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.

numen-retro - Prompt - OpenSmartRoute

Name: numen-retro
Author: THEjacob1000

Prompt file imported from THEjacob1000/dotfiles (.codex/prompts/numen-retro.md). Copyright stays with the author.

/retro

Retrospective on a coding session that proposes fixes to the agent environment (doctrine, rules, skills, checks, reviewer, tooling) so the next run goes better. Use when the user says "retro", "retrospective", or asks what to change after a session went badly.

Retro

You are improving the environment future agents run in, not the code the session produced. Read the session, find where it lost time or got something wrong, and propose the smallest change to the environment that would have prevented it.

Read the session

Default to the current session. For another one, read its transcript:

OMP: ~/.omp/agent/sessions/<cwd relative to $HOME, / as ->/*.jsonl, newest first. The archive global in eval searches past sessions and prompts.
Claude Code: ~/.claude/projects/<absolute cwd, / as ->/*.jsonl.
Codex: ~/.codex/sessions/.

Note each wrong turn: a file it couldn't find, a mistake a check would have caught, a rule it broke, an expensive call, a fact it never had.

Where a fix can live

Every fix lands in the numen harness, so it reaches every tool through numen sync. Pick the cheapest home that would have worked:

Automated check: a lint, type, test or numen check rule. Best when the mistake is mechanical; it costs no context at all.
Reviewer rule: a line in harness/canonical/agents/code-reviewer.md. The reviewer reads a diff with little context pressure, so standards belong there, not in the implementer's prompt.
Path-scoped rule: harness/canonical/rules/<name>.md, loaded only when the agent touches the files it claims.
Skill: harness/canonical/skills/, or a pin in harness/canonical/skills.toml. Only its description sits in context, so it suits a workflow reached on demand.
Doctrine: harness/canonical/doctrine.md or a repo AGENTS.md. Always loaded, so it costs every turn of every session; use it for navigation pointers and hard boundaries only.
Tooling or information access: an OMP extension under harness/canonical/omp/extensions/, an MCP entry in harness/canonical/mcp.toml, or a ripwire verb the agent should have used.

Also look for

Doctrine or AGENTS.md lines that changed nothing in this session's behaviour and could move into a rule, a skill or a check.
Tool calls that returned far more than the agent needed.
Instructions the agent followed that contradict another file.

Report

List the candidates, most severe first. For each, give the session evidence (what happened, and where in the transcript), the file to change, and the exact change. Don't apply them until the user picks.

Install numen-retro 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.

# /retro Retrospective on a coding session that proposes fixes to the agent environment (doctrine, rules, skills, checks, reviewer, tooling) so the next run goes better. Use when the user says "retro", "retrospective", or asks what to change after a session went badly. # Retro You are improving the environment future agents run in, not the code the session produced. Read the session, find where it lost time or got something wrong, and propose the smallest change to the environment that would have prevented it. ## Read the session Default to the current session. For another one, read its transcript: - OMP: `~/.omp/agent/sessions/<cwd relative to $HOME, / as ->/*.jsonl`, newest first. The `archive` global in eval searches past sessions and prompts. - Claude Code: `~/.claude/projects/<absolute cwd, / as ->/*.jsonl`. - Codex: `~/.codex/sessions/`. Note each wrong turn: a file it couldn't find, a mistake a check would have caught, a rule it broke, an expensive call, a fact it never had. ## Where a fix can live Every fix lands in the numen harness, so it reaches every tool through `numen sync`. Pick the cheapest home that would have worked: - **Automated check**: a lint, type, test or `numen check` rule. Best when the mistake is mechanical; it costs no context at all. - **Reviewer rule**: a line in `harness/canonical/agents/code-reviewer.md`. The reviewer reads a diff with little context pressure, so standards belong there, not in the implementer's prompt. - **Path-scoped rule**: `harness/canonical/rules/<name>.md`, loaded only when the agent touches the files it claims. - **Skill**: `harness/canonical/skills/`, or a pin in `harness/canonical/skills.toml`. Only its description sits in context, so it suits a workflow reached on demand. - **Doctrine**: `harness/canonical/doctrine.md` or a repo `AGENTS.md`. Always loaded, so it costs every turn of every session; use it for navigation pointers and hard boundaries only. - **Tooling or information access**: an OMP extension under `harness/canonical/omp/extensions/`, an MCP entry in `harness/canonical/mcp.toml`, or a ripwire verb the agent should have used. ## Also look for - Doctrine or AGENTS.md lines that changed nothing in this session's behaviour and could move into a rule, a skill or a check. - Tool calls that returned far more than the agent needed. - Instructions the agent followed that contradict another file. ## Report List the candidates, most severe first. For each, give the session evidence (what happened, and where in the transcript), the file to change, and the exact change. Don't apply them until the user picks.

{ "prompt": "# /retro\n\nRetrospective on a coding session that proposes fixes to the agent environment (doctrine, rules, skills, checks, reviewer, tooling) so the next run goes better. Use when the user says \"retro\", \"retrospective\", or asks what to change after a session went badly.\n\n# Retro\n\nYou are improving the environment future agents run in, not the code the session produced. Read the session, find where it lost time or got something wrong, and propose the smallest change to the environment that would have prevented it.\n\n## Read the session\n\nDefault to the current session. For another one, read its transcript:\n\n- OMP: `~/.omp/agent/sessions/<cwd relative to $HOME, / as ->/*.jsonl`, newest first. The `archive` global in eval searches past sessions and prompts.\n- Claude Code: `~/.claude/projects/<absolute cwd, / as ->/*.jsonl`.\n- Codex: `~/.codex/sessions/`.\n\nNote each wrong turn: a file it couldn't find, a mistake a check would have caught, a rule it broke, an expensive call, a fact it never had.\n\n## Where a fix can live\n\nEvery fix lands in the numen harness, so it reaches every tool through `numen sync`. Pick the cheapest home that would have worked:\n\n- **Automated check**: a lint, type, test or `numen check` rule. Best when the mistake is mechanical; it costs no context at all.\n- **Reviewer rule**: a line in `harness/canonical/agents/code-reviewer.md`. The reviewer reads a diff with little context pressure, so standards belong there, not in the implementer's prompt.\n- **Path-scoped rule**: `harness/canonical/rules/<name>.md`, loaded only when the agent touches the files it claims.\n- **Skill**: `harness/canonical/skills/`, or a pin in `harness/canonical/skills.toml`. Only its description sits in context, so it suits a workflow reached on demand.\n- **Doctrine**: `harness/canonical/doctrine.md` or a repo `AGENTS.md`. Always loaded, so it costs every turn of every session; use it for navigation pointers and hard boundaries only.\n- **Tooling or information access**: an OMP extension under `harness/canonical/omp/extensions/`, an MCP entry in `harness/canonical/mcp.toml`, or a ripwire verb the agent should have used.\n\n## Also look for\n\n- Doctrine or AGENTS.md lines that changed nothing in this session's behaviour and could move into a rule, a skill or a check.\n- Tool calls that returned far more than the agent needed.\n- Instructions the agent followed that contradict another file.\n\n## Report\n\nList the candidates, most severe first. For each, give the session evidence (what happened, and where in the transcript), the file to change, and the exact change. Don't apply them until the user picks.", "variables": [], "notes": "Retrospective on a coding session that proposes fixes to the agent environment (doctrine, rules, skills, checks, reviewer, tooling) so the next run goes better. Use when the user says \"retro\", \"retrospective\", or asks what to change after a session went badly.", "metadata": { "source": { "provider": "github-codex-prompts", "repository": "https://github.com/THEjacob1000/dotfiles", "path": ".codex/prompts/numen-retro.md", "ref": "8cb1feccf64642f418d3c1670e2f7119c201bdee", "url": "https://github.com/THEjacob1000/dotfiles/blob/8cb1feccf64642f418d3c1670e2f7119c201bdee/.codex/prompts/numen-retro.md", "key": "THEjacob1000/dotfiles/.codex/prompts/numen-retro.md" } } }

numen-retro

About

/retro

Retro

Read the session

Where a fix can live

Also look for

Report

Install numen-retro 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