Skip to content

PHASE1-PICKS.md

Latest commit

 

History

History
34 lines (28 loc) · 5.93 KB

PHASE1-PICKS.md

File metadata and controls

34 lines (28 loc) · 5.93 KB
Raw

v1.9 Phase 1 — 12 Regression Test Picks

Branch: feat/v1.9-phase1-tests (off origin/main @ v1.8.3). Constraint: test-only. No impl changes. If a test reveals a bug, file an issue. Selection rubric: v1.8.x regressions (#864 #876 #856 #816 #881 #867) or P0 use-case scenarios from USE-CASES-AND-TESTS.md (S-LC-1, S-MS-3, S-FAIL-7, S-CLI-1, S-CLI-4) or parity gaps that #898 / #900 abstractions exposed but didn't cover.

# Case ID Layer Maps to What it asserts
1 prof-001 internal/session T2 / #881 / J3 GetEffectiveProfile falls back to config.json default_profile when no env vars are set (priority 4 of the resolution ladder). Today only priorities 1–3 have tests — config-default is the fallback users hit on a clean machine, and a regression here would silently re-route every read to "default".
2 prof-002 internal/session T2 / #881 / J3 Full table-driven precedence chain: explicit > AGENTDECK_PROFILE > CLAUDE_CONFIG_DIR > config default > literal "default". Locks the contract documented in config.go:301-336 so no re-ordering accident slips through.
3 prof-003 internal/session T2 / #881 / TUI K2 profileFromClaudeConfigDir directly: every documented mapping (~/.claude-work → "work", ~/.claude-personal → "personal", ~/.claude → "" no-inference, /opt/claude-prod → "prod") plus negative cases (empty, no dash, just trailing dash). Today this helper is only exercised through indirect paths in 2 tests.
4 web-001 internal/web #867 / J2 (joeblubaugh) GET /api/menu applies refreshSnapshotHookStatuses — fresh "waiting" hook lifts a stale "error" snapshot. Today only GET /api/sessions has this end-to-end assertion (TestParity_WaitingStatusFlowsThroughHandler); handleMenu calls the same helper at handlers_menu.go:41 but is untested at the handler boundary.
5 web-002 internal/web #867 / J2 GET /api/session/{id} applies refreshSnapshotHookStatuses — same divergence-lift assertion for the per-session endpoint at handlers_menu.go:72. Without this, a refactor that drops the call site from one handler but not the other would slip.
6 web-003 internal/web #867 defensive refreshSnapshotHookStatuses tolerates nil snapshot, items with nil Session, items that are groups, and an empty loader return. The current implementation has guards but no test pins them — a future refactor that removes them would crash on real production snapshots that always interleave groups and sessions.
7 send-001 cmd/agent-deck #876 / S-CLI-4 messageDeliveryToken (the helper that gates body-in-pane delivery evidence at session_cmd.go:2123): empty/short → "", >64 chars → truncated to 64, whitespace-trimmed. Currently has zero direct tests.
8 send-002 cmd/agent-deck #876 / S-CLI-4 (large-prompt receipt) sendWithRetryTarget(verifyDelivery=true) accepts a 100KB-class message body match in the pane as positive delivery evidence — exercises messageDeliveryToken's 64-char cap on the real call path. The existing TestSendWithRetryTarget_VerifyDelivery_AcceptsMessageInPane only tests a 19-char token; the silent-drop bug class manifests on multi-KB conductor prompts.
9 rebind-001 internal/session #856 boundary clearRebindMtimeGrace boundary: candidate.mtime - current.mtime ≥ 5s rebinds even if smaller; gap of 4s rejects. Existing TestSessionIDFromHook_ClearRebindWinsOnMtimeGap (instance_test.go:3576) covers the happy path with a generous gap; nobody pins the threshold itself, so a future "make it 30s" tweak would silently change behavior.
10 status-001 internal/web J1 / S-CLI-1-SCHEMA-STABILITY Every session.Status value serializes to its documented JSON literal through GET /api/sessions. The existing TestParity_AllStatusesPreservedThroughGetSessions only asserts round-trip; this asserts exact wire format ("status":"running", etc.) so any future Status-marshal refactor that changes casing or symbol breaks the CLI/JS shape contract loudly.
11 route-001 internal/web F8 / negative routing GET /api/session/{id} for an unknown id returns 404 with code:"NOT_FOUND", not 200/null or 500 — the per-session endpoint must be a hard negative under both no-overlay and overlay-present branches. Pins the contract that webMutations and overlay state never leak into the read-error path.
12 status-stop-001 internal/web J5 / B14 / #867 negative GET /api/menu: a "stopped" session is never flipped by a fresh "waiting" hook overlay. Asserts the user-intentional rule (Instance.UpdateStatus + applyHookStatusToMenuSession StatusStopped short-circuit) at the handler boundary so nobody wires hookStatusLoader past the stopped-stickiness guard.

Skipped on purpose

  • 🚫 Untestable in CI: WEB-K1/K2 (host-emulator paste/cmd-click), WEB-J4 multi-client tmux size negotiation, TUI G1 XTVERSION leak (host-PTY).
  • ❓ Product decision pending: WEB-A1–A10 topbar redesign, S-MS-4 sort-by-actionable, S-FAIL-13 multi-instance lock spec, mobile/tablet SLA.
  • Already covered: #864 (conductor/tests/test_python_compat.py), #856 happy path (TestSessionIDFromHook_ClearRebindWinsOnMtimeGap), #876 verifyDelivery error (TestSendWithRetryTarget_VerifyDelivery_ErrorsWhenNoEvidenceOfReceipt).
  • Needs new infrastructure not in scope: TUI lifecycle B1 (needs fakeInotify+teatest+fakeClock), hook event drop simulator (needs Linux real-fsnotify harness), CLI ↔ web cross-process fixture (needs separate test binary).

Execution discipline

  • One commit per case for clean review.
  • After each commit: go test -race -count=1 on the touched package.
  • If a test fails on main (i.e. the case turns up a real bug), file an issue and add an ACTION: PHASE1_FOUND_BUG line to RESULTS.md. Do NOT fix the impl in this PR.
  • LEFTHOOK=0 only for pre-existing lint failures, documented per commit.