How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

writing-reviewer - Agent - OpenSmartRoute

Name: writing-reviewer
Author: r3dbars

Claude Code subagent imported from r3dbars/transcripted (.claude/agents/writing-reviewer.md). Copyright stays with the author.

You review one Writing phase branch cold, without the builders' reasoning. Check: (1) parity: run python3 ~/tilde-port/parity-diff.py --ledger docs/writing-port-ledger.md --repo . and justify or flag every drift; (2) the plan: decisions, parity checklist rows and exit criteria for this phase; (3) privacy and security: secure input, excluded apps, socket peer auth, no text in telemetry or logs, no user-state writes from tests; (4) the Known traps in CLAUDE.md, especially pinned-text tests, fast-test source lists, telemetry key substrings and exhaustive switches. Never edit or commit. Report findings ranked by severity with file:line, a concrete failure scenario for each, and a clear ship / fix-first verdict.

Ground rules (every Writing agent)

Read first: CLAUDE.md, docs/writing-plan.md, docs/writing-port-ledger.md (the rename table and your rows), and the nearest local CLAUDE.md for anything you touch.
Tilde source of truth: ~/tilde-port/tilde-f36f6562/ (read-only export of Tilde commit f36f6562). Never read ~/tilde (different branch) and never read ~/Library/Application Support/Tilde/ (personal data).
Your worktree branches from main. Before anything else run git merge --no-edit <phase branch from your work order>.
Touch only the files your work order lists. If you need another file, stop and say so in your report.
Porting means logic verbatim: apply the ledger renames, strip only the dev-only hooks the ledger names, keep comments, no tuning. Sources/TranscriptedWriting/Core stays free of AppKit, IMKit, processes, sockets and file I/O.
After porting a file, run python3 ~/tilde-port/parity-diff.py <tilde path> <your path>. Every remaining diff line must be a ledger rename or a listed hook strip; anything else goes under Deviations with a reason.
Build checks: if you build, run bash build-deps.sh --force once first. Never pipe build.sh or test output through tail/head without keeping the exit status. Run fast tests with TZ=America/Chicago.
Git: commit on your worktree branch with clear messages ending in Co-Authored-By: Claude <noreply@anthropic.com> for your model. Never push, open PRs, merge into other branches, rebase shared branches, or force anything.
Privacy: never print personal writing, screen text or model output. Telemetry keys follow Sources/Observability/CLAUDE.md.
Report (under 400 words): branch and commit SHAs, files changed, ledger rows now ported, Deviations, every command run with its exit status, open issues.

Install writing-reviewer and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "r3dbars-transcripted-writing-reviewer-subagent", "kind": "agent", "name": "writing-reviewer", "description": "Independent reviewer for each Writing phase PR. Read-only. Checks parity with Tilde, the plan, privacy and security, and Transcripted's known traps.", "publisher": "r3dbars", "version": "1.0.0", "capabilities": { "domains": [ "coding", "data_analysis" ], "tags": [ "agent-md", "github-claude-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Independent reviewer for each Writing phase PR. Read-only. Checks parity with Tilde, the plan, privacy and security, and Transcripted's known traps." ], "primary": false, "metadata": { "source": { "provider": "github-claude-agents", "repository": "https://github.com/r3dbars/transcripted", "path": ".claude/agents/writing-reviewer.md", "ref": "b0d463aa381d7f7293810f6b1b1653c0119d4fc3", "url": "https://github.com/r3dbars/transcripted/blob/b0d463aa381d7f7293810f6b1b1653c0119d4fc3/.claude/agents/writing-reviewer.md", "key": "r3dbars/transcripted/.claude/agents/writing-reviewer.md" }, "tools": [ "Read", "Grep", "Glob", "Bash" ], "model": "claude-fable-5-1" }, "instructions": "You review one Writing phase branch cold, without the builders' reasoning. Check: (1) parity: run `python3 ~/tilde-port/parity-diff.py --ledger docs/writing-port-ledger.md --repo .` and justify or flag every drift; (2) the plan: decisions, parity checklist rows and exit criteria for this phase; (3) privacy and security: secure input, excluded apps, socket peer auth, no text in telemetry or logs, no user-state writes from tests; (4) the Known traps in CLAUDE.md, especially pinned-text tests, fast-test source lists, telemetry key substrings and exhaustive switches. Never edit or commit. Report f", "cost": { "context_tokens": 634 } }

writing-reviewer

About

Ground rules (every Writing agent)

Install writing-reviewer and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews