Luci · Real code in `~/workspace/mc-orchestrator-build-mc/`.
Luci
Real code in `~/workspace/mc-orchestrator-build-mc/`. Let me invoke dev-loop and inspect the target functions.
Luci
Let me inspect the target code. Read the key files.
Luci
Two repos. mission-control is live (port 3001). Let me find FINAL-ASSESSMENT.md and read the actual function in the live repo.
Luci
Found it — `mission-control/models.py:4968`. Let me read the function and surrounding profile/registry infrastructure.
Luci
Now let me read the catalog, state, and the new runtime_registry module.
Luci
rtk is mangling grep output. Let me read the catalog region directly.
Luci
Now the catalog and `_runtime_profiles_for_purpose`. Let me read the region before 4660.
Luci
Now the catalog itself and runtime_registry. Let me read both.
Luci
Full picture now. Let me check existing tests in the live repo for these functions.
Luci
Let me enumerate every caller and its purpose to size regression risk.
Luci
Callers use only persistent/interactive/default — safe to enforce. Let me confirm the two ticket_runtime resolve sites won't crash on a raise.
Luci
Enough context. Let me check an existing scheduler task format and the profile-switch resolve site, then write the plan.
Luci
Plan clear. Tier 2-3 change, 3 features. Writing plan doc, then implementing.
Luci
Now implement. First the models.py changes — alias layer, purpose enforcement, probe.
Luci
models.py done. Now make preflight an inspection path (`enforce_purpose=False`).
Luci
Now graceful fallback in `_profile_for_ticket`:
Luci
Now the CLI probe script:
Luci
Now the tests:
Luci
Now the scheduler task, then run tests.
Luci
Now run tests + smoke.
Luci
Probe merges on-disk `api_keys.env`, so env={} still finds the key on this host. Make those probe tests isolate auth env.
Luci
All 17 pass. Now smoke the CLI script + regression-check existing runtime tests.
Luci
CLI good — gated = the 3 active state profiles, all PASS. Now regression-check existing runtime tests in the live repo.
Luci
Removing the speculative `explicit_profile_key` field (scope discipline — not needed for the 3 requirements).
Luci
All green. Let me review the full diff before committing.
Luci
Pre-existing lint (line 8273, not mine). My regions clean. Let me verify mc_interactive's resolve call won't raise uncaught.
Luci
Same stale-profile surface as `_profile_for_ticket`. Apply matching graceful fallback for resilience.
Luci
Let me run the broader runtime/dispatch/ticket test subset to confirm no regression.
Luci
Pre-existing collection error (unrelated `shared_console` packaging). Run root-level tests, ignoring the broken module.
Luci
One failure + one e2e timeout. Verify they're pre-existing (not from my edit) by running against stashed HEAD.
Luci
Confirmed pre-existing (fails on HEAD too). My change adds 30 passing tests, zero new failures. Committing.
Luci
Now commit the scheduler task file in the workspace repo.
Luci
Both pushed. Verify the new task passes the runtime-profile lint (it declares a direct_python override).
Luci
Lint clean. Work complete. Reporting via mc-coord.