Templates
Launch profile
Pick a launch profile for a coding agent session before it starts, from a list the host has approved.
Pick a launch profile for a coding agent session before it starts, from a list the host has approved. Pattern: intent_routing. Model: jev-1.13.0. Outage rule: review.
When to use it
- A host starts coding agent sessions, such as Claude Code, and can set the model and effort at launch but not change them mid-session.
- Many tasks are mechanical and a few need the strongest model, and you want the pick made once, before the first turn, from profiles you have reviewed.
When not to use it
- The session is already running. A hook inside it cannot change its model; use model-tier for advice instead.
- The task names the model or the effort itself. Read it in code.
- You want the set to return a model name or a flag. It only picks a profile id; the host maps the id to a model and effort from its own reviewed file.
Questions
| Id | Type | Asks | Answers | Gating |
|---|---|---|---|---|
profile | choice | A developer is about to start a coding agent session in their repository with this task. Which launch profile fits it best? | light, standard, deep, deep_review, unclear | yes |
Reading the result
- The answer is a profile id:
light,standard,deepordeep_review. The host keeps a reviewed profiles file that maps each id to one or two sessions, each with a model and an effort, plus a default. The set never names a model or a flag. unclearis not a profile id, so it resolves to the default, as does any answer that is not a profile id in the host's file. So do an outage, an error or a timeout: a Bandwise problem never stops a session from starting.- Start in
shadow: the host always launches the default and records what the set would have picked. Incontrolledonly a high band pick is used. - A wrong
lightcosts the most (a weaker model on a hard task), so it needs a higher bar than the other options. deep_reviewnames a second session for an independent review. Whether a host starts it is the host's decision;bandwise launch --printonly reports it.- Outage rule:
onUnavailableisreview. During a System One outage the host launches the default.
Spec
Save it as bandwise/sets/launch-profile.json and edit it for your data.
{
"schemaVersion": 1,
"model": "jev-1.13.0",
"input": {
"schema": {
"type": "object",
"required": [
"task"
],
"properties": {
"task": {
"type": "string",
"maxLength": 8000
}
}
}
},
"stages": [
{
"id": "pick",
"questions": {
"profile": {
"type": "choice",
"instructions": {
"question": "A developer is about to start a coding agent session in their repository with this task. Which launch profile fits it best?",
"developer_task": "`task`"
},
"criteria": {
"light": "Mechanical work with every detail given and no design choice: rename a symbol, format files, move files, bump a version, run a given command, make a small edit spelled out in full, or find where something is defined.",
"standard": "Normal work with a clear goal in ordinary product code: build a described feature, fix a bug whose cause is known or easy to find, write tests for existing code, or refactor one module.",
"deep": "Work that needs design or investigation: choose an architecture, debug a failure whose cause is unknown, change code across many parts of the system, or settle a goal that is vague or open to several readings.",
"deep_review": "Deep work where a subtle mistake is costly, so an independent review of the result is worth a second session: authentication, permissions, secrets, payments, data deletion or migrations, production infrastructure, or a security fix.",
"unclear": "The task is too short or vague to tell what it needs, such as a single word or a greeting."
},
"meta": {
"label": "Launch profile"
}
}
}
}
],
"policies": {
"profile": {
"type": "choice",
"gating": true,
"thresholds": {
"high": 0.7,
"medium": 0.4
},
"perOption": {
"light": {
"high": 0.8,
"medium": 0.5
}
},
"actions": {
"high": {
"kind": "auto"
},
"medium": {
"kind": "review"
},
"low": {
"kind": "review"
}
}
}
},
"routes": [
{
"when": {
"q": "profile",
"eq": "light"
},
"output": "light"
},
{
"when": {
"q": "profile",
"eq": "deep"
},
"output": "deep"
},
{
"when": {
"q": "profile",
"eq": "deep_review"
},
"output": "deep_review"
}
],
"defaultRoute": "standard",
"savings": {
"comparatorModel": "claude-haiku-4-5",
"estOutputTokensPerQuestion": 30,
"kind": "decision"
},
"onUnavailable": "review"
}Example states
The expected outcome is what a person would decide. It is not a recorded model answer.
Rename across the repo
Expected: light: every detail is given and nothing needs a design choice.
{
"task": "Rename the function getUserName to getDisplayName everywhere, including the tests. No other changes."
}Described feature
Expected: standard: a clear feature in ordinary product code.
{
"task": "Add a Download CSV button to the invoices table. It should export the rows currently shown, with the same columns, and use the existing toCsv helper in src/lib/csv.ts."
}Intermittent failure in the auth flow
Expected: deep_review: the cause is unknown and the code decides who can sign in.
{
"task": "About one in twenty sign-ins fail with a 401 right after the token refresh, only in production. Find out why and fix it."
}Borderline cases
One case near the line for each question. Use them to test your wording before you trust the thresholds.
profile
A small, well-described edit that still needs one judgment call about where the check belongs: light or standard.
{
"task": "The settings page crashes when the user has no avatar. Add a null check so it shows the initials instead."
}Try it
Save an example state as state.json, then run the spec locally. Local mode makes no network call and needs no key; answers are synthetic unless a recorded fixture matches.
pnpm bandwise run --local bandwise/sets/launch-profile.json state.jsonBandwise is an independent product built on TypeSafe's System One models. It is not TypeSafe's documentation. For the System One models themselves, see docs.typesafe.ai.