[webandseo]

With no --model, muse exec runs Muse's catalog default, muse-spark-1.3-contributor, whose catalog entry says content may be used for product improvement. So every review, critique, delegate run and transfer made through the plugin lands on that model unless the user remembers --model spark each time, and the repository content in those prompts goes with it.

This changes the default, so it is your call as maintainer; I've kept it to one commit to make it easy to take or leave.

The bridge now always names a model: --model if given, else MUSE_CC_MODEL (an alias or a full catalog id), else the spark alias (muse-spark-1.3). --model contributor or MUSE_CC_MODEL=contributor still selects the contributor model explicitly.

• muse.mjs: resolveModelSelection(); runHeadlessAgent resolves the model for every launch, so review, critique, delegate runs (foreground and background workers), transfer, the stop-time review gate and check --probe are all covered.
• /muse:check gains a model: line with the model that will be used and why (--model, MUSE_CC_MODEL, or plugin default), plus the catalog note for that model; JSON gains models.selected. check accepts --model, and --probe uses the selected model. The catalog list labels Muse's own default "(Muse default)".
• README model sections and Environment table, commands/*.md, the delegate agent and runtime skill notes, CHANGELOG (Unreleased).
• Tests: every exec path defaults to muse-spark-1.3; MUSE_CC_MODEL takes aliases and full ids and --model beats it; check reports the selected model and its source.

Independent of #2 and #3. It shares the runEveryExecPath test helper with #3 (identical text, same place) and, like both, adds a line at the top of the CHANGELOG's Unreleased section; I'll rebase whichever lands later.

Tests: 113/113 locally on Windows.

🤖 Generated with Claude Code

[rtravellin]

This looks right for new runs: everything goes through runHeadlessAgent, background workers inherit MUSE_CC_MODEL, and check --model matches real runs. One question first.

Resume may switch models (scripts/lib/muse.mjs:707). --model is now always passed, so /muse:delegate --resume without --model gets spark even when the original run used another model. For example, a run started with --model contributor and continued with --resume changes model partway through. The resume path (executeTaskRun with resumeSessionId) doesn't look up the earlier run's model, and no test covers resume.

This only matters if Muse otherwise keeps a session's model when --model is omitted. If it does, could you skip the default on resume, or reuse the model recorded on the run being resumed?

This will also need a rebase: #2 and #3 are merged, and this one overlaps with both (muse-bridge.mjs, lib/muse.mjs, CHANGELOG, README).

[webandseo]

Muse does keep the session's model: on 1.3.0-R3401.1, a session started with --model muse-spark-1.3 and resumed with the same --session-id and no --model stayed on muse-spark-1.3, while a new session without --model got muse-spark-1.3-contributor. Fixed in 5ed83b3: runHeadlessAgent takes resume, and on resume only an explicit --model is forwarded, so neither the default nor MUSE_CC_MODEL switches a resumed session. Test covers resume with and without MUSE_CC_MODEL, and with an explicit --model. Rebased on main (merged with #3's foreign-context change).