Main-suite hygiene: the 10 amico-run + 16 extension pre-existing failures skip cleanly or get fixed — no baseline-diff tax per campaign

[aarontrowbridge]

Important

Problem — amicode main currently carries pre-existing suite failures that every campaign must baseline-diff around: the amico-run suite fails 10 tests (unreachable cloud endpoints, curlSotaFetch, the gh shim, a loadRegistry seed, parseDoctorArgs, a reviewSpec guard) and the extension suite fails 16 (service auth/bootstrap, contract, engine-proxy, fleet-data-plane, runner/wiring, staged-bins, one OPENCODE_DB-dependent open-threads case). They are environment-dependent — green on some hosts, red on others — which forces every campaign to prove "identical to baseline" by hand and makes CI verdicts ambiguous.
Approach — Make each failing test honest about its environment requirement: tests that need external endpoints or credentials skip cleanly (explicit skip reason, not a bare fail) when the requirement is absent; tests that are genuinely broken on main get fixed or explicitly marked with a tracked reason. No test is deleted; no assertion is weakened — a skipped test states what it needs and how to run it.
Scope — in: the 10 amico-run failures + the 16 extension failures on current main; out: any feature work, the scores/skill-drift real-checkout portability issue (#1531), the re-land PRs (#1526–#1530).

Acceptance Criteria

• From a fresh worktree off main, the amico-run suite reports 0 failures (skips allowed with explicit reasons)
• From a fresh worktree off main, the extension suite reports 0 failures (skips allowed with explicit reasons)
• Every skip carries a stated environment requirement (endpoint/credential/binary) — no silent skips
• Tests that were genuinely broken (not env-dependent) are fixed, with the fix described in the PR

Testing Decisions

The existing suites themselves are the surface; no new test files except where a skip-reason contract needs pinning.

Prior Art

• The baseline-diff records on PRs feat(amico-run): re-land the strategy-brief renderer (living-sota slice 6, #1310) #1526–docs(skills): re-land the spec-reviewer + analyze rewire off the dead P-list model (#1326) #1530 (the re-land campaign's verification logs enumerate the exact failing sets on main @ 244e6a9)
• The session-curation campaign's S3 baseline-diff proof (the 16-extension-fail set was first proven pre-existing there)

Notes

Found repeatedly during the 2026-09-24 re-land campaign; every implementer cast spent effort proving these failures pre-existing. Label afk: implement + merge unattended per house AFK semantics, merges still sequential-green only.