.example + CLAUDE.md env registry: API_PUBLIC_URL/BACKEND_URL documented
  as REQUIRED in production for OAuth callback flows (Clio + MCP).
- DURABLE_LESSONS: env-derived callback bases degrade silently; guard fails
  closed; OAuth deploy verification needs a real round-trip from a non-dev
  machine. Includes the debugging signature.
- Tests: 4 guard cases (prod+fallback -> not configured/no URL; prod+base ->
  normal; dev+fallback -> unchanged). Backend vitest 604/604, tsc + prettier
  clean.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-1069.md from inside the repo you want the changes in.

⬇ Download capture-thread-1069.md