/set-models

/scv:set-models

Apply a model policy to all SCV commands and remember the choice in .env.

Language preference

Resolve the user's preferred language with this priority, then use it for ALL user-facing output (AskUserQuestion text, summaries):

1. ~/.claude/settings.json (or project .claude/settings.json / .claude/settings.local.json) — language key (Claude Code official).
2. Project .env — SCV_LANG.
3. Auto-detect from the user's most recent message language.
4. Default to English.

Technical identifiers stay as-is: policy names (recommended, all-opus, all-sonnet, all-haiku, session-default), env var names (SCV_MODEL_POLICY, SCV_LANG), file paths, model names (haiku, sonnet, opus).

What each policy does

Policy	Effect
recommended	Per-command mapping: status/report/update/install-deps → haiku · sync/help/promote/codegen/regression/work → opus. Matches v0.11.5 baseline. Recommended default.
all-opus	Every command uses opus. Highest quality, highest cost.
all-sonnet	Every command uses sonnet. Balanced.
all-haiku	Every command uses haiku. Lowest cost; some reasoning-heavy commands may degrade.
session-default	Removes the model: line entirely. Each command runs on whatever model the current Claude Code session is using.

Step 1 — Resolve the chosen policy

If $ARGUMENTS is exactly one of recommended, all-opus, all-sonnet, all-haiku, session-default, treat that as the chosen policy and skip to Step 2.

Otherwise, fire AskUserQuestion with these 5 options (the first is the recommended default — put it first in the list):

Question: "Which model policy do you want for SCV commands?"
options:
[1] "recommended (Recommended)"
    description: "Per-command mapping. status/report/update/install-deps run on haiku (cheap, fast). sync/help/promote/codegen/regression/work run on opus (heavy reasoning). Matches v0.11.5 baseline."
[2] "all-opus"
    description: "Every SCV command uses opus. Highest quality. Significantly higher cost — use only if your team's budget allows."
[3] "all-sonnet"
    description: "Every SCV command uses sonnet. Balanced between cost and quality. Good middle ground if you don't want per-command policy."
[4] "all-haiku"
    description: "Every SCV command uses haiku. Lowest cost. Reasoning-heavy commands (work, codegen, promote) may produce lower-quality output."
[5] "session-default"
    description: "Remove model: lines entirely. Each command inherits the model of the current Claude Code session. Use this if you want full manual control."

Map the user's selection to the corresponding lowercase identifier: recommended / all-opus / all-sonnet / all-haiku / session-default. Call this value POLICY from now on.

Step 2 — Apply the policy via the helper script

Run exactly one Bash call with the resolved POLICY value substituted as a real string (never as a <placeholder>):

"${CLAUDE_PLUGIN_ROOT}/scripts/apply-model-policy.sh" --policy <POLICY>

For example, if the user picked all-haiku, the actual call must be:

"${CLAUDE_PLUGIN_ROOT}/scripts/apply-model-policy.sh" --policy all-haiku

The script will print per-file changes (status: haiku -> opus / report: ok (haiku) / etc). Capture stdout and surface a short summary to the user — one line per file that changed (skip the ok lines).

Step 3 — Persist the choice in .env

The choice must survive /scv:sync (which re-renders frontmatter from templates). Persist it:

1. Read the project root's .env file (path: $PWD/.env).
   • If .env does not exist: create it with the single line SCV_MODEL_POLICY=<POLICY>.
2. If .env already contains a line matching ^SCV_MODEL_POLICY=:
   • Use Edit to replace the existing line's value with the new <POLICY>.
3. If .env exists but has no SCV_MODEL_POLICY line:
   • Use Edit to append SCV_MODEL_POLICY=<POLICY> on a new line at the end (preserve existing content).

The value MUST be the literal policy identifier (e.g., recommended), never quoted, never with spaces.

Step 4 — User notice

Print a short final summary in the resolved language. Include:

• Which policy was applied.
• How many files changed (from Step 2 output).
• A reminder: "Restart Claude Code or run /reload-plugins so the new model frontmatter takes effect."
• A note: "Whenever /scv:sync runs and template frontmatter changes, this policy will be reapplied automatically from .env."

Example summary (English):

Applied model policy: all-haiku. 4 files updated. Run /reload-plugins (or restart Claude Code) for the change to take effect. Saved SCV_MODEL_POLICY=all-haiku in .env — /scv:sync will keep this in force on future template updates.

Non-negotiable rules

• Never use <chosen> / <POLICY> style placeholder strings inside a shell command that will actually execute. When generating the Bash call in Step 2, the resolved policy identifier must be inlined as a real literal (e.g., --policy all-haiku). The <POLICY> notation in this file is documentation only.
• Never run apply-model-policy.sh without first resolving POLICY from Step 1. If Step 1 cannot resolve a valid policy (e.g., AskUserQuestion was cancelled), stop and report — do not guess.
• Never modify command files directly. All frontmatter updates go through the helper script for idempotence and validation.
• Never write SCV_MODEL_POLICY= with an empty value to .env. If the user picks session-default, the value is the literal string session-default, not empty.