Summary

Slice 2 of #438, based on latest main and built on the slice-1 profile layer merged in #465. Adds the five remaining user-scope harness adapters behind the same contract and closes #438.

Design notes

• The shared profile engine is reused, not forked. It gains two additive surfaces on top of the schema-v2 state sections reserved in slice 1: whole-file generated artifacts (digest-owned, prune-on-uninstall) and a single managed entry inside a co-owned JSON hook config.
• Cursor instructions are not a plain marked block: the managed brigade-loop.mdc rule uses whole-file ownership so its alwaysApply frontmatter stays at the top of the file; plugin.json and the session-start hook script are generated artifacts; hooks.json keeps exactly one managed sessionStart entry and preserves foreign entries.
• Kimi paths are version-sensitive: a capability probe selects ~/.kimi when it exists and falls back to ~/.kimi-code; the profile carries the probed MCP path instead of blindly copying the static adapter path.
• _server_targets_harness aliases grok/kimi/opencode to their *-user adapters, matching the existing cursor/cursor-user alias. Slice-1 Claude/Codex behavior is unchanged (slice-1 suites pass unmodified except the two --target all assertions, which now expect seven profiles).
• Legacy cursor_user_cmd state (version 1) and profile state (schema_version 2) share ~/.cursor/brigade/install-state.json; each refuses the other's schema as a conflict, so they never clobber each other.

Contract honored per harness

• Dry-run default; --write required to touch disk.
• Only Brigade-owned files/marked blocks are written; hand-authored user instruction sections survive sync (tested for the four AGENTS.md harnesses; foreign Cursor hook entries and foreign MCP servers survive too).
• Idempotent: second sync writes nothing (no state/receipt churn).
• Uninstall removes exactly the Brigade-owned artifacts via the receipt model, including prune of created directories.
• doctor reports drift/conflicts read-only.
• --target all now covers all seven harnesses (Pi keeps its own bridge).

Tests

New tests/test_harness_user_scope_slice2.py (48 cases) drives everything through a temp HOME/tmp_path: dry-run vs --write, existing-config preservation, idempotent re-sync, uninstall-only-owned, doctor drift, Kimi probe selection, Kimi MCP foreign-key round-trip, Cursor managed-surface + conflict handling, stdio gate, and MCP uninstall per harness. No test touches the real ~/.openclaw, ~/.kimi-code, ~/.grok, ~/.cursor, or ~/.config/opencode.

Verification (brigade-wired)

brigade work verify run --target . --command "pytest -q tests/test_harness_user_scope.py tests/test_harness_user_scope_slice2.py tests/test_harness_profile_cmd.py" --capture brigade-work
# status: completed — exit=0 (90 passed)

brigade work verify run --target . --command "./scripts/verify" --capture brigade-work
# status: completed — exit=0
# ruff: All checks passed! / 532 files already formatted
# mypy: Success: no issues found in 318 source files
# version=0.25.1 checked=13 locations; managed snapshot: ok (6 manifests)
# 4182 passed, 3 skipped in 382.24s

Closes #438