mirror of
https://github.com/callstack/agent-device.git
synced 2026-09-14 20:06:34 +08:00
perf/1961-cli-compile-cache
30 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Michał Pierzchała
|
cb65d6ca1f |
refactor(tests): replace the test-utils barrel with direct module imports (#1956)
* refactor(tests): replace the test-utils barrel with direct module imports The barrel re-exported 13 modules, so every importer evaluated all of them (store-factory alone drags 16 daemon session-store files; property-arbitraries drags fast-check). Importing the backing modules directly cuts the unit suite's aggregate eager module evaluations from 153,401 to 144,344 (-5.9%), measured with the eager-import-closure walker. Deleting the barrel makes the tax unrepresentable instead of pinning it with a guard test. * docs(testing): point fixture guidance at the test-utils modules, not the deleted barrel * test: extract replay session fixture |
||
|
Michał Pierzchała
|
5676d5ff8a |
refactor(apple): drop dead runner code and collapse duplicated helpers (#1936)
* refactor(apple): drop dead runner code and collapse duplicated helpers Removes declarations with no consumers (findScopeElement, interactiveTypes, two unused PresentedNode convenience inits) and collapses copy-pasted logic: DataPayload now relies on the synthesized memberwise init, TvRemoteButton is String-raw-valued, point-hit sorting shares smallestElementFirst, command-id trim-or-nil lives once on RunnerCommandJournal, scroll/desktopScroll share direction and durationMs validators, and the seven inline NSError refusals use unsupportedOperationError. elementTypeName reads a table pinned by the visibility-fold parity test. The packager now skips files whose unit-test blocks were their whole body, so 10 test-only files stop shipping (and stop compiling on user machines) as empty translation units. Packaged Swift: 432.3 kB -> 427.8 kB; 56 files instead of 66. Net -144 lines. * test(apple): pin skeleton-file exclusion in the packaging guard The strip fixture always kept runtime content, so reverting the skeleton-skip branch left every guard green. The new fixture's whole body is unit-test blocks; the packaged path must be absent while a non-skeleton sibling still ships. Observed red with the skip branch disabled before re-enabling it. * refactor(apple): tighten runner cleanup boundaries |
||
|
Michał Pierzchała
|
30de1597d3 |
ci: attribute native package size and trim Apple runner (#1934)
* ci: attribute npm package size by shipped component * refactor: modularize size reporting and trim Apple runner * ci: preserve size reporter modules across base checkout |
||
|
Michał Pierzchała
|
40e4b0dd3e |
docs(agents): restore and enforce progressive disclosure (#1888)
* docs(agents): restore and enforce progressive disclosure * test(maestro): pin typed selector fallback signal * docs(agents): address progressive disclosure review * docs(agents): restore orphaned traps and close guidance-gate bypasses - AGENTS.md: skills carry a minimal start/routing card; command semantics stay in versioned CLI help (the skills contract enumerates two skills by hand, so prose retains ownership for the rest) - testing.md: restore the two local-only XCTest snags CI never hits (unsigned-bundle policy refusal signature + first-run automation permission) - scripts/gate/routing.ts: record GitHub's 300-changed-file path-filter limit at the paths-ignore assertion it bounds - agent-guidance-contract.test.ts: recurse docs/agents so nested guidance cannot evade the byte budgets while the gate stays green |
||
|
Michał Pierzchała
|
250e30a578 |
test(bench): falsification fixtures for oracles + typed runner outcomes (#1893)
* test(bench): falsification fixtures for oracles + typed runner outcomes
Two deterministic PR-time quality gates for the help-conformance bench
(the repo's single non-gating small-model planning oracle):
- Every EXPECTATION_SCORERS entry in help-conformance-case-checks.mjs
now has a falsification fixture (a minimal passing witness plus at
least one known-bad counterexample, and a metamorphic variant where
useful) in the new help-conformance-expectation-fixtures.ts, run
through the real validatePlanCommands/scoreExpectations pipeline.
help-conformance-expectation-falsification.test.ts is the "what
enumerates N" completeness gate: a new named expectation with no
fixture fails it. Counterexamples cover swallowed lifecycle command
prefixes, unsupported flags/selectors, pseudo refs, shell operators,
and invalid positional ordering.
- help-conformance-runner-output.mjs now returns a discriminated
RunnerOutcome ({kind:'success',commands}|{kind:'runner-error',
message,reason}) instead of a raw-string success inference. Only a
'success' outcome ever reaches validatePlanCommands/scoreExpectations
in runCase, so a runner-error result can no longer also carry
model-validation checks, and an all-runner-error aggregate now
reports passRate: null (rendered as "N/A") instead of "0/0 (0%)".
Fixes #1481
* refactor(bench): dedupe RunnerOutcome construction, drop leftover narrowing
Thermo-nuclear pass over 4b2df0a38's diff:
- help-conformance-bench.mjs's runOutcome() catch block was hand-building
the exact {kind:'runner-error', raw, message, reason} shape that
runner-output.mjs's private runnerError() helper already constructs for
its own two error paths. Export it as runnerErrorOutcome so the
discriminated union has exactly one constructor for its error variant,
reused by both error sources instead of duplicated.
- runCase's two return branches repeated the same
{runner, caseId, trial, outputPath} fields; pulled into a shared `base`
object.
- Reverted bench.test.ts's rateLimitedOutcome block: it had an explicit
`: RunnerOutcome` annotation and an if/throw narrowing guard, added only
to give fallow's dead-code checker a "real consumer" of the type before
the actual fix (adding the .d.mts to .fallowrc.json's ignorePatterns,
matching the existing sample-outputs.d.mts precedent) was found. That
workaround is now unnecessary scaffolding — replaced with the same
flat assert.deepEqual style the surrounding assertions already use.
|
||
|
Michał Pierzchała
|
735ab7672a |
refactor(daemon): one capture-input builder and one admit-then-bind step (#1876)
Behaviour-neutral. No descriptor changes platform execution, the cutover table is untouched, and no contract surface is added. - buildRuntimeCaptureInput moves to its own module so every request-bound capture consumer builds CaptureSnapshotInput one way. - The admit-then-bind sequence in the snapshot/diff resolver becomes one named step, ready for the selector units' second caller. - handlers/find.ts splits into focused target-capture and match-resolution concepts (600 -> 346 lines); behaviour unchanged. Co-authored-by: agent <agent@local> |
||
|
Michał Pierzchała
|
d07b837621 |
test: classify the runner XCTests — pure decisions to a macOS host lane, simulator semantics gated os(iOS) (#1781 A7) (#1861)
Every declared AgentDeviceRunnerUITests method now belongs to a lane, and the #if guard is the classification: AGENT_DEVICE_RUNNER_UNIT_TESTS alone means a pure runner decision (runs on the macOS host on every PR — ci.yml's existing compile job now executes the bundle it builds), '&& os(iOS)' means runner/XCTest semantics (simulator lanes only). check:xctest-selection evaluates the guards per platform, derives each lane's reach, and fails on a flagged identifier that is undeclared or uncompiled on that lane, on a declared test no lane reaches (found the two tvOS-only tests, dark since birth — widened to os(tvOS) || os(macOS)), and on testCommand reaching any lane. The host and nightly lanes assert executed == derived reach, so a missing -D flag or a guard that compiles a file out reads red, not as a smaller green. One duplicate test deleted (sparse-verdict assertions folded into its twin). |
||
|
Michał Pierzchała
|
f03c0309a1 |
fix: derive iOS transition snapshots from visible presentation (#1831)
* fix: project iOS transition semantics * fix: derive iOS transition semantics from visible state * fix: preserve iOS presentation context for scoped snapshots * fix: confirm broad iOS transition settlement * ci: run coordinate input regression on pull requests * test: mock migrated snapshot capture seam * fix: confirm transitions across snapshot backends * fix: arm transition confirmation after first capture * fix: settle against immutable action baseline |
||
|
Michał Pierzchała
|
6a8beb653e |
feat(mcp): compact server instructions in both eras + MCP-only help tool (#1839)
* feat(mcp): compact server instructions in both eras + MCP-only help tool (#1833) MCP-only clients got no workflow guidance: server/discover carried two sentences, legacy initialize carried nothing, and the CLI guides (agent-device --help, help <topic>) were unreachable over MCP. - MCP_SERVER_INSTRUCTIONS: one MCP-phrased workflow card (<2 KB, the Claude Code truncation limit) returned by server/discover and legacy initialize alike. - help tool, router-owned (not a command descriptor): no topic -> the CLI decision card; topic -> agent-device help <topic|command> text, prefixed with the one-line CLI->tool-property mapping; unknown topic -> isError listing the topics. listCommandTools() stays descriptor-only for the AI SDK; the router composes descriptors + help. - Move src/cli/parser/cli-help{,-overview}.ts to src/cli-schema/ so src/mcp (rank 3) can import the renderers without a layering back-edge into src/cli (rank 6). * fix(mcp): name terminal-only commands in help guides; colocate cli-help tests with their sources - The MCP guide preamble claimed every `agent-device <command>` line is a tool of that name; `help web` tells the reader to run `web setup` / `web doctor` and no `web` tool exists. The preamble now lists the exact CLI-only set (listCliCommandNames minus listMcpExposedCommandNames) — derived, not scanned out of prose where `device`/`web` are ordinary words. Regression: help web names `web` as terminal-only, and the listed set equals the registry difference. - cli-help-*.test.ts move from src/cli/parser/__tests__ to src/cli-schema/ to mirror the moved sources. * perf(mcp): tighten the guide card, tool description, and preamble Instructions card 1572 -> 1378 bytes (paid every session), tool description and preamble trimmed, HELP_TOOL built once as a const. Bundle delta vs main 3189 -> 2715 bytes; the remainder is the guide text itself, which the bundle carried in no MCP-phrased form before. |
||
|
Michał Pierzchała
|
142d156338 |
ci(ios): run the full XCTest suite nightly and check the PR test list (#1781 A7) (#1789)
* ci(ios): run the full XCTest suite nightly and check the PR test list (#1781 A7) * fix(ci): skip the runner server entry point in the nightly and validate both test flags * docs(ci): restate the nightly lane cost and timeout honestly * docs(ci): stop quoting XCTest counts that drift between commits * ci(ios): tighten the nightly timeout to the measured suite duration |
||
|
Michał Pierzchała
|
801734d433 |
feat(ai-sdk): add agent-device/ai-sdk tool set and document the MCP zero-code path (#1804)
* feat(ai-sdk): add agent-device/ai-sdk tool set and document the MCP zero-code path
Adds `createAgentDeviceTools()` under a new `agent-device/ai-sdk` subpath,
built from the same command registry the MCP server uses so both stay in
lockstep without a hand-maintained tool list. Introduces a `frameworkTier`
descriptor facet ('core' | 'extended') so the factory can default to a
curated perceive/act loop instead of handing a model dozens of tools.
`ai` is wired as an optional peer dependency, imported lazily inside the
factory rather than at module scope, so importing the subpath itself never
requires `ai` to be installed - only calling it does. The package's own
publishing gate (scripts/lib/shipped-imports.ts) is extended to recognize
peerDependencies as a valid resolution source, since this is the first
optional peer this package has shipped.
Also restructures the AI SDK doc around three tiers (zero-code via
@ai-sdk/mcp, the new typed tool set, hand-written tools) and fixes a stale
`needsApproval` reference in favor of the current `toolApproval` API.
* fix(layering): classify src/ai-sdk as a rank-4 zone
The layering guard requires every src/<folder>/ to be explicitly ranked or
unranked; the new src/ai-sdk/ subpath (added in the prior commit) was left
unclassified, failing CI's Layering Guard job. It sits at the same tier as
client/compat/daemon-server/metro/remote/sdk - a public integration surface
consuming mcp (3) and core (2), imported by nothing else in the tree.
* fix(ci): cover, exempt, and pack the new ai-sdk subpath
Fixes the remaining CI failures on the ai-sdk subpath commit:
- Coverage: src/ai-sdk/index.ts had no dedicated unit test (only manual/
integration verification), so changed-line coverage sat at 6.9% against
the 70% gate. Adds src/ai-sdk/__tests__/index.test.ts (core vs 'all' tool
filtering, session/platform pinning and schema hiding, error
normalization, toolApproval passthrough) with createCommandToolExecutor
and createAgentDeviceClient mocked the same way command-tools.test.ts
does, plus a dedicated missing-peer-dependency.test.ts that mocks `ai`
itself to throw, isolated to its own file so it doesn't affect the other
tests' use of the real, installed `ai` package. Changed-line coverage is
now 29/29 (100%).
- Fallow Code Quality: src/ai-sdk/index.ts and examples/sdk/ai-sdk-tools.ts
are entry points with no in-repo importer (reached only via package.json
exports / run directly), and the new subpath's exports are unused
internally by design - both need the same treatment src/sdk/*.ts and its
examples already have in .fallowrc.json.
- Integration Tests: test/integration/installed-package-metro.test.ts and
src/__tests__/package-exports.test.ts each hand-list every published
subpath and smoke-check it from a real packed install; added ./ai-sdk to
both so the new subpath is actually exercised, not just silently passing.
* fix(ai-sdk): hide MCP transport/config fields from the model too
createAgentDeviceTools() only removed session and mcpOutputFormat from tool
schemas. stateDir was still model-visible and reached the shared executor
as client configuration, letting a tool call redirect into a different
daemon state directory - defeating the "one pinned session" guarantee the
factory exists to provide. includeCost and responseLevel are MCP
tool-config knobs in the same category, irrelevant to this adapter.
Widens the hidden-field set to session/stateDir/mcpOutputFormat/
includeCost/responseLevel, and now strips them from the runtime input
inside execute() too (not just the schema), so the guarantee holds even if
a caller bypasses schema validation. The schema-properties filter and the
input filter now share one omitHidden() helper instead of two near-
duplicate implementations.
Addresses the P1 review comment on #1804.
|
||
|
Michał Pierzchała
|
04613ae8d3 |
ci: keep Bundle Size job green on transient GitHub comment failures (#1795)
* ci: keep Bundle Size job green on transient GitHub comment failures The size measurement and job summary had already succeeded on PR #1789 (run 32050847506) when the PR comment write got a 503 during a GitHub incident and failed the whole lane. --post-comment now retries 5xx / 429 / network errors (4 attempts, 1s/2s/4s backoff) on both the list and write calls. If it still fails, it prints a ::warning::, appends a note to $GITHUB_STEP_SUMMARY, and exits 0. Other 4xx (bad token, missing permissions) stay fatal. * refactor: split GitHub response classification to satisfy fallow complexity gate * fix: reconcile uncertain comment creates instead of re-POSTing; add regressions Retry now wraps the whole list -> write cycle rather than each request, so a create whose response was lost (network error / 5xx) is re-listed on the next attempt and turned into a PATCH of the marker comment instead of a duplicate POST. Splits the retry/classify helpers under the fallow complexity gate. Adds scripts/__tests__/size-report-post-comment.test.ts (unit-core): spawns the real script against a stubbed fetch and pins uncertain-create reconciliation, transient exhaustion (warn + exit 0), and fatal 4xx (nonzero, no retry). SIZE_REPORT_RETRY_BASE_MS lets the tests skip real backoff. |
||
|
Michał Pierzchała
|
c7565cb1f8 |
refactor(snapshot): clean snapshot ownership (#1754)
* refactor(snapshot): clean snapshot ownership * fix(snapshot): address ownership review feedback |
||
|
Michał Pierzchała
|
7f5dbd2e50 |
chore: drive unused production exports to zero (#1743)
* chore: drive unused production exports to zero `pnpm check:production-exports` has been failing on main with 21 findings. Each was investigated rather than blanket-suppressed; they split three ways. Genuinely dead, deleted: - `androidDeviceForSerial` (android/adb.ts) had zero references anywhere, tests included. - `streamAndroidLogcatWithAdb` (android/logcat.ts) had no production consumer and only a guard-clause test; its `captureAndroidLogcatWithAdb` sibling is the published SDK surface. Removed with its options type and test. Test-only aliases over live siblings, collapsed: - app-log-resource-store re-exposed four bound store methods; production used only `resolvePath`, tests used the other three. The sibling screen-recording-resource-store exports just the store, so this now matches: one export, all consumers call `appLogResourceStore.x`. - device-claims re-exported `canonicalLocalDeviceKey` for a single test, while production imports it from device-claim-paths directly. Dropped the re-export and pointed the test at the canonical module. Real consumers the analysis cannot see, exempted with the reason: - The nine remaining `src/cli/commands/*Command` handlers are reached only through `dedicatedCliCommandHandlerLoaders`, the dynamic import() table in router.ts. `deviceCommand` already carried an inline suppression for exactly this; replaced it with one config entry naming the table that enumerates all ten, matching the existing daemon route-handler entry. - `resolveVitestMaxWorkers` (vitest.config.ts), `DEVICE_CLAIM_IN_USE_SAMPLE` (bench sample producers) and the capture-kit `createAppLogLiveHandle` facade export joined the existing entries that already record their exact shape. - `**/*.fixtures.ts` is now an ignorePattern: all 15 build doubles for co-located tests, several import `vi`, and none is imported by production source. Pattern-matching them as test infrastructure also keeps this class of finding from recurring. Gate now reports zero. Unit suite, layering, fallow audit, MCP metadata, build, bundle-owner and package checks all pass. * chore: scope the fixture exemption to unused exports Review feedback on #1743: `ignorePatterns` removes a file from every Fallow mode and rule, but the false positive here is only production-unused-exports. Moved *.fixtures.ts to an ignoreExports entry so fixtures stay inside health, dead-code and cycle analysis. Kept `exports: ["*"]` rather than today's three symbols because the property is per-file — no fixture module has a production consumer — so a new fixture symbol should not reopen the finding. check:production-exports still reports zero, and a full `fallow --summary` returns identical totals (2 dead-code / 4 dupes / 123 health) with and without the change, so nothing was newly surfaced or newly hidden. * test(fallow): prove fixture policy scope |
||
|
Michał Pierzchała
|
f5d9789764 |
feat: enforce local device claims and reconcile stale owners (#1735)
* feat: enforce local device claims * fix: address device claim review feedback * fix: persist canonical daemon claim state directory |
||
|
Michał Pierzchała
|
b0d4b40467 |
chore: stop publishing skills to npm (#1730)
* chore: stop publishing skills to npm * fix: align simulator skill startup * docs: align agent setup with open-first workflow |
||
|
Michał Pierzchała
|
c2c81549d9 |
feat: add simulator verification skills (#1716)
* feat: add simulator verification skills * chore: simplify simulator skills * docs: refine simulator skill guidance * test: guard simulator skill workflows * style: format simulator skill contract test |
||
|
Michał Pierzchała
|
cdc754e6ed |
perf: speed up iOS agent recovery and streamline CLI guidance (#1700)
* Avoid interactive children in parent taps * docs: streamline no-skill CLI help * perf: recover faster from sparse iOS trees * fix: preserve selector context for blocked parent taps * fix: preserve coordinate text-entry focus * fix: preserve thin parent touch targets * fix: fail closed for unscoped iOS typing * test: isolate replay lock fixture * test: share node integration process |
||
|
Michał Pierzchała
|
18291ba8e2 |
perf: collapse app-driving startup turns (#1693)
* perf: collapse app-driving startup turns * fix: align foreground open guidance |
||
|
Michał Pierzchała
|
9c25bc66f4 |
docs(cli): advertise open --foreground and snapshot --actions in the workflow card (#1682)
* docs(cli): advertise open --foreground and snapshot --actions in the workflow card open --foreground (#1670/#1671) and snapshot -i --actions (#1665) shipped with no mention in the compact `help workflow` card, so a planning model never discovers either. Add one terse line each: the foreground fast-path in Bootstrap, and the merged-element custom-action guidance in Validation and evidence. Stays under the 9,000-byte compact-card budget (8493 -> 8908 bytes). Adds two help-conformance bench cases per the repo's changed-guidance rule: foreground-attach-single-sim (correct plan starts with `open --foreground` in an unambiguous single-sim scenario, fail-closed alternative forbidden) and merged-card-actions-not-directly-invokable (a merged Bluesky-style feed card's actions list is evidence, not a selector). Both use a real pinned sample rebuilt through the production snapshot renderer. * fix(scripts): accept flag order in the foreground-attach conformance matcher Flag order after `open` isn't semantically meaningful (`open --platform ios --foreground` is exactly as correct as `open --foreground --platform ios`), but startsWithForegroundOpen required --foreground to be the literal next token after `open`. Rescoring the completed repeat=3 bench report shows this docked codex:gpt-5.4-mini on all 3 trials even though its plan was config-order noise, not a real deviation -- the no-positional/no-device guarantee already comes from the forbidden checks. Loosened to require --foreground anywhere on the open line; foreground-attach-single-sim now scores 54/54 across both runners. * fix: close workflow help conformance gaps |
||
|
Michał Pierzchała
|
a158434a9c |
feat(cli): compact workflow help card + version header (#1663)
* feat(cli): compact workflow help card + version header Shrinks the per-task agent protocol tax of the help/skill surface. `agent-device help workflow` drops from 41025 to 8466 bytes (-79%) by moving depth into new `help scripting` (save-script, secret-safe fills, batch JSON, replay divergence/repair) and `help gestures` (multi-touch shapes/quirks) topics, and folding a few paragraphs into topics that already owned the subject (help debugging, help physical-device, help validate). Content is moved, not deleted. Every `help <topic>` first line is now `agent-device <version> — <topic>`, so the skill router reads the CLI version off the mandatory first help read instead of a separate `agent-device --version` call. SKILL.md is updated to do that and stays a thin router otherwise. The compact card also gains two terse behavioral rules: chain confident consecutive steps with `&&` (falling back to one command at a time when uncertain), and confirm the requested end state is actually visible on screen before declaring a task done. help-conformance-bench (22 cases x 2 runners) improves after the change: 29/44 -> 32/44 passing checks. * fix(cli): review follow-ups on the compact workflow card (#1663) Three fixes from PR review: - Extend the help-conformance plan validator to split a command line on unquoted && and validate each chained segment independently, so a plan that follows the workflow card's "chain confident consecutive steps with &&" guidance is accepted instead of rejected as one shell-projection violation. && inside a quoted selector value (e.g. label="A && B") is not a chain boundary and does not split. Adds unit tests for the splitter and a chains-confident-consecutive- settle-steps conformance case. batch stays out of this: it is deliberately stop-only. - Replace the literal @ref placeholder the compact card used in its own "snapshot -s @ref" example with a concrete ref (snapshot -s @e12 (the current concrete ref)), matching the same card's rule against placeholder targets. Reverts the test to demand the concrete shape. - Give help scripting and help gestures real conformance cases instead of waivers: a secret-safe recorded-fill + publish case, and an Android transform-then-verify case whose exact verification text only appears in the gestures topic. Removes both waivers. help-conformance-bench (25 cases x 2 runners, repeat=1) after these fixes: two full runs landed at 32/50 and 33/50. That is on par with the pre-change baseline (29/44) once the topic-untouched cases' run-to-run swings are accounted for (confirmed noise: one case with zero exposure to any change here flipped 10/10 -> 1/10 on a runner API error, and another swung across all three post-fix runs). The new scripting case now passes 8/8 for both runners; the new chaining case correctly reports the model's choice not to chain as a soft signal, not a validator failure. * fix(cli): update session.test.ts help pointer for moved script-authoring content * fix(cli): reject empty && chain operands in the plan validator (#1663) splitOnUnquotedAnd() previously trimmed and filtered out empty segments, so a plan with a leading (`&& press ...`), trailing (`press ... &&`), or doubled (`a && && b`) operator passed validPlanCommands even though a real shell rejects all three as a syntax error. The validator would bless a plan that fails at execution. Empty segments are now surfaced as an `empty-chain-operand` issue instead of being silently dropped. The quoted-&& non-split behavior (label="A && B") is unchanged, and a normal single command with no chain still parses identically to before. Adds regression tests for all three empty-operand shapes plus the quoted-&& case. |
||
|
Michał Pierzchała
|
3937036e5e |
feat: support --settle on scroll and back (#1638) (#1650)
* feat: support --settle on scroll and back (#1638) Scroll-then-observe and back-then-observe are legitimate agent pairs, but the post-action observation registry never grew past the touch commands, so `--settle` on either was rejected with INVALID_ARGS — burning a tool call each in AppControlBench's bsky-16. Both commands now carry the `settle` descriptor trait, and every surface derives from it rather than a hand list: CLI allowed flags, MCP/SDK input fields, the flag-sourced timeout envelope, and MCP ref-pinning. The CLI flag/metadata helpers moved out of the interaction family into post-action-observation-grammar.ts (back is a system command), and SETTLE_REF_ISSUING_TOOLS became a derivation — a hand list would have silently stopped pinning the new commands' refs. settleAfterInteraction and the new settleObservationCommand are two entry points over one engine: same loop, storage, hints, and diff bounds, with the target-less path supplying its own baseline and no proximity point. The daemon reaches that command through the runtime surface, never by importing `commands/` (R2) — the same seam the touch handlers use for press/fill — and generic-settle.ts is loaded through a lazy `await import` returning a closure, so the interaction runtime subgraph stays out of this dispatcher's static graph (a static edge folded ~18 files into the daemon-server type cycle; R10 caught it). Both of generic-settle's orderings are load-bearing and tested: the baseline is frozen before dispatch (and before the Android dialog preflight), and the observation runs after markDeferredInteractionOutcome so settle's first capture folds in the #1542 stabilization rather than racing it. The ADR 0014 "a settled diff publishes refs" rule moved to settle-ref-issuance.ts, shared by both routes. One divergence is deliberate: scroll/back resolve no element, so the diff baseline is the session's STORED pre-action tree — "settled tree vs the last tree you observed" — not press's freshly resolved pre-action capture. Both commands also switch to preserve-daemon on timeout, which changes the non-settle path too: with --settle their dominant hang mode is now a wedged accessibility bridge, and a timed-out capture must not reset the daemon and lose every session (#1105). The reviewed-set gate records it. Live-validated on an iOS 26.2 simulator (Settings): scroll --settle settled in 1786ms with a +6/-6 diff carrying fresh refs; back --settle in 771ms with +15/-6. Alternating cost runs, one call vs the pair it replaces: scroll 2.9-3.0s vs 5.3-5.6s, back 3.1-3.2s vs 4.7-5.1s. Those include the #1627 deep-capture extension. * fix: render settled-diff refs paste-ready in CLI output A settled diff activates a PARTIAL ref frame (ADR 0014), which admits only the pinned `@eN~s<gen>` form of the refs it issued. The unchanged-interactive tail already rendered that way, but the diff's own added lines rendered the bare `@eN` embedded in the snapshot line — so a CLI caller who copied the ref the diff just handed them got `plain_ref_requires_complete_frame` and had to append the generation by hand. Added lines now render pinned when the response carries `refsGeneration`, exactly like the tail. Removed lines render verbatim: they name elements that just left the screen, and `SettleDiffLine` never gives them a ref. This is not new to scroll/back — press/click/fill/longpress had the same gap since #1101. MCP was never affected: its ref-pin store rewrites plain refs on the way in, which is why the model never sees a suffix. Live: `scroll down --settle` now emits `+ @e14~s218078 [cell] "Game Center"`, and `press @e14~s218078` copied straight out of that line taps successfully. * test: record the pinned-diff-ref bytes in the output-economy baseline Rendering added diff-line refs pinned costs 8 bytes in the two settle CLI text samples (two `~s<gen>` suffixes). The output-economy baseline is the tripwire for exactly this, so the increase takes an explicit reviewed waiver rather than a silent baseline bump — the same one the settled TAIL's pins already carry, for the same ADR 0014 reason. Only `bytes` moves: lines, refs, hints, and shape are unchanged, which is the evidence that this is a suffix on existing refs and not a new payload. Caught by CI, not locally: `pnpm test:unit` runs unit-core and subprocess-stub only, while the Coverage lane runs every vitest project. * test: prove the generic settle degrades when its runtime cannot be built `createGenericSettleRuntime` catches and returns undefined so an observation that cannot even start does not fail an action that already succeeded. That was a claim in a docstring with nothing behind it — the one changed line the coverage gate reported uncovered (95/96). The test puts the session in the state the catch exists for: the router handed us a session that is no longer in the store, so building the settle runtime throws SESSION_NOT_FOUND. The response keeps its scroll result and simply carries no settle payload. Removing the try/catch fails it. * build: teach fallow that vi.mock reaches pinOwnProcessStartTime dynamically Not from this PR: #1642 added `pinOwnProcessStartTime` on main, and its three consumers reach it the only way a Vitest module mock can — `vi.mock(path, async (importOriginal) => (await import('...')).pinOwnProcessStartTime(...))`. Dependency analysis cannot follow that dynamic import to a consumer, so the export reads as dead the moment any PR pulls that file into its audit scope. This PR is the one that did. The entry records the consumers by path and the reason, matching the daemon route-handler entry directly above it, which exists for the same dynamic-`import()` limitation. * refactor: adopt the best of the parallel #1653 implementation Two sessions independently built #1638 (PR #1650 and PR #1653) and converged on the same architecture — trait in the registry, one engine with two entry points, runtime-command seam, lazy import, preserve-daemon, stored-baseline honesty. #1650 continues; this folds in what #1653 did better: - The agent-facing help core loop (cli-help.ts) now names scroll and back as settle-capable. Without this, the benchmarked closed-grammar help line kept instructing agents that --settle is only for press/click/fill/longpress — actively steering the AppControlBench models away from what #1638 shipped. - issueSettleRefs moves into session-snapshot.ts, beside the partial-frame primitive it wraps, deleting the single-function settle-ref-issuance module. - Their seam tests: back reader→writer settle plumbing, back CLI settle rendering, and a trait-less generic command (home) ignoring a stray settle flag rather than observing or rejecting. What #1650 had that #1653 lacked, for the record: the SETTLE_REF_ISSUING_TOOLS registry derivation (without it, MCP never pins a scroll/back settle diff's refs and the partial frame rejects every follow-up), BackCommandResult.settle in contracts, back's MCP output schema, paste-ready pinned diff refs, and the docs/changelog/baseline surfaces. * bench: help-conformance case for settled scroll-to-find planning The #1638 extension of the closed --settle grammar to scroll/back is the feature's entire payoff — collapsing scroll-then-observe into one call — and the closed command list is an enumerated N whose enumerator is this bench. The regex over the help text proves the sentence exists; this case checks whether a model plans differently because of it. One focused case, deliberately not coached: a pinned visible-first snapshot (rendered by formatSnapshotText, pinned by the sample-producers gate) whose wanted row is summarized off-screen with no ref anywhere in the output. The tempting pre-#1638 plan is `scroll` plus a separate `snapshot -i`; acceptance is the single settled call. Scoring was verified against eight plan shapes in both directions before recording. Model-backed record (claude-haiku-4-5, 3 trials, current help): 0/3 — but the decomposition is the finding. Settle eligibility GENERALIZED (3/3 trials put --settle on scroll unprompted; the mutation-suffix framing concern did not materialize) and the two-call habit is residual (1/3). All three trials failed on `scroll @e3 down --settle` — the pre-existing #1366 scroll-takes-no-target confusion, which the live CLI recovers with a dedicated hint but a single-shot bench cannot. The recorded gap is therefore a first-30 doc gap (nothing teaches that scroll takes no target), not a settle-eligibility gap; tuning the case until it passes would just delete the evidence. |
||
|
Michał Pierzchała
|
8ba5f9b8de |
fix: surface AMBIGUOUS_MATCH candidates and name find's supported actions (#1602)
* fix: surface AMBIGUOUS_MATCH candidates and name find's supported actions (#1597) AMBIGUOUS_MATCH errors now list the matching candidates (ref, role, label/identifier) rendered the same way as snapshot -i lines, capped at 5 with a "+N more" marker. buildAmbiguousMatchError (the single producer, src/daemon/handlers/find.ts) reuses formatSnapshotLine to build the list; formatAmbiguousMatchCandidateLines (src/utils/output.ts) renders it unconditionally on both text surfaces an agent actually reads (CLI printHumanError and MCP formatToolErrorText) — previously the candidates lived only in details, which neither surface printed. find's "Unsupported find action: X" (e.g. from `find <text> press`) now attaches a hint naming every action find actually supports and the two-step recovery shape: run find "<text>" to resolve the ref, then dispatch the gesture as its own command (press @eNN). The hint is a single exported constant (UNSUPPORTED_FIND_ACTION_HINT) shared by both throw sites — packages/selectors' raw-token parser and the CLI's typed reader (src/commands/interaction/selectors.ts) — so they can't drift. Matching semantics are unchanged; ambiguous rejection stays by-design. The help-conformance corpus's AMBIGUOUS_MATCH quiz is updated: its premise ("candidate refs were not shown") no longer holds, but with 3 identically-labeled candidates the lesson (don't guess a specific ref) still holds. * fix: guard the AMBIGUOUS_MATCH candidate renderer against device-domain shapes Review on #1602 (P2): formatAmbiguousMatchCandidateLines ran for every normalized error and stringified details.candidates unconditionally, but device-domain AMBIGUOUS_MATCH/APP_NOT_INSTALLED errors (findBootedAppleSimulatorWithApp, src/core/dispatch-resolve.ts) reuse that key for { id, name } device objects with no `matches` field — CLI and MCP would have printed "Candidates: [object Object]" for those. The renderer now requires numeric details.matches AND every candidate to be a string before rendering anything, restricting it to buildAmbiguousMatchError's element-match shape; unrecognized shapes render nothing, same as before this feature existed. Added regression tests against the exact device-error shape on both text surfaces. Also unexports AMBIGUOUS_MATCH_CANDIDATE_LIMIT (fallow flagged it as an unused production export) — it has no consumer outside find.ts. |
||
|
Michał Pierzchała
|
80feff42d6 |
build: verify the published tarball instead of grepping the bundle (#1578)
* build: verify the published tarball instead of grepping the bundle Replaces the bundle-dependency grep with one gate that packs the tarball npm would publish and proves it sound from a clean consumer install: publint and attw on the tarball, a two-way dependency-closure audit, an import of every `exports` subpath, and the CLI smoke run — all from outside the workspace, where no pnpm link can mask an unresolvable specifier. Also stops the build from emitting a publishable bundle in the first place: a missing workspace link now fails `pnpm build` instead of warning and exiting 0, which is how 0.20.4 shipped an unresolvable `@agent-device/ad-script` import. publint found 12 real defects in the current package — every `exports` entry listed `types` after `import`, so TypeScript resolved declarations by accident rather than by condition. The dependency audit found `pngjs` declared as a runtime dependency while tsdown inlines it, an install every user paid for and no shipped code reached; it moves to devDependencies. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NamFJUgn9DGHrT2za11JbD * fix(ci): run the package gate without pnpm on the Node floor pnpm 11.17 requires Node >= 22.13, so `pnpm check:package` could not start on the 22.12 floor the Packaged CLI job exists to cover. The gate needs only `node` and `npm`, so the job invokes the script directly. Splits the dependency-closure audit into a collector and a message builder to clear Fallow's complexity threshold, and classifies both packaging linters in ignoreDependencies: they are subprocess CLIs with no importable API here, which dependency analysis cannot follow to an import. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NamFJUgn9DGHrT2za11JbD * fix(publishing): read every literal resolution form in the closure audit The dependency-closure audit derived shipped imports from the ESM module record alone, so it could not see a package resolved through `require` or a `createRequire` result: neither produces a module-record entry. A lazy `createRequire('@agent-device/…')` would therefore clear the audit, the all-export probe and the exercised CLI paths, reintroducing the 0.20.4 published-install failure class for another command. Measuring the built bundle turned up a second, larger hole in the same reader. The shipped files are minified, and the minifier rewrites every string literal to a no-substitution template literal, so the dynamic-import extraction — which accepted quoted strings only — matched 0 of the 99 dynamic imports the bundle contains. The lazy `import()` path that broke 0.20.4 was reported as covered while checking nothing. Specifiers now come from the module record plus an AST walk over every literal runtime-resolution form: `import()`, `require()`, `require.resolve()`, an immediately-invoked `createRequire(...)`, and calls through a `createRequire` result under any import or minified alias. Both spellings of a string literal count everywhere, and `.cjs` joins the scanned extensions. Computed specifiers stay explicitly out of scope, and are pinned as such. Rejecting them is not available: minifiers reuse short identifiers across scopes, and the packed bundle really does contain an unrelated `a(h[t],f,g,l,e,m)` that no name-based match can distinguish from a require call. Those are covered by the gate's runtime half instead, which resolves them for real. Bare-identifier calls need the one-string-argument shape for the same reason. The audit moves to scripts/lib/shipped-imports.ts so fixture packages can exercise it. The gate needs a real `npm pack` behind minutes of Swift and Android builds, so every check that runs it can only watch a healthy package pass — which is how a reader that matched nothing looked covered. The new fixtures assert the failure direction per resolution form: 16 of the 22 fail against the previous reader, and the 6 that pass are the quoted-spelling and pinned-limitation cases. A wiring assertion keeps the audit and both runtime probes attached to the gate, since fixtures alone would stay green if the call were deleted. Verified against the real built bundle: the closure resolves to exactly the two declared dependencies, so the stricter reader adds no false positives. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NamFJUgn9DGHrT2za11JbD --------- Co-authored-by: Claude <noreply@anthropic.com> |
||
|
Michał Pierzchała
|
2e74b789fd |
feat: verify device cloud connections (#1564)
* feat: verify device cloud connections * refactor: unify connect provider adapters * refactor: separate connect verification facts * fix: tighten connect provider verification * fix: use neutral cloud connection wording * perf: deduplicate local affected checks * refactor: simplify affected check runner * refactor: derive connect workflow from verification |
||
|
Michał Pierzchała
|
76453add71 |
refactor: pnpm workspace + @agent-device/kernel pilot (#1490 W0) (#1494)
* refactor: pnpm workspace + @agent-device/kernel pilot (#1490 W0) Extend the workspace with packages/* and move the kernel behind an enforced public API: packages/kernel with nine consumer-earned subpath exports (errors, device, snapshot, contracts, collections, rect, redaction, daemon-error, bounds — the last absorbed from utils as Rect vocabulary). Every kernel import repo-wide becomes the @agent-device/kernel/<sub> specifier; kernel tests move to src/__tests__/kernel/ and exercise the package surface. The root declares the package in devDependencies (workspace:*), tsdown bundles it (noExternal) so the published artifact and its runtime dependency manifest are unchanged. Gate rewiring in the same change, per the W0 brief: - R1 kernel-sink retires (physically subsumed); new R11 package-boundaries guards no-root-back-imports, relative tunnelling past exports maps, undeclared workspace deps, and non-exported subpaths, with runtime resolution pins via import.meta.resolve. - resolveImportEdges and mutation ownership follow workspace specifiers through exports maps, keeping R4 cycle checks, depgraph, and derived test ownership connected across the seam (kernel-errors still owns 495 tests). listSourceFiles includes packages/*/src. - kernel becomes an unranked zone; mutation registry, stryker mutate globs, and the mutation-affected workflow path filter move to packages/kernel/src/errors.ts. - check:affected gains packages/ ownership (manifests fail open); vitest and coverage include packages/*/src; fallow ignores packages/** (its resolver cannot follow workspace specifiers). - The affected-selector CI job installs dependencies: its closure now crosses workspace specifiers, and the R8 relative exception is unsafe for production src files (Node ESM does not realpath, so dual specifier/relative loads would instantiate modules twice). The R8 zero-dep set is pinned empty with that rationale. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FUv7bvbWNryuXgSBuqTtep * fix: address W0 review — mutation sandbox, exports-map resolution, tsc -b Review findings on #1494, all five: 1. contracts-schema-public.test.ts reads the kernel source at its packages/ path (fs access invisible to the codemod and typecheck). 2. Mutation lane: Stryker sandboxes the tree but pnpm's node_modules symlink resolves @agent-device/* back to the real repo, so mutants in the sandbox never load and vitest.related finds no tests. vitest.mutation.config.ts now aliases each EXPORTED specifier to its source (derived from exports maps, never a wildcard), keeping resolution inside the mutated tree. Validated: kernel-errors module runs end to end (dry run 3,984 tests, mutants killed, exit 0). 3. Layering/depgraph resolve workspace specifiers through the exports-derived map (workspaceSpecifierTargets) instead of reconstructing paths, so '.'-facade packages resolve; the positional fallback remains only for map-less fixtures (P0 pin). 4. Per-package project references implemented: packages/kernel is composite (emitDeclarationOnly -> dist-types, gitignored), the root references it, and typecheck becomes tsc -b — probed to catch type errors on both sides under TypeScript 7 native. 5. R11's relative-route exception now requires membership in an actual R8 zero-dep job closure (zeroDepClosureFiles walks entries), not mere scripts/ placement — closing the dual-instantiation bypass. Also from review discussion: daemon-error moves out of the kernel package to src/client/ — its consumers (cli, client facade) rehydrate wire DaemonErrors client-side; the daemon only produces them. Kernel drops to 8 exported subpaths before any of them ship. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FUv7bvbWNryuXgSBuqTtep * refactor: one exports-map reader for mutation alias and ownership Fallow flagged workspaceExportAliases (cognitive 15, CRAP 90). The manifest-reading logic already exists as workspaceSpecifierTargets in scripts/layering/package-boundaries.ts, so both the Stryker sandbox alias table and the mutation ownership walker now consume it instead of carrying near-clones. Behavior unchanged; mutation suite 45/45 and changed-code fallow green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FUv7bvbWNryuXgSBuqTtep * fix: composite kernel without a root references edge FreeRange runs plain `tsc -p tsconfig.json`, and a root `references` entry makes non-build-mode TypeScript demand the referenced project's built declarations (TS6305) — a standing "build first" tax on every plain -p consumer (fr, editors). Keep the per-package composite project and build it in typecheck (`tsc -b packages/kernel` before the root and examples/sdk passes), but drop the root references edge: root consumption resolves through exports to source, identical to runtime and to the bundler. Probed: plain -p green with no prebuilt output; kernel-side type errors still caught by its own build. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FUv7bvbWNryuXgSBuqTtep * fix: R11 uses the layering parser; mutation config is a fallow entry Review blockers on #1494: - R11's private single-quote regex could miss a double-quoted or re-export route into packages/*/src. specifierSites now delegates to the layering model's parseImports (both quote styles, side-effect imports, re-exports, dynamic imports), with direct regressions for each formerly-invisible form. - vitest.mutation.config.ts becomes a declared fallow entry instead of a tolerated unused-file finding: the full-repo audit now reports it reachable (unused files 2 -> 1; the remainder predates this PR). FreeRange clean-checkout evidence: with packages/kernel/dist-types and every *.tsbuildinfo deleted, `pnpm check:freerange` reports 0 findings on this head — the TS6305 topology died with the root references edge in the previous commit; check:freerange has no build precondition. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FUv7bvbWNryuXgSBuqTtep --------- Co-authored-by: Claude <noreply@anthropic.com> |
||
|
devin-ai-integration[bot]
|
7402a40bac |
test: enumerate error-code recovery quizzes in a unit-lane gate (#1445)
* test: enumerate error-code recovery quizzes in a unit-lane gate Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * test: mark recovery quizzes structurally and derive retriability from the enumeration Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Michał Pierzchała <thymikee@gmail.com> Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> |
||
|
Michał Pierzchała
|
2d1d70613f |
feat(bench): renderer-pinned samples, topic-coverage gate, error-recovery quizzes; trim skillgym to agentic checks (#1411)
* feat(bench): renderer-pinned samples, topic-coverage gate, error quizzes; trim skillgym to agentic checks The help conformance bench's quoted CLI output is now sourced from scripts/help-conformance-sample-outputs.mjs, and every sample is rebuilt through the real production renderers (settle output formatters, printHumanError, formatSnapshotText, refMutationAdmissionResponse) by scripts/__tests__/help-conformance-sample-outputs.test.ts — a rendering or message change fails deterministically instead of leaving the bench grading against output the CLI no longer prints. This retires the fabricated recoverable-failure envelope (production never throws a textual settle timeout; that case is replaced by a real DEVICE_IN_USE recovery quiz). Bench cases move to scripts/help-conformance-cases.mjs and are enumerated against the help-topic registry: helpTopicIds() is exported from cli-help, and scripts/__tests__/help-conformance-topic-coverage.test.ts fails when a help topic has neither a bench case nor an explicit waiver. New case families: error-envelope recovery quizzes (device-in-use, stale pinned ref, ambiguous find match, app-not-installed) pinned to real error text, topic coverage for tv/web/react-native/debugging/workflow, and a metamorphic twin of the settled-diff quiz. The skillgym smoke suite shrinks from 119 cases to the 5 that measure what only an agentic runner can show: skill routing plus output interpretation with a proven local CLI help probe (local-cli-help-policy). Its embedded samples now import the same pinned constants, replacing hand-transcribed output that had already drifted from the renderer. Knowledge checks belong to the bench; live fixture behavior belongs to the iOS simulator e2e suite. * review: drive error samples through the real producers; enforce local-help on the routing smoke The DEVICE_IN_USE, AMBIGUOUS_MATCH, and APP_NOT_INSTALLED parity tests no longer hand-author the producer message before rendering: each drives the actual producer — buildDeviceInUseBySessionError (extracted in session-open.ts and called by the handler), buildAmbiguousMatchError (now exported from find.ts), and buildAppNotInstalledError (extracted in app-resolution.ts and thrown by the resolver). Because each factory is exported from its producer file and called by the production path, dropping the production call would make it test-only and fail check:production-exports — the wiring is gate-enforced, not conventional. open-and-snapshot now sets requireLocalCliHelp and allowOnlyLocalCliHelpCommands, so the 'skill plus local help' claim is observed rather than assumed; without them the case can pass on model prior alone. |
||
|
Michał Pierzchała
|
ab913c9720 |
feat: strengthen agent help benchmarks (#1404)
* feat: strengthen agent help benchmarks * fix: harden help benchmark review findings * fix: close help benchmark validation bypasses * fix: make selector scoring quote-insensitive |
||
|
Michał Pierzchała
|
983625fc5d |
feat: fix codex runner, add --override-doc grading, port skillgym quiz cases to help-conformance bench (#1176)
* feat: fix codex runner, add override-doc grading, port skillgym quiz cases The 2026-07-09 evaluation of scripts/help-conformance-bench.mjs found it structurally right but broken for the codex runner (two bugs), thin on coverage (4 cases), sequential, and unable to grade a draft help rewrite without a rebuild. - Fix runCodex: (1) codex exec reads stdin until EOF when not attached to a TTY, and execFile never closes the child's stdin, so every codex call hung until RUN_TIMEOUT_MS with empty output — close stdin right after spawn. (2) `-o outFile` writes the same final JSON that codex also prints to stdout, so concatenating both produced two back-to-back JSON objects that broke every JSON.parse candidate and silently zeroed extractCommands() — prefer the clean -o payload, fall back to stdout only when it's empty. - Add `--override-doc <topicId>=<path>` (repeatable): loads a topic's text from a file instead of shelling out to `node bin/agent-device.mjs help <topic>`, so a draft help rewrite can be A/B graded with zero rebuild. - Port three cases from test/skillgym/suites/agent-device-smoke-suite.ts (settle-diff-is-observation, sample-output-settled-diff-next-target, sample-output-not-settled-needs-observe) as self-contained "next-command quiz" cases, generalizing the scorer to support regex matchers/forbidden patterns alongside the existing named expectations. Fixture output text matches the CURRENT settle rendering in src/commands/interaction/output.ts, including the "unchanged interactive (N):" tail added by #1167/#1172. - Parallelize the runner x case matrix with a concurrency cap (HELP_BENCH_CONCURRENCY, default 4); results still print in the original matrix order. - Extend test/skillgym/README.md's existing pointer to this bench with the new flags. Validated with real LLM calls (both runners, all 7 cases, 14 calls, ~$0.25 total): 13/14 pass; the one fail (claude-haiku-4-5 on dogfood-mode) is a genuine model miss (returned an empty command plan asking for the app name instead of committing to a generic plan), not a bench bug. `--override-doc` demonstrated live: stripping the dogfood doc's evidence-command examples regresses codex:gpt-5.4-mini from 3/3 to 2/3 on the same case, showing the flag both loads and changes grading. * fix: apply live-doc post-processing to --override-doc, fail fast on bad overrides Review findings on the initial version (all reproduced): - HIGH: an override for the --help:first30 doc id skipped the live path's firstLines(text, 30) cap, so a 49-line draft leaked lines 31-49 into the prompt — grading content a live run never shows, on the doc id every case uses. loadDoc now splits source (live shell-out vs override file) from post-processing, and the post-processing applies to both, so an override differs ONLY in where the text comes from. - MEDIUM: an --override-doc topic id no selected case uses was silently ignored (exit 0, real doc graded). Now fails fast listing the valid doc ids for the selection. - LOW: a missing override file threw a raw ENOENT stack trace; expected failures now print one clean Error line. Added --help usage text that documents last-wins semantics for repeated same-topic overrides and the post-processing parity. Guard tests (scripts/__tests__/help-conformance-bench.test.ts, wired into the unit-core vitest project by explicit path): a 49-line fixture whose prompt must keep line 30 and drop line 31, unknown-topic fail-fast with valid ids listed, clean no-stack error for a missing file, and last-wins for repeated overrides. All spawn the script in --dry-run with every required doc overridden, so they need no LLM calls and no built CLI. Live re-validation: a 33-line override of --help:first30 whose lines 31-33 instruct the model to emit a sentinel command; neither claude-haiku-4-5 nor codex:gpt-5.4-mini emitted it (both scored 4/4, matching the live-doc baseline), proving the cap applies end-to-end. |