Reviewed test inventory for ADR 0221
This is the ongoing review for VUH-1925,
under ADR 0221. The inventory
is not complete. Files without a recorded review remain unchanged. In particular,
plain .test.ts names do not establish that a test is a unit test: several exercise
real HTTP, filesystem, native executable and producer/consumer boundaries.
Keep trust boundaries, published contracts, regressions grounded in actual bugs, and all E2E, integration and golden tiers. Exact values are useful when they are the contract or retained real evidence; they are incidental when they merely repeat today's implementation or presentation choice.
First batch: Linear test data
The provider SDL shrank from 52,706 to 2,528 lines without deleting assertions. The schema review and measurements record the retained upstream types and signatures. Keep all five consumer files:
File under apps/clankie/test/ | Keep reason |
|---|---|
linear-api-tracker.test.ts | Provider documents, credential renewal/revocation, redaction and pagination contracts. |
linear-graphql.integration.test.ts | Lane authority, destructive confirmation and request admission across localhost HTTP. |
linear-request-budget.integration.test.ts | Quotas, actor isolation and reserved write/API/MCP accounting. |
linear-request-budget-autonomous.integration.test.ts | Integration tier; its existing skip remains, and is not an executed pass. |
worker-call-receipts.integration.test.ts | Durable uncertain receipts, replay refusal and authority fences. |
TUI presentation batch
Retain these five mixed files, removing only the reviewed incidental cases:
File under apps/tui/test/ | Cut | Keep reason |
|---|---|---|
banner.test.ts | Mascot glyph, palette RGB, row count, gutter, exact ASCII art, fixed condensed layout and padding snapshots. | Actual component width, hidden state, non-TTY/NO_COLOR/COLORTERM behavior, escape-free color opt-out and ASCII-only fallback. |
footer.test.ts | Token abbreviation strings, internal warning levels, exact context copy, extras separators and fixed row composition. | Actual terminal width across wrapped rows and very narrow output. |
connections-menu.test.ts | Relative-age/path/transcript formatting helper snapshots and exact ordered hub-option arrays. | Native resume dispatch, refused reads, durable errors and original-conversation reuse without a new process. |
herdr-menu.test.ts | Option order and exact labels. | Save before restart, cancellation, bundled/disabled configuration and retained session selection. |
tool-render.test.ts | Refusal pretty-print line count, fixed ten-line helper arrays and flattened argument copy. | Real captured Linear envelope and its formerly ineffective collapse regression; lossless non-envelope/text handling and Pi/MCP renderer delegation. |
Do not re-add these deleted kinds as unit snapshots. A mascot, palette, separator, menu order, padding or abbreviation can change without breaking a published contract. Prove intended presentation with inspected output or real evidence. Renderer delegation and safe session reuse remain contracts even when exercised with a controlled transport.
The first after-cut gate found six helper functions and one type exported solely for the removed snapshots. The lead approved making them private to their two implementation modules: export-only, no behaviour change. Their bodies and internal runtime callers are unchanged. Knip remains active; no replacement snapshots or dead-code exemptions are added. The same approval covers future helpers that become unexported or dead after their sole test consumer is removed; any behaviour change needs the lead's decision.
These source visibility changes broaden native Vitest dependency selection, so gate time must not be presented as a matched speedup against the five-file baseline.
Keep native-tool-render-integration.test.ts unchanged: native JSONL passes through
the actual parser, shell and Pi components, including retained VUH-1661 goldens,
malformed output, running sessions, empty results and collapse/expand behavior.
Keep fresh-page.test.ts unchanged: real shell frames exercise restored history
and startup ordering. Neither file is replaced by helper snapshots.
The batch evidence records line counts, gate selection and retained failed runs.
Command-picker presentation batch
This is the next clear landable cut among the reviewed candidates, not a claim that the remaining inventory has been exhausted. Four files lose 209 test lines.
File under apps/tui/test/ | Cut | Keep reason |
|---|---|---|
autocomplete.test.ts | Preview colors, spacer/description copy, outline glyph and full ranked-match arrays. | Actual canonical selection, alias collision and Kitty release regressions; argument/skill completion boundaries, width, row budgets, dismissal and keyboard submit/cancel. |
interactive-flow.test.ts | Outline and hint copy, status-row alignment/color and current-value labels. | Secret masking, multiline input, width, filtering, selected/current value submission, duplicate-title regression, right/left arrow and close behavior. |
provider-commands.test.ts | Fixed auth-status table, default slot labels, model-order helper snapshot, setup labels/order and readiness-footer copy. | Credential redaction/broker storage, hosted approval refusal, real provider/model/effort persistence, endpoint fallback, restart intent and authoritative cross-face config changes. |
shell-assembly.test.ts | Default-state constructor snapshot, pending/picker/result copy and duplicated ten-line tool collapse assertions. | Prompt preservation and uncertain delivery, steer/queue and pasted input, streamed words, scoped conversation/fresh-context/close handling, side transcript restoration and server interrupts. |
The one source edit is export-only, no behaviour change: newestFirst loses
an export used only by the deleted ordering snapshot. The implementation body
and its internal model-picker caller stay identical. readinessFooter remains
exported for its real entrypoint consumer. No exemptions or replacement
implementation snapshots are added. Native render goldens, actual startup frames
and guided-setup integration tests stay unchanged.
Batch evidence records LOC, before/after gate wall time, selection and their comparison limits.
Menu presentation and duplicate-helper batch
Eight mixed TUI files keep their behavior and boundary coverage while shedding menu titles, status/hint copy, presentation order and duplicate helper assertions. The batch review records each cut and keep reason. No integration, golden or contract case is removed.
external-activity.test.ts retains expanded payload content;
machines-menu.test.ts retains discovered transport and command dispatch,
default-session protection, grants, reconnect and refusal;
product-navigation.test.ts retains exact membership and selected identity,
without asserting picker order; settings-menus.test.ts retains real persistence;
setup-commands.test.ts retains readiness, cancellation, draft and command routing;
voice-commands.test.ts retains actual wizard/API/broker integration and marker
redaction; persona-commands.test.ts retains defaults, alias recovery and stale
write refusal; project-menu.test.ts retains its revision-bearing API integration.
The deleted describeVoice helper case asserted labels without supplying a secret
marker; actual wizard redaction remains covered at the API/settings boundary.
Its sole test-consumer export becomes private with its body and runtime caller
unchanged: export-only, no behaviour change, under the existing lead ruling.
Other reviewed files retained
These reviews authorize no deletion in the listed files. Mixed files can receive a later case-level review; their boundary coverage must survive it.
Package and files under its test/ directory | Keep reason |
|---|---|
agent-hosts: hosts.test.ts, pi-roots.test.ts | UTF-8 byte ranges, native profile selection, SSH injection refusal, shell quoting, symlink/path containment and malformed remote responses. |
vec3: vec3.test.ts | Existing Minecraft CommonJS/pathfinder consumer contract, mutation versus copying and zero-vector normalization. |
work-items: backends.test.ts, convention.test.ts, format.test.ts, linear-pagination.test.ts | Tracker/CLI/provider contracts, owner-text preservation, fenced Markdown, token redaction, origin-only pagination, bounded continuation and refusal to guess competing trackers. |
model-registry: model-registry.test.ts, hire-model.test.ts | Published price/catalog inputs, disk cache and environment overrides, malformed/forward-compatible catalog data and refusal of retired/wrong-harness hires. |
observability: body-telemetry.test.ts, hosted-body-telemetry.test.ts, body-telemetry-shipper.test.ts, support-audit-http.test.ts, observability.test.ts | Secret/prompt redaction, tenant stamping, signed requests, durable retry cursors, mandatory support audit and real shell/HTTP producers. |
persona-images: persona-images.test.ts, video.test.ts | Real art/FFmpeg producers, bounded decoded content, symlink refusal, cache recovery and explicit image-data versus instruction priority. |
interactive-environment: activity-observation.test.ts, discord-transport.test.ts | Published observation/presence schemas, bounded data, room membership and body-swap address compatibility. |
rendered-surface-client: activity-frame-sink.test.ts | Producer bearer/wire contract, stale frame/audio refusal, latest-status replay and explicit-close reconnect behavior. |
play-voice: client.test.ts | Bearer and bounded narration/utterance contracts, unreachable refusal, room-state reset and malformed wire rejection. |
apps/clankie: captain-lane-prompt.test.ts | Authenticated HTTP prompt/card reads stay inside the bearer's lane; unauthenticated or cross-lane access never assembles private notes. |
Further reviewed files retained:
Package and files under its test/ directory | Keep reason |
|---|---|
api-client: api-client.test.ts, gateway-encryption.test.ts | Bearer/lane/live-session API contracts, delivery-ID polling and encrypted refusal without replaying uncertain sends. |
settings: attachments.test.ts | Actual attachment-path producer/reader regression, including blank overrides. |
settings: codex-rate-limits.test.ts | Read-only RPC/profile isolation, bounded owned-child cleanup, and no automatic login or hook trust acceptance. |
settings: project-worktrees.test.ts, project-enrollment.test.ts, settings-fence.test.ts | Approved workspace namespace, actual Git identity, path containment and settings-generation authority races. |
settings: minecraft-settings.test.ts, linear-wake-migration.test.ts | Approved destinations and play budgets; owner identity, edited wake rules and malformed persisted data preservation. |
settings: skill-roots.test.ts, bundled-skills.test.ts | Shipped/workspace/personal precedence and plugin projection contracts; retired names do not reappear. |
Package and files under its test/ directory | Keep reason |
|---|---|
credential-broker: account-credential.test.ts, credential-broker.test.ts, credential-store.test.ts, linear-app.test.ts, linear-oauth.test.ts, activity-producer-credential.test.ts, captain-credential.test.ts, discord-user-session-provider.test.ts, operator-credential.test.ts, play-voice-credential.test.ts, provider-bearer.test.ts, public-gateway-credential.test.ts, discord-bot-provider.test.ts, discord-bridge-credential.test.ts | Credential and lane boundaries, secret redaction, token rotation and lost-reply recovery, sign-in versus retryable refusal, corrupt-data preservation, private file modes and real independent-process refresh races. |
settings: agent-hosts.test.ts, codex-accounts.test.ts, persona.test.ts, store-isolation.test.ts, linear-follow.test.ts, linear-wake.test.ts, machines.test.ts, projects.test.ts | Host injection refusal, credential-profile isolation, owner-authored persona preservation and hallucination regression, real process isolation, wake author identities, approved machine/project grants and revocation without migration resurrection. |
Package and files under its test/ directory | Keep reason |
|---|---|
interactive-environment: rendered-surface.test.ts | Published overlay v1/v2 wire compatibility, honest stale-producer normalization and incompatible version refusal. |
media-connector: media-connector.test.ts | Provider schemas and credential transport, exact persisted/hash artifacts, edit/generation routing, video-origin/status refusal and reference preservation. |
model-provider: subscription-policy.test.ts, readiness.test.ts, routing.test.ts, local-endpoint.test.ts, anthropic-configured-model.test.ts, xai-configured-model.test.ts, configured-model.test.ts | Hosted subscription approval and credential boundaries, routing money/escalation policy and VUH-1391 regression; persisted endpoint configuration, SDK credential separation and actual Astra reasoning-effort transport regression. |
All 28 files in packages/protocol/test/*.test.ts were reviewed and retained.
They prove published producer/consumer schemas, old-client response compatibility,
immutable owner/occupant/incarnation correlation, delivery uncertainty, private
push and voice data boundaries, gateway route/header allowlists, real ingress/wake
cryptographic binding, bounded terminal bytes and lossless replay/recovery. Actual
pmset input goldens and the node-free React Native import contract stay.
Three mixed protocol files have later incidental-assertion candidates, without
any changes in this batch: app-telemetry.test.ts pins the exact ["zod"] import
array; host-power.test.ts pins advice-copy fragments; protocol.test.ts pins role
array order and helper-generated appearance diversity. Their privacy, wire,
cryptographic, compatibility and data-preservation coverage stays.
The lifecycle tab-name and SSH settings-race regressions named by the assignment remain protected. No evaluator, tldraw, skin-manifest, Swift-deduplication or ops mockup cleanup is included here; those VUH-1897 items remain unassigned.
Further full-file reviews in this batch retain these files unchanged:
packages/settings/test/settings.test.ts: secret/input/HTTPS/consent boundaries, execution-grant migration, private file mode, concurrent-write preservation, explicit environment precedence and MCP deny-by-default.- Under
apps/tui/test/,voice-commands.test.ts,desktop.test.ts,settings-menus.test.ts,project-menu.test.ts,owner-command-layer.test.tsandhost-power-surfaces.test.ts: real settings/API writes, preserved owner state, credential separation, stale-draft refusal and honest native failures. - Under
apps/tui/test/,seat-context-selection.test.ts,linear-follow-status.test.ts,stance-command.test.ts,file-command.test.tsandagents-command.test.ts: exact conversation/native identity, scoped bearers, read-only/redacted status, bounded claims, named artifact publication and request refusal before external calls. Their request arrays are contracts. - Under
apps/tui/test/,catalog-watch.test.ts,skill-catalog.test.ts,skills-command.test.ts,claude-tool-catalog.test.tsandsetup-flow-integration.test.ts: native discovery, consumer agreement, preserved failed-read state, real subprocess/HTTP catalog evidence and guided setup integration.face-bash.test.tskeeps real subprocess output, bounds, cancellation and failure propagation. - Under
apps/clankie/test/,captain-render-notice.test.tsandoperator-tool-detail.test.ts: room/private-memory scope, honest render failure, secret redaction, bounded native detail and exact-only invocable skills.packages/discord-presence-core/test/voice-tone-text.test.tskeeps arbitrary stream-split behavior and withholding of malformed or partial voice directions.
Further read-only reviews retain these files unchanged:
interactive-environment:interactive-environment.test.ts,emulator.test.ts: published Discord/GBA/environment schemas, frozen fixtures, credential/telemetry refusal, transport authority and bounded/versioned leases.play:free-play-mind-prompt.test.ts,free-play-character.test.ts,play-story.test.ts,free-play-mind-timeout.test.ts,free-play-voice.test.ts: real AI SDK prompt validation, live invalid-prompt and drained-stream wedge regressions, cold-start/decoder evidence, bounded structured output and journal/story identity without inner-monologue disclosure.apps/clankie:captain-room-guidance.test.ts,persona-caption.test.ts,captain-loadout-tools.test.ts: async one-use private guidance, separate room captures, real art-file captions, role separation and body-specific tool exposure.apps/tui:services.test.ts,claude-plugin.test.ts,codex-plugin.test.ts,discord-room-view.test.ts,status-connection.test.ts,linear-follow-status.test.ts,herdr-roster.test.ts,next-step.test.ts: real owned-process lifecycle, native hook trust and CLI consumer wiring, honest pending deliveries, redacted read-only wake state, native fleet cursors/identity and canonical recovery commands. These arrays describe consumer wiring.discord-presence-core:addresses-character.test.ts,body-voice-lease.test.ts,body-voice-reconcile.test.ts,presence-grant.test.ts,voice-consent.test.ts,voice-control.test.ts,voice-address.test.ts,voice-audio.test.ts,voice-room-evidence.test.ts,shutdown.test.ts,discord-rest.test.ts,minecraft-login-code.test.ts,transcript-store.test.ts,room-text.test.ts,receipt-store.test.ts,presence-session.test.ts: consent/account/guild and lease fences, exact native termination, HTTP/URL/mention contracts, guarded secret-bearing DM delivery, PCM/WAV boundaries and real room-tone regression, private append/tail/symlink modes, content-free receipts and synchronous publication-loss fencing.- Mixed later candidates stay unchanged:
check-report.test.tsunderdiscord-presence-corekeeps the JSON CLI output contract; its human table alignment is a future presentation cut.captain-model-card.test.tsunderapps/clankiekeeps per-run refresh and unresolved-model behavior; its exact sentence-formatting case is a future incidental cut.
Further retained consumer contracts: elevenlabs-tts.test.ts,
voice-ingress.test.ts and voice-floor.test.ts under discord-presence-core
keep provider WebSocket/PCM transport, bounded secret-safe failures, public
speech-rate caps, speaker attribution and withholding ambient approval payloads.
play-voice/test/listener.test.ts keeps real WebSocket bearer/route/size behavior,
Unicode-safe attributed words, no absent-client replay and EADDRINUSE recovery.
model-provider/test/openai-codex.test.ts keeps the real local OAuth callback,
state/PKCE binding, device polling, broker refresh races and revoked credentials.
Remaining inventory
The read-only candidate scan also includes native test filenames and Rust files
with inline test attributes. Whole Rust source-file lines are not test-only LOC.
It records file hashes and repository revisions in the worker's ignored
.local/vuh-1925/inventory-candidates.json; unreviewed status is explicit. This
scan is a discovery aid, not a finished classification or a deletion rule.
Private app and hosted-service reviews stay in their own repositories or private issue evidence. They do not publish private implementation details here. No private tests have been deleted by this batch. Evaluation fixtures are outside this pruning and remain manual-only.
Service presentation helpers
Two mixed files lose 24 test lines and one case. Product code, exports and fixtures are unchanged.
File under apps/clankie/test/ | Cut | Keep reason |
|---|---|---|
captain-model-card.test.ts | Exact model-card sentences, token abbreviations and input-list copy. | Per-run resolution refresh, original system prompt and honest unavailable-model behavior. |
play-execution-shared.test.ts | Exact ordered room-event text snapshot. | Overlay length cap and blank-event omission. The retained play-voice.test.ts runtime cases exercise the authored thought, observed outcome, goal and next intent across the actual play/voice/journal seam. |
The room-event fields remain behavior covered by that runtime file; only the helper's incidental field order and line layout disappear. The model-card text can change without invalidating model selection or prompt refresh.
Additional full-file reviews retain captain-memory.test.ts,
captain-model.test.ts, personas.test.ts, runtime-health.test.ts,
captain-native-subagents.test.ts, codex-accounts.test.ts,
captain-body-identity.test.ts, captain-browser-tools.test.ts,
browser-authority.test.ts, system-authority.test.ts, desktop.test.ts,
operator-auth.test.ts, runtime-terminals.test.ts and
runtime-connections.test.ts. They guard lane-scoped private memory and file
publication, broker preservation and serialized credential rotation, durable
owner identities and exact account homes, bounded public boot identity,
transcript/presence consumer agreement, original turn authority and revocation,
browser shell grants, room audience grants, owner quiet hours, HTTP credential
revocation, pinned terminal routing and runtime/workspace approval. Exact
request arrays and native identities in these files enforce contracts, rather
than presentation order. The empty-memory label is explicitly documented in
docs/cli.md, so its case remains. No E2E, integration or golden case is removed.
The batch measurements record counts and the before gate. Final gate output and timings are retained with the batch's issue evidence; the all-repository inventory remains incomplete.
On resumption, eleven additional full-file reviews keep presence-service,
memory-capacity-api, lead-coding-helper, voice-transcripts,
discord-captain-actions, discord-tool-progress, discord-turn-trail,
discord-sender-standing, herdr-head-seat, herdr-parent-edges and
pi-native-capability under apps/clankie/test/. They protect real captain/API
projection, the historical memory quota bug, bounded no-follow filesystem and
clean-shell operations, private retained transcript access, persisted dispatch
guards after restart, content-free progress, one-shot native turn provenance,
verified owner/grant separation, native head identity, corrupt-cache recovery
and inode/hash-bound native capability lifetime. Ordinary filenames do not
make these producer/consumer integrations removable unit coverage. No test is
removed from these files and the helper fixtures do not run evaluations.
Further boundary review after VUH-1948
These 50 additional files are retained. Reviews use the file bytes on
3ef50411d, based on 6f7e23aa; no deletion is authorized by this list.
Rust whole-file line counts include production code, not only tests.
apps/tui/test/linear-publishing-command.test.ts: Keep published CLI/provider request decoding, fail-closed refused/uncertain/malformed envelopes and malformed-input no-call boundary.apps/tui/test/workspace.test.ts: Keep real filesystem checkout/root/path resolution, non-directory refusal and conversation workspace scope contract.apps/discord-bridge/test/subcommand-authority.test.ts: Keep Discord command authority, owner-only settings and revocable voice-consent fences. Source-wiring assertions remain because full dispatch duplication is not established; no speculative cuts.apps/tui/test/work-command.test.ts: Keep published work CLI request mapping, criteria/status/owner/evidence preservation and malformed argument refusal.apps/discord-user-session/test/stream-discovery.test.ts: Keep Discord stream producer/consumer opcodes, stream keys, credential events, DAVE channel identity and ended-stream cleanup.apps/tui/test/pair-routes.test.ts: Keep pairing CLI reachability disclosure and JSON route contract across gateway/direct/no-route offers, including App Store plain-HTTP refusal.apps/clankie/test/project-worktree-membership.test.ts: Keep native linked-worktree admission and cwd/process/authority/settings race refusal across Git observation.apps/tui/test/support-http.test.ts: Keep real CLI/console/HTTP support-grant create/list/pair/revoke integration, expiration bound and invalid scope refusal.apps/clankie/test/gmail-mcp-canary.test.ts: Keep credentialless no-network refusal and explicitly opt-in real broker/Google MCP canary. No live canary run authorized or performed.apps/clankie/test/codex-seat.test.ts: Keep native foreground/open-file session resolution, exact parent preference versus child rollout and absent/replaced native session behavior.apps/clankie/test/hosted-wake-key.test.ts: Keep real device-pairing HTTP authentication/schema/revocation boundary, local revocation through hosted cleanup failure and self-hosted route absence.apps/clankie/test/operator-conversation-workspace.test.ts: Keep actual conversation-store turn workspace/native seat/run correlation and absolute-existing-directory refusal.apps/clankie/test/http-journey.test.ts: Keep real HTTP/restart integration, pairing-code secrecy and operator-only offer refusal without event-log mutation.apps/clankie/test/local-fleet-discovery.test.ts: Keep private/production filesystem isolation across publication/cleanup failures, actual native discovery reader and authenticated link boundary.apps/tui/test/harness-enable-idempotency.test.ts: Keep captured native enable result regression, exact profile/source and native error proof, retarget/race/wrong-profile refusal.apps/clankie/test/inbound-seat-binding.test.ts: Keep fresh native session identity before receipt/POST acceptance, alias/replacement refusal and no legacy unbound delivery.apps/clankie/test/work-items-linear-results.test.ts: Keep actual MCP host/work service integration, lossless large JSON and explicit oversized UTF-8 byte refusal without truncation.apps/clankie/test/voice-receipt-activity.test.ts: Keep real JSONL receipt consumer integration, open/closed room stay correlation and spoken/suppressed/token accounting.apps/clankie/test/codex-catalog-refresh.test.ts: Keep real filesystem notification coordinator integration, bounded failures/no duplicate native mutation and private config symlink/home refusal.apps/clankie/test/project-routes.test.ts: Keep HTTP settings mutation trust boundary, explicit workspace/revision and authority races, tracker binding protection and unrelated data preservation.apps/tui/test/project-settings-cli.test.ts: Keep CLI operator-bearer/revision request contracts, conflict no-retry and exact seat/occupant membership without injected project.apps/clankie/test/discord-durable-room.test.ts: Keep untrusted backlog isolation versus carried history, durable room/native voice session compatibility and silence sentinel streaming contract.apps/clankie/test/captain-turn-metrics-api.test.ts: Keep operator-only bounded metrics API, no transcript disclosure and unknown legacy execution/usage without invented token accounting.apps/clankie/test/seat-ledger.test.ts: Keep real durable file/restart accounting, seat/day isolation and honest observed run outcomes; disappeared panes never become success.apps/tui/test/seat-hook.test.ts: Keep native hook CLI wire/auth/session correlation, actual transcript fallback and absent/unowned pane no-report boundary.apps/clankie/test/hosted-device-security.test.ts: Keep real filesystem key identity/private modes, rollback/session revocation and lost-response recovery, unreachable authority and symlink refusal.apps/tui/test/metrics-cli.test.ts: Keep documented CLI bounds/query/auth contracts and honest legacy unknowns; credentialless read never becomes anonymous network access.apps/tui/test/project-onboarding-cli.test.ts: Keep reviewed proposal/hash/incarnation/revision contract, uncertain no-retry and malformed no-dispatch; review never implicitly confirms.apps/vox/src/media_sink_wants.rs: Keep Discord opcode 15 producer/consumer payload stream/quality/pixelCounts contract. Whole-file LOC includes production code, not test-only LOC.apps/clankie/test/agent-sessions.test.ts: Keep native transcript producer/consumer formats, bounded UTF-8 tails/cursors, replacement/partial-record lossless recovery and retargeted host cache isolation. No PC work executed.apps/clankie/test/agent-work.test.ts: Keep real SQLite/file/restart goal and assignment contracts, exact occupant persistence, legacy-client compatibility and read-only unsupported/remote store isolation.apps/clankie/test/accounts.test.ts: Keep real loopback OAuth/broker/HTTP integration, provider/account/consent/revocation/PKCE isolation and replay refusal, real subprocess stdout/events/telemetry credential redaction, hosted token separation.apps/clankie/test/body-lease-router.test.ts: Keep real durable lease mutual exclusion/recovery, grant/identity rechecks before effects, uncertain no-replay, explicit scoped wake/head routing and overlapping receipts.apps/clankie/test/body-lease-routes.test.ts: Keep authenticated HTTP writable conversation authority, scoped browser leases/refused close, actual-stop-before-release, stale/revoked token fences and explicit head designation.apps/clankie/test/body-leases.test.ts: Keep real durable exclusive lease/restart and abrupt-death lock recovery, stale/expired token fencing, uncertain effects and corrupt/unwritable-state fail-closed.apps/clankie/test/herdr-runtime.test.ts: Keep real socket ownership refusal/closure and real pane shell environment isolation, selected installed/bundled binary and harness-marker boundary.apps/clankie/test/operator-conversation-input.test.ts: Keep actual durable conversation queue/steer producer-consumer ordering and replay behavior for both owner and internal turns.apps/discord-bridge/test/voice-presence.test.ts: Keep Discord live-body authority/target/consent boundary, idempotent join and cross-guild leave refusal.apps/clankie/test/channel-projection.test.ts: Keep Discord webhook wire/room/thread/credential transport contract, mentions disabled, content bounds and honest provider refusals.apps/clankie/test/fleet-membership-route.test.ts: Keep authenticated HTTP/CLI membership integration, revoked-authority post-observation refusal and honest unavailable native card.apps/clankie/test/herdr-session.test.ts: Keep exact owner-selected external runtime versus service-owned runtime, launch environment isolation and absent/native observation fallback.apps/clankie/test/linear-fleet-admission.test.ts: Keep actual MCP/provider/HTTP write fences after asynchronous author attribution and credential/header selection; revoked fleet cannot dispatch.apps/tui/test/telemetry-support-http.test.ts: Keep real CLI/loopback metadata/signed transport integration, mandatory support audit independent of diagnostic consent.apps/clankie/test/project-membership-latency.test.ts: Keep native authority proof bounded-scan regression, two checkpoints and no cross-request stale process/cwd/failed observation reuse.apps/tui/test/restart.test.ts: Keep delayed restart CLI/runtime contract: final complete answer versus partial/tool-use transcript, preserved service/Discord launch and harness environment isolation. No service restarted.apps/tui/test/send-stdin.test.ts: Keep CLI UTF-8 stream/newline preservation and duplicate/empty input no-dispatch boundary.apps/clankie/test/lead-census.test.ts: Keep fixture isolation from actual owner transcripts, native goals/work/subagents and ambient process access.apps/tui/test/harness-profile-alias.test.ts: Keep canonical profile/symlink/source-manager boundaries and post-consent/native/skill-build races; no unapproved alias retarget/update.apps/clankie/test/connect-tools.test.ts: Keep operator-only inbox trust boundary and sender-authored content untrusted marking without treating refusal as sender text.apps/clankie/test/pokeagent-mmo-boundary.test.ts: Keep published player versus world-host dependency/transport boundary; no duplicate socket host implementation.
Discord helper presentation batch
Reviewed both complete test files, their product functions, real readiness/live-proof
CLI consumers and original changes (71066d70c, f60806d86) at main 3ef50411d.
Neither deleted detail is a documented actual bug regression or machine contract.
packages/discord-presence-core/test/check-report.test.ts: remove the human PASS/FAIL table snapshot of exact padding, title and remediation placement. Keep the JSON CLI payload case byte-identical: machine-readable output is a published consumer contract. The writer and all real consumers remain.apps/clankie/test/discord-music.test.ts: remove the exact authored sentence for an unreachable body. Keepok: falserefusal, selected live-body routing, author attribution and lossless provider replies; all three cases remain.
These two files shrink from 103 to 74 lines, five to four cases: 29 lines and one case removed. No product code, exports, fixture or E2E/integration/golden tier changed. The actual text writer is run and its output inspected alongside the retained tests. Measurements and evidence record the baseline and landing gates; temporary baseline selectors are absent from the committed files. Running subtotal: six unit-pruning batches, 688 lines and 39 cases removed; fixture bulk remains separate. The inventory is incomplete.
Further full-file boundary reviews on the same main retain:
apps/clankie/test/minecraft-host-authority.test.ts: Keep final conversation/individual grant versus guild-only authority, post-await revoke and synchronous stale identity dispatch fences.apps/clankie/test/tracker-isolation.test.ts: Keep real config/native subprocess boundary disabling inherited tracker connectors and listing failure fail-closed.apps/clankie/test/remote-lead-delegations.test.ts: Keep launch-secret exact host/pane/occupant/process lifetime proof, retained grant revoke/downgrade/unavailable/post-await refusal. Reviewed only; no remote machine operation.apps/clankie/test/host-power.test.ts: Keep real pmset-format producer/consumer schema and health API/CLI fresh opt-in, unknown/sleep-gap versus ordinary stall observation contracts.apps/clankie/test/minecraft-destination.test.ts: Keep DNS/IP/SRV pinning and exact approved public host/port trust boundary, mixed/nonunicast/failure refusal and endpoint redaction.apps/clankie/test/minecraft-host-invite.test.ts: Keep recipient/enrollment-bound ephemeral capability, revoke-before-dispatch, immutable origin/public credential redaction and private/invalid endpoint refusal. Mixed line-count/copy remains later incidental candidate.apps/clankie/test/composer-catalog.test.ts: Keep real skill roots/native harness catalog syntax and enabled-only plugin/schema quick-action validation shared across API/seat producers.apps/clankie/test/email.test.ts: Keep real credential/settings boundary and IMAP/SMTP/MIME consumer contract, absent credentials refuse. No live mail read/send.apps/clankie/test/opencode-profiles.test.ts: Keep real native SQLite/file/API source lifetime, original prepared identity/fresh proof and retarget/session/foreign machine refusal; metadata cannot launch.apps/clankie/test/remote-herdr-transcript.test.ts: Keep exact listed host/native session versus prefix/ambiguous/unlisted path boundary and lossless append/cache/concurrent producer reads. Reviewed only; no remote operation.apps/clankie/test/fleet-host-fence.test.ts: Keep actual HTTP MCP/WorkerMcp/host credential and final-admission awaits revocation fence; changed/disabled authority cannot reach provider.apps/clankie/test/fleet-project-membership-routes.test.ts: Keep independently authorized bounded no-store API, injected native proof/project refusal and post-await reauthorization, honest schema errors/private detail redaction.apps/tui/test/accounts-command.test.ts: Keep secret-stdin authenticated account route/redaction, persisted OAuth settings and provider device-code poll interval/request contract. Mixed prompt copy is a later candidate.apps/tui/test/installer.test.ts: Keep real installer/tar/checksum immutable version rollback with preserved owner state, one login PATH entry and linked harness refresh contract.apps/tui/test/project-role-cli.test.ts: Keep operator revision API whole-role/null/default preservation and model validation no-write; alternate native profile registration does not modify login.apps/clankie/test/codex-hook-trust.test.ts: Keep native discovered-hash trust limited to own installed worker hooks, owner-home and missing/guessed hash fail-closed, unrelated hooks/config untouched.apps/clankie/test/minecraft-capture.test.ts: Keep renderer loopback/session/generation/stale/dimension/body bounds and post-await ended-session fence, broker reconnect sink recovery.apps/clankie/test/minecraft-host-tools.test.ts: Keep host-proven identity versus forged tool args, typed admin injection/op refusal, diagnostics secret scrubbing and raw/delegated MCP hosting access denied.apps/clankie/test/channel-turns.test.ts: Keep actual observed Discord PASS leak regression, silent versus spoken sentinel semantics, lossless bounded native transcript and truthful unavailable versus quiet outcomes. Mixed exact prose/tie ordering remains later incidental candidate; stale PTY explanation warrants provenance review.apps/clankie/test/hosted-security-wire.test.ts: Keep real signed nonce/body digest/tenant/installation/lifetime schema state, persisted-before-register key rotation and device revoke versus account identity.apps/clankie/test/lead-native-claude-launch.test.ts: Keep evaluation harness native containment/lifetime safety fixture cases. Evaluation-related inputs are outside pruning and were not executed.apps/clankie/test/captain-lane-prompt-assembly.test.ts: Keep lane authority/private address/fleet preference exposure and VUH-1391/VUH-1456 documented instruction contracts, native harness project file compatibility. Mixed exact whitespace/section order remains an incidental case-level candidate.
VUH-2149 packages and Discord bridge, slice 1
Reviewed at fetched origin/main 990a774a4cceaba7a5b3f6d2939d5529a70b8041.
Only tests change. No assertion or timeout is weakened. The table records every
removed case and both merged cases. Retained test names below are exact selectors.
| File | Removed or merged case | Reason and retained coverage |
|---|---|---|
packages/discord-presence-core/test/voice-floor.test.ts | wakes on being addressed, for free, and the speaker takes the floor | Duplicate: voice-session.test.ts / an addressed wake briefs, opens, seeds an attributed ring, and responds exercises the real floor through session transcription and checks attributed wake/engagement. Remaining floor re-address tests check holder identity. |
| Same | under the all policy, any speech is offered without fabricating engagement | Duplicate: session / keeps an all-policy silent turn dormant so the next line gets a fresh offer checks the same offer/dormancy and subsequent audible engagement through the actual consumer. |
| Same | the floor holder continuing without naming him is an offer, and does not refresh decay | Duplicate: session / the floor holder continuing gets another response without reopening or re-briefing and a same-breath pivot is an offer the model may refuse, and silence does not refresh decay. Exact decay threshold coverage remains in floor / uses the documented default window when none is configured. |
| Same | a goodbye with his name in it is an address he gets to answer, not a cut-off | Duplicate: session / no phrase releases the floor: a goodbye is answered and decay ends the exchange. Holder transfer remains in floor / a re-address from anyone holds and moves the floor to that speaker. |
| Same | releases on silence alone, with no phrase at all, under an explicit clock | Duplicate: session / decay fires from the timer tick with no phrase at all exercises the consumer's actual timer. Floor's documented-default test retains the exact inclusive/exclusive threshold. |
| Same | a suppressed offer stays dormant and is counted | Duplicate: session / an empty response is him passing: suppressed, still dormant, session parked on the hold checks the actual response-to-floor accounting and dormant state. Floor's monotonic counters and duplicate-outcome checks remain. |
packages/discord-presence-core/test/shutdown.test.ts (deleted file) | runs every step and rethrows the first failure | Duplicate: bridge vox-process.test.ts / publishes presence-off and records stopped after leave failure, then reports the first error calls the real shared helper through bridge shutdown, injects two failures, checks every cleanup step and the first error. |
| Same | collapses concurrent signals onto one run | Duplicate: apps/discord-user-session/test/shutdown.test.ts / disconnects roles and listeners before closing the shared Vox child once calls the real shared coalescer through its consumer with concurrent SIGTERM/SIGINT. This consumer file is unchanged. |
apps/discord-bridge/test/commands.test.ts | exposes person-memory controls | Duplicate: keeps person-memory mutation proposal-only and excludes ambient export/delete looks up the same registered command and asserts its actual choice values and forbidden operations. Namespace, consent, tools and Discord's cap remain covered. |
packages/discord-presence-core/test/voice-session.test.ts | Merge clears the current session when the voice transport disconnects and clears the current session when matching DAVE state is cleared | Duplicate setup/assertions: table drives both original matching native events, keeping active=false and exactly one left receipt for each, plus the original shared-Vox-not-closed assertion. Two runtime cases remain; there is no removed integration case. |
Measurements and checks
| Test file | Before lines | After lines | Before runtime cases | After runtime cases |
|---|---|---|---|---|
voice-floor.test.ts | 468 | 398 | 37 | 31 |
shutdown.test.ts | 33 | 0 | 2 | 0 |
voice-session.test.ts | 5,750 | 5,739 | 215 | 215 |
Bridge commands.test.ts | 62 | 58 | 7 | 6 |
| Changed files | 6,313 | 6,195 | 261 | 252 |
Changed test files: 4 → 3; 118 lines and 9 runtime cases removed. The focused before/after run also includes unchanged bridge Vox ownership and user-session shutdown consumers: 6 → 5 files, 271 → 262 passing runtime cases. The merged voice-session table preserves both cases; static declaration count drops by one.
Complete assigned-path inventory (packages/*/test/**/*.test.ts plus
apps/discord-bridge/test/**/*.test.ts): 187 → 186 files, 46,862 → 46,744 lines,
1,610 → 1,600 static test declarations. Declaration counts count each it/test
registration once, including each table registration, rather than expanding rows;
runtime counts above come from Vitest JSON. The inventory includes the unchanged
manual-eval corpus file; it was neither collected nor executed. No area-wide
runtime count or speedup is claimed.
Protected voice-session coverage remains: speaker consent and PCM disposal, post-await body/grant revocation, immutable gateway attribution, provider and playback correlation, stale generations, failure/uncertainty receipts, mute/stop confirmation, and bounded graceful ending. Existing exact disclosure wording, provider configuration, protocol schemas and recorded regressions remain.
Focused check evidence is recorded in the slice evidence.
VUH-2149: TUI slice 1 — incidental display cases
Reviewed at 990a774a4, applied to current main 586ccd31b, against the existing TUI pruning above. This is a
case-level cut, not permission to delete the remaining files. All four mixed
files retain their behavioral cases. No integration, golden, credential,
authority, lossless-output or documented bug regression is deleted.
File under apps/tui/test/ | Removed or merged case | Reason and retained coverage |
|---|---|---|
activity-command.test.ts | names Emerald from the runner-owned environment id | Pins the human game title and playing suffix. The same file retains authored-versus-observed separation, control-byte stripping, authenticated-client dispatch and real settings persistence. No game/session schema or runner evidence is removed. |
connect-commands.test.ts | tells the owner how to finish each connection (merged) | Merge its existing formatter assertion into the retained argument-routing case; it does not need a separate test registration. Argument routing, authenticated provider probes, real follow-settings integration, stale writes, secret entry and preserved owner state remain in this file. |
devices.test.ts | reports an empty list clearly | Checks only exit zero and the authored empty-list sentence. devices-commands.test.ts retains successful empty-list command handling and the actionable pairing path; this file retains device IDs/grants/review provenance, JSON output, revoke, required ID, operator refusal, malformed/401/503 responses and redaction. |
herd-lead-companion.test.ts | says so when the console is not inside herdr | A formatter-only assertion for an error tone and the phrase herdr pane. The same file retains the actual outside-Herdr no-execution fence, missing binary, exact focus/split environment, self-pane-safe close and /board dispatch. |
The formatter assertion remains after merging, so its existing export stays
unchanged. No product code changes. The earlier queued variant
dcfe1725-bda3-4de6-b4a8-f02019cf059c is cancelled under the owner's revised
pruning rule: local package typecheck, deadcode and lint, then rebase and direct
main push. Product edits still use the merge queue.
Counts for apps/tui/test/**/*.test.ts: 155 → 155 files, 36,694 → 36,663
lines, 1,000 → 996 test registrations (four cases removed). Registrations
count literal it/test call sites, with each parameterized registration counted
once; this is not a claim that the entire suite ran. The four changed files'
actual focused Vitest cases are 47 → 43. The post-cut run additionally checks
both retained devices-commands cases. Focused tests and TUI package typecheck, deadcode and lint run locally under
clankie heavy; the owner-approved pruning path uses a direct main push. Raw focused reports and lint
results remain in the worktree's .local/pruning/.
Direct slice 1 checks passed: TUI source and all four test partitions, package
lint, deadcode for TUI plus its actual service consumers, and 45 focused cases.
The narrower TUI-only deadcode run reported UpdateAuthority as unused;
apps/clankie/src/runtime-update-routes.ts consumes it. Including that consumer
passes without any source edit or exemption. Direct-slice evidence.
VUH-2149: TUI slice 2 — duplicate helper coverage
Builds on slice 1. The merges retain unique assertions and remove repeated fixture invocations. No product code or test default changes in this slice.
Original file/case under apps/tui/test/ | Decision and reason | Retained coverage |
|---|---|---|
model-cli.test.ts: is a headless command | Merge the isolated classification case into the endpoint/persistence case; remove the separate test registration, preserving its assertion. | The headless recognition assertion now precedes actual command execution in the same case; endpoint probing/persistence, unreachable fallback/refusal, authenticated set/status refusal, catalog refresh, argument validation and unknown-command refusal remain. |
connect-commands.test.ts: is a bot+commands install for the stored application id | Merge into discord-commands.test.ts: builds a bot invite that does not use a signed 32-bit shift. Both invoke the same invite writer. | Every unique scope, permissions, positive-number and non-administrator assertion moves into the retained case. Client ID, exact permission value and explicit admin mode remain; no assertion is dropped. |
herdr-jump.test.ts: focuses the target through the herdr CLI | Merge into herdr-connection.test.ts: routes viewer, board, and jump commands to the authenticated service's binding. The broader case already invokes the same focus writer and checks its exact argv. | Move the complete success/target and formatted success-tone assertions into that binding case, which additionally proves selected socket, authenticated lookup and environment isolation. |
herdr-jump.test.ts: finds the id the click landed on | Move unchanged to herdr-connection.test.ts; consolidate the transcript-to-pane consumer with its native dispatch coverage. | Same exact click hit/miss and non-pane assertions. |
herdr-jump.test.ts: counts columns past styling and wide glyphs | Move unchanged to the same file; no separate fixture or module is needed. | Same ANSI and wide-glyph column assertions. |
herdr-jump.test.ts: reports herdr's own refusal | Move unchanged to the same file. | Same native refusal body and error result; no swallowed failure or success substitution. |
herdr-jump.test.ts: stays inert outside herdr | Move unchanged to the same file. | Same no-dispatch fence and skipped reason. |
Delete herdr-jump.test.ts only after those transfers. Counts for this slice:
155 → 154 files, 36,663 → 36,636 lines, 996 → 993 test registrations.
Actual focused Vitest cases: 62 → 59, all passing; this includes unchanged
next-step and status consumers. Registration counting follows slice 1's method.
Cumulative TUI cut: one file, 58 lines and seven cases, with all unique
trust-boundary and invite-contract assertions preserved. Formatter and lint
pass on explicit touched files. The TUI source and all four test partitions
pass the package typecheck on the final tree. The final merged model case also
passes its focused seven-case rerun. The owner-approved pruning path runs local package typecheck, deadcode and lint,
then rebases and pushes directly to main.
Raw focused reports remain in .local/pruning/; no full suite runs locally.
The 155-file inventory was screened for candidates. This bounded review does not establish that every remaining assertion earns its place. Files and cases without a specific removal/merge above stay unchanged, including evaluator inputs, native hooks and hire receipts, release/rollback, account consent, operator/worker authority, settings revisions and real HTTP/native integration.
VUH-2149 packages, final slice and retained inventory
Based on current origin/main after bf7fe53ec (the formatting repair).
This final slice removes two additional duplicate helper cases. No source, export,
timeout or remaining assertion changes.
| File | Removed case | Reason and retained coverage |
|---|---|---|
packages/settings/test/persona.test.ts | renders without character notes rather than inventing a personality | Duplicate authority sentence, already checked in states in every register that voice changes and authority does not. Its other assertion pins the incidental introductory sentence You are Clankie.; it never actually asserts absence of invented notes. describes each register instead of quoting a line he could say still renders all three registers with empty notes and retains the observed false-presence regression. Owner-authored verbatim notes, reply policy and authority/approval checks remain. |
packages/discord-presence-core/test/addresses-character.test.ts | stays quiet for ordinary conversation | Duplicate: real text-ingress.test.ts / stays quiet in an admitted channel until it is actually addressed checks ordinary room chatter is refused before any turn; still drops a message carrying neither text nor an image he can see checks empty text. Alias/case/punctuation, embedded-name/URL refusal, absent names and policy parsing remain in the helper file. |
| Test file | Before lines | After lines | Before runtime cases | After runtime cases |
|---|---|---|---|---|
packages/settings/test/persona.test.ts | 94 | 88 | 9 | 8 |
packages/discord-presence-core/test/addresses-character.test.ts | 40 | 35 | 5 | 4 |
| Final slice | 134 | 123 | 14 | 12 |
Final slice: 2 files → 2; 11 lines and 2 runtime cases removed. Its focused gate
also includes unchanged text-ingress.test.ts (44 cases) and the reviewed retained
model-provider.test.ts (62 cases): 120 → 118 passing cases. The config helper's
returns undefined when no .clankie.json exists up the tree case was considered
but retained: findRepoConfigPath is an exported, documented function and the
consumer fallback alone would not prove its direct absent-path return contract.
Assigned-area totals and scope
The inventory covers all assigned test paths, including the excluded manual-eval corpus as a static file. Runtime counts come only from focused collected tests; no eval, live-account operation or area-wide suite was run. Candidate review examined the removed cases, their producer/consumer paths and the retained cases named above. This is a complete path/registration inventory, not a claim that all 46,862 baseline lines received an exhaustive case-by-case review. Untouched entries stay retained; filenames alone do not justify removing them.
| Area | Before files | After files | Before lines | After lines | Before declarations | After declarations |
|---|---|---|---|---|---|---|
apps/discord-bridge | 21 | 21 | 5612 | 5608 | 152 | 151 |
packages/agent-hosts | 2 | 2 | 187 | 187 | 9 | 9 |
packages/api-client | 3 | 3 | 675 | 675 | 15 | 15 |
packages/credential-broker | 14 | 14 | 2063 | 2063 | 94 | 94 |
packages/discord-presence-core | 28 | 27 | 13896 | 13777 | 508 | 498 |
packages/fleet-resources | 17 | 17 | 4717 | 4717 | 109 | 109 |
packages/interactive-environment | 5 | 5 | 789 | 789 | 24 | 24 |
packages/lead-skin | 1 | 1 | 263 | 263 | 5 | 5 |
packages/media-connector | 1 | 1 | 425 | 425 | 17 | 17 |
packages/model-provider | 11 | 11 | 2705 | 2705 | 106 | 106 |
packages/model-registry | 2 | 2 | 380 | 380 | 16 | 16 |
packages/observability | 5 | 5 | 603 | 603 | 18 | 18 |
packages/persona-images | 2 | 2 | 405 | 405 | 14 | 14 |
packages/play | 11 | 11 | 3636 | 3636 | 146 | 146 |
packages/play-voice | 2 | 2 | 549 | 549 | 20 | 20 |
packages/protocol | 28 | 28 | 4061 | 4061 | 146 | 146 |
packages/rendered-surface-client | 2 | 2 | 380 | 380 | 9 | 9 |
packages/settings | 21 | 21 | 3101 | 3095 | 133 | 132 |
packages/vec3 | 1 | 1 | 66 | 66 | 4 | 4 |
packages/vox-client | 1 | 1 | 653 | 653 | 17 | 17 |
packages/work-items | 9 | 9 | 1696 | 1696 | 48 | 48 |
| Total | 187 | 186 | 46,862 | 46,733 | 1,610 | 1,598 |
Together the two source slices remove 129 test lines and 11 runtime cases; 6 changed test files become 5. Static declarations fall by 12, including the voice-session table merge, which keeps both runtime inputs and all assertions.
Protocol stays intact: response compatibility and immutable old-client shapes, strict requests, delivery/refusal/uncertainty, public field secrecy, namespace routing, lifecycle transitions, byte limits and the Node-free mobile import boundary are published contracts. Reviewed settings cases retain credential-free configuration, native account/profile identity, missing/malformed/replaced-file fences, private modes, migrations, owner policy, shared skill roots and verbatim persona notes. Bridge source-wiring tests remain where equivalent startup/cleanup coverage was not demonstrated. Voice disclosures, native media lifecycle, consent, grant revocation and audience boundaries remain. Other package files, including fleet-resource integrations, credential/provider transports, play records/eval inputs, observability redaction, vector consumer contracts and media bounds remain unchanged. No speculative dead-code deletion or export edit.
Landing incident and recovery
Slice 1 b68e3d898 passed focused tests, package typechecks, Knip and lint, but its
new review tables had not been formatted. The next CI run failed only this doc's
formatting. A local revert was prepared, then withdrawn before any push after
the lead directed a format-only repair. bf7fe53ec formats the complete shared
doc; its CI passed.
The tested cuts remained on main. This used the second of three permitted pushes;
the third groups the final source cuts, complete review counts and archived
checks. Whole-file review/doc/evidence formatting is now part of the focused gate.
Final measurements and archived reports include both failed attempts and the passing scoped gates. The ordinary root landing/merge queue was skipped under James's explicit pruning assignment.
VUH-2149 TUI slice 1: remaining display snapshots
Reviewed against 990a774a4cceaba7a5b3f6d2939d5529a70b8041. The assigned
area contains 155 test files. Existing VUH-1925 review decisions above remain
in force; file names and mocked seams alone do not justify deletion. This slice
removes only the following complete cases, with every retained assertion and
timeout unchanged. No production source, fixture, integration or golden is changed.
File under apps/tui/test/ | Deleted case | Reason and retained coverage |
|---|---|---|
activity-command.test.ts | names Emerald from the runner-owned environment id | Incidental game display copy. Keep authenticated activity dispatch, authored/observed separation, terminal-control stripping, actual persisted play toggles and budget writes. |
provider-commands.test.ts | labels xAI as SuperGrok or API key instead of a bare connected | Incidental provider hint and confirmation wording. Keep SuperGrok OAuth broker storage/code redaction, signed-in readiness, API credentials, hosted refusal, provider allowlists and authoritative model writes. |
voice-transcripts.test.ts | reprints a dim room header only when the body or channel changes | Incidental row layout: exact header placement, separator and repeated age strings. Keep overlay speaker/body/content rendering, missing-name attribution, distinct human/ack/answer identity, incomplete-speech disclosure, paging/schema/401 refusal, retention clearing and follow recovery. |
next-step.test.ts | asks to start Clankie when he is not answering | Incidental not answering prose; it asserts no recovery command. Keep the restart, pairing, remote-access and direct-route command cases. Service/doctor tests retain honest unavailable-state behavior. |
devices.test.ts | reports an empty list clearly | Incidental No paired devices prose; successful list exit is already asserted by table and JSON list cases. Keep device identity, grants, review provenance, revocation, malformed/refused reads and credential/secret boundaries. The TUI empty-list next-step case also stays. |
Measurements count physical lines in tracked *.test.ts files and static
it/test declarations (one declaration per .each, without expanding rows).
Support files and fixtures are excluded and unchanged. No new tests or merges.
| Scope | Before files | After files | Before lines | After lines | Before declarations | After declarations |
|---|---|---|---|---|---|---|
| Five changed files | 5 | 5 | 1,658 | 1,597 | 66 | 61 |
| Entire TUI test area | 155 | 155 | 36,694 | 36,633 | 992 | 987 |
Validation is recorded with the landed slice report on VUH-2149. The initial focused test startup failed because this new worktree had no dependencies; it is not a test pass. A governed frozen-lockfile install precedes the actual checks.
Slice 1 checks passed through clankie heavy: pnpm --filter @clankie/tui typecheck, pnpm lint apps/tui, pnpm deadcode, and the five changed files
with pnpm exec vitest run (64/64 retained registered cases passed). Formatting
ran on those five explicit files. Scoped Knip's UpdateAuthority warning was
a cross-workspace false positive: apps/clankie/src/runtime-update-routes.ts
and apps/clankie/src/index.ts consume it; the normal repository dead-code
check includes those consumers and passed without exemptions or source changes.
The assignment's typecheck/deadcode/lint gates replace the usual merge-queue
landing gate for these test-only slices.
VUH-2149 TUI final slice: duplicate recovery helper check
The first TUI push (c0f7f7dd9) was reverted immediately in 586ccd31b when
CI rejected review-document formatting. The reverted tree exactly matched
Ivy's b68e3d898 and also failed that same formatting check. This final slice
restores the five reviewed cuts above, adds the duplicate cut below, and
formats/checks the complete shared document while preserving Ivy's content.
This uses this worker's third and final push; no product source changes.
File under apps/tui/test/ | Deleted case | Reason and retained coverage |
|---|---|---|
next-step.test.ts | leads with signing the Mac back in when the doorway is signed out | Duplicate of retained signed-out recovery through status-connection.test.ts, install-doctor.test.ts (asks for a sign-in when the Mac is shut out of its own doorway) and pairing.test.ts (says a gatewayless code needs the sign-in when remote access is signed out). These consume the real helper from their command/doctor flows. The remote-access CLI command string remains in the helper's bare-install case. |
All other retained assertions and timeouts remain unchanged. Entire-area measurements use the same static declaration method above; registered counts for the focused runs are reported separately.
Retain model-cli.test.ts's headless-recognition assertion: command-handler and
help tests do not exercise the launcher's recognition decision. Concurrent
commit efe108509 merges it into the endpoint/persistence case without deleting
the assertion, preserving this published CLI contract. A real headless launcher proof against
an isolated loopback models endpoint also passed, including persisted model
selection; no owner configuration or live service was changed.
Deferred incidental candidate: voice-transcripts.test.ts's relative-age
formatting case is the only outside consumer of the formatVoiceTranscriptAge
export. Removing it needs a separate production visibility change through the
merge queue. It stays here; no unused-export exemption is added. Mixed files
with incidental copy beside boundary assertions stay intact. The prior protected
review inventory remains in force, and no integration/E2E/golden file is removed.
| Scope | Before files | After files | Before lines | After lines | Before declarations | After declarations |
|---|---|---|---|---|---|---|
Six reviewed cuts in five files, relative to 990a774a4 | 5 | 5 | 1,658 | 1,588 | 66 | 60 |
This final push's three files, relative to efe108509 | 3 | 3 | 1,295 | 1,240 | 50 | 46 |
Whole area, relative to efe108509 | 154 | 154 | 36,636 | 36,581 | 985 | 981 |
Whole area, including concurrent cuts, relative to 990a774a4 | 155 | 154 | 36,694 | 36,581 | 992 | 981 |
Net reduction: six complete cases and 70 test lines. The final helper cut
adds nine lines/one case to the first reviewed slice's reduction.
Five support/fixture files (128 lines) remain unchanged. Registered parameter
rows in unchanged .each declarations are preserved. The first focused
selection had 69 registered cases before the cut (64 retained plus five removed).
Concurrent TUI work landed as 2124e083b during this worker's rebase. It
already includes the activity-title and empty-device-copy cuts reviewed above,
plus the separate connect-status merge and board-copy cut documented in its own
section. Both sections and all its assertions are preserved. This final push
therefore contributes four cases/55 lines in three test files; the whole area's
combined change after efe108509 is one fewer file, eleven fewer declarations
and 113 fewer lines. That commit also consolidates the invite, model-recognition
and Herdr jump coverage with their unique assertions preserved.
The table uses this review's AST/line-start declaration method consistently;
parameterized cases are counted once, and no full-suite pass is claimed.
The final retained selection is the three changed files plus the real
doctor/status/pairing consumers. A syntax-tree comparison confirms that all
remaining case bodies in this review's changed files are byte-identical to
990a774a4; no assertion or timeout was weakened. The first 64-case focused
run and the later 71-case consumer run passed. model-cli.test.ts was restored
unchanged after its recognition case was found to protect the launcher contract;
all eight model cases passed separately. The real headless launcher proof also
passed against an isolated loopback endpoint and isolated persisted config.
Final rebased checks: the three changed files pass all 46 registered cases.
TUI typecheck (source plus four test partitions), TUI package lint, repository
deadcode and formatting of the three test files plus the complete shared
review passed through clankie heavy. A normal Knip configuration hint about
an unmatched entry glob remains informational; no unused-code exemption was added.