MCPcopy Create free account

hub / github.com/JuliusBrussee/caveman / functions

Functions12,853 in github.com/JuliusBrussee/caveman

FunctionTestParseUsage_TruncatedBedrockStreamsStampRawUsageIncomplete
Both truncation sites in this file zero OutputTokensReported themselves, and neither can rely on providers.ParseUsageBytes to stamp the raw blob: merg
proxy/providers/bedrock/usage_test.go:525
FunctionTestPartialEnvPairFailsClosed
(t *testing.T)
shared/platform/awscreds/awscreds_test.go:87
FunctionTestPayloadAllocationsDoNotScaleWithFiringRules
TestPayloadAllocationsDoNotScaleWithFiringRules guards the fix for the per-firing-rule full-body copies. Redaction runs inline in the proxy handler un
shared/platform/redact/payload_test.go:360
FunctionTestPayloadAppliesOrgRules
(t *testing.T)
shared/platform/redact/payload_test.go:126
FunctionTestPayloadBudgetRefusalReportsPayloadBudget
(t *testing.T)
engine/ccr/store_full_test.go:277
FunctionTestPayloadBuiltinGroupsExistInTheirPatterns
A group index past a pattern's capture count indexes apply()'s loc slice out of range — the rule works until a body finally contains the secret it exi
shared/platform/redact/payload_test.go:719
FunctionTestPayloadCatchesEveryBuiltin
(t *testing.T)
shared/platform/redact/payload_test.go:20
FunctionTestPayloadDeterministicAcrossRuleOrder
(t *testing.T)
shared/platform/redact/payload_test.go:158
FunctionTestPayloadDoesNotMutateInput
(t *testing.T)
shared/platform/redact/payload_test.go:441
FunctionTestPayloadDoesNotOverRedactOrdinaryNumbers
(t *testing.T)
shared/platform/redact/payload_test.go:86
FunctionTestPayloadDoesNotRedactNumericFieldsNamedLikeCredentials
The same family on the redaction side: a numeric field named like a credential must survive, or every usage record in the corpus loses its counters.
shared/platform/redact/payload_test.go:731
FunctionTestPayloadDoesNotShredCodeIdentifiers
Allowing a name fragment between keyword and delimiter made every identifier reference whose NAME contains a keyword look like a credential assignment
shared/platform/redact/payload_test.go:749
FunctionTestPayloadExhaustionReportsPayloadBudget
(t *testing.T)
engine/ccr/store_full_test.go:180
FunctionTestPayloadFailsClosedOnBadRules
(t *testing.T)
shared/platform/redact/payload_test.go:202
FunctionTestPayloadLabelledGovernmentIDs
(t *testing.T)
shared/platform/redact/payload_test.go:452
FunctionTestPayloadLabelledSSNGapWindow
The labelled-SSN gap used to be two `{0,12}` windows with an optional newline between them — 24 non-newline characters of reach, far enough to weld a
shared/platform/redact/payload_test.go:694
FunctionTestPayloadLargeAdversarialBodyAtDefaultCaptureCeiling
TestPayloadLargeAdversarialBodyAtDefaultCaptureCeiling runs at capture's default CAVE_DATA_COLLECTION_MAX_BODY_BYTES setting of 4 MiB. Operators may r
shared/platform/redact/payload_test.go:331
FunctionTestPayloadNeedleIntroducersAreMarked
TestPayloadNeedleIntroducersAreMarked pins the bookkeeping the optimisation depends on, so a future placeholder rename cannot silently unmark one.
shared/platform/redact/payload_test.go:561
FunctionTestPayloadNilAndEmptyBody
(t *testing.T)
shared/platform/redact/payload_test.go:428
FunctionTestPayloadOrgReplacementIsLiteral
(t *testing.T)
shared/platform/redact/payload_test.go:147
FunctionTestPayloadOrgRuleCannotDisableBuiltin
(t *testing.T)
shared/platform/redact/payload_test.go:99
FunctionTestPayloadOverlappingOrgRulesResolveByPriority
(t *testing.T)
shared/platform/redact/payload_test.go:188
FunctionTestPayloadPrescreenPremisesHold
TestPayloadPrescreenPremisesHold pins the two structural premises that make replIntroducesNeedle sound. TestPayloadNeedleIntroducersAreMarked cannot d
shared/platform/redact/payload_test.go:585
FunctionTestPayloadPreservesSurroundingShape
(t *testing.T)
shared/platform/redact/payload_test.go:70
FunctionTestPayloadRedactsVendorPrefixedTokens
A vendor-published prefix is direct evidence, so it gets its own rule rather than relying on the assignment shape or on entropy. None of these was red
shared/platform/redact/payload_test.go:655
FunctionTestPayloadRedactsWeldedKeyAssignments
--- Boundary-screen bypasses closed 2026-07-31 ----------------------------- The key-assignment pattern required the keyword to be ADJACENT to the del
shared/platform/redact/payload_test.go:627
FunctionTestPayloadRejectsOversizeBody
(t *testing.T)
shared/platform/redact/payload_test.go:229
FunctionTestPayloadReportNeverEchoesMatches
(t *testing.T)
shared/platform/redact/payload_test.go:243
FunctionTestPayloadRuleSetHashTracksRules
(t *testing.T)
shared/platform/redact/payload_test.go:286
FunctionTestPayloadStalePrescreenCannotDropAMatch
TestPayloadStalePrescreenCannotDropAMatch is the guard on the optimisation in finding 3. The prescreen is derived once and re-derived only after a rul
shared/platform/redact/payload_test.go:525
FunctionTestPendingSetsAreKeptPerRequestShape
A side request starting with the same message as a compacted conversation keeps its own pending set: the continuation still takes the session over.
proxy/internal/gateway/route_effort_test.go:1648
FunctionTestPerMessageBetaHeader
(t *testing.T)
proxy/internal/gateway/route_effort_test.go:518
FunctionTestPerMessageEffortCountsTheAgentsOwn
The agent's own per-message effort counts as in force, and a new mark goes after it, never before.
proxy/internal/gateway/route_effort_test.go:316
FunctionTestPerMessageEffortFreshConversationSetsTheTopLevel
A fresh conversation needs no mark: its top-level effort is the routed one, and stays fixed for the session even though the agent keeps sending its ow
proxy/internal/gateway/route_effort_test.go:268
FunctionTestPerMessageEffortHealsARefusedMark
Marks refused: one retry with top-level effort only (no marks, no beta); once that is served the session's latch is on and the next ask says so.
proxy/internal/gateway/route_effort_test.go:339
FunctionTestPerMessageEffortHealsTheThinkingBinding
A broken thinking binding: one retry asking the provider to drop the unbound thinking blocks, marks kept.
proxy/internal/gateway/route_effort_test.go:367
FunctionTestPerMessageEffortLatchesOnRefusalWithoutMarks
Refusal wording on a request without marks latches the session at once; an unrelated 400 neither heals nor latches. Neither retries more than once.
proxy/internal/gateway/route_effort_test.go:410
FunctionTestPerMessageEffortMarksReplayByteIdentical
(t *testing.T)
proxy/internal/gateway/route_effort_test.go:192
FunctionTestPerMessageEffortSideRequestsAndChildren
Compaction and side requests get the session's marks back but never a new one; a forked child resending the parent's history inherits its marks.
proxy/internal/gateway/route_effort_test.go:284
FunctionTestPerMessageHealReadsACompressedRefusal
A compressed refusal is read decoded: the heal and the latch still fire.
proxy/internal/gateway/route_effort_test.go:597
FunctionTestPerMessageMarksComeBackOnEveryAnswer
A session's marks come back whatever the answer: a Cloud failure, routing off, or a top-level answer without an effort.
proxy/internal/gateway/route_effort_test.go:714
FunctionTestPerMessageMarksSurviveMovingCacheControl
Agents and the breakpoint planner move cache_control every request; the anchors ignore it, so the marks stay put.
proxy/internal/gateway/route_effort_test.go:560
FunctionTestPerRequestAsksEveryTime
A compaction or side request is asked every time and carries no ask text; a turn's tool loop asks once.
proxy/internal/cloudlink/route_effort_test.go:197
FunctionTestPersistedChoicesSurviveARevisionChange
Scope identity no longer includes the policy revision, so an upgrade that changes the revision keeps every persisted choice and its exact bytes.
proxy/internal/middleware/protocol_v11_test.go:427
FunctionTestPersistentRuntimePrunesAbandonedCorrelationEntries
(t *testing.T)
proxy/internal/nativeruntime/runtime_test.go:1452
FunctionTestPersistentStoreBudgetRefusesNewRecoveryAndPreservesExisting
(t *testing.T)
engine/ccr/store_test.go:350
FunctionTestPiChatGPTMalformedZstdFailsOpenByteExact
(t *testing.T)
proxy/internal/gateway/chatgpt_test.go:198
FunctionTestPiChatGPTOAuthZstdRoutesAndCompresses
(t *testing.T)
proxy/internal/gateway/chatgpt_test.go:97
FunctionTestPiChatGPTSubscriptionDetection
(t *testing.T)
proxy/internal/gateway/chatgpt_test.go:60
FunctionTestPiChatGPTZstdTransformed4xxRetriesExactOriginalWire
(t *testing.T)
proxy/internal/gateway/chatgpt_test.go:153
FunctionTestPiRootsHonorRelocationVariables
Pi reads PI_CODING_AGENT_DIR as its agent directory and PI_CODING_AGENT_SESSION_DIR as an outright session-storage override (both tilde-expanded via e
proxy/internal/store/agent_roots_test.go:108
FunctionTestPiSessionSourceHonorsRelocatedSessionsDirs
TestPiSessionSourceHonorsRelocatedSessionsDirs proves the scanner follows Pi's relocation variables instead of only <CAVEMAN_PI_ROOT>/agent/sessions.
proxy/internal/store/source_pi_test.go:154
FunctionTestPiSessionSourceReadsUsageAndCompletesToolCalls
(t *testing.T)
proxy/internal/store/source_pi_test.go:80
FunctionTestPiTaskSpawnsCountsOnlySpawnTools
(t *testing.T)
proxy/internal/store/source_pi_test.go:259
FunctionTestPiUsageReadsOneSpellingNeverSums
(t *testing.T)
proxy/internal/store/source_pi_test.go:272
FunctionTestPingNeverSplitsAPacedRelayedEvent
(t *testing.T)
proxy/internal/translate/relay_framing_test.go:87
FunctionTestPingNeverSplitsARelayedEvent
(t *testing.T)
proxy/internal/translate/relay_framing_test.go:57
FunctionTestPinnedFailureRetriesTheEntryUnpinned
A pinned provider that fails: once more on the same OpenRouter entry without the pin, so the request stays on the pool model.
proxy/internal/gateway/route_cache_test.go:327
FunctionTestPinnedTransportErrorIsNeverResent
A transport error on a pinned send may come after OpenRouter read the request: never sent to OpenRouter again; the pin goes and the asked model runs.
proxy/internal/gateway/route_cache_test.go:349
FunctionTestPitchSixGeometry
TestPitchSixGeometry pins the reference pitch-6 page geometry: page height is 2*pad + (rows-1)*pitch + glyphH (descenders/ascenders both fit), and a h
engine/pixel/render_density_test.go:69
FunctionTestPixelModeAnthropicImagesAndStoresOriginal
(t *testing.T)
proxy/internal/gateway/pixel_test.go:69
FunctionTestPixelModeDisallowedModelPassesThrough
(t *testing.T)
proxy/internal/gateway/pixel_test.go:118
FunctionTestPixelModeGeminiImagesAllowedModel
(t *testing.T)
proxy/internal/gateway/pixel_test.go:284
FunctionTestPixelModeKeepsPreviousTurnStable
TestPixelModeKeepsPreviousTurnStable: pixel renders the live message, which the provider then caches as images. The next turn carries that message bel
proxy/internal/gateway/pixel_test.go:478
FunctionTestPixelModeMalformedJSONPassesThrough
(t *testing.T)
proxy/internal/gateway/pixel_test.go:259
FunctionTestPixelModeOpenAIImagesGPT56
(t *testing.T)
proxy/internal/gateway/pixel_test.go:310
FunctionTestPixelModeOpenAIKeepsPreviousTurnStable
TestPixelModeOpenAIKeepsPreviousTurnStable: OpenAI caches every long prompt implicitly, so the user message rendered on turn 1 must come back as the s
proxy/internal/gateway/pixel_test.go:511
FunctionTestPixelModeOpenAIResponsesStringInput
(t *testing.T)
proxy/internal/gateway/pixel_test.go:338
FunctionTestPixelModeOpenAIResponsesUserAndToolOutput
(t *testing.T)
proxy/internal/gateway/pixel_test.go:359
FunctionTestPixelModeRecordPassesThrough
(t *testing.T)
proxy/internal/gateway/pixel_test.go:149
FunctionTestPixelModeStoreFailurePassesThrough
(t *testing.T)
proxy/internal/gateway/pixel_test.go:231
FunctionTestPixelModeStreamingAnthropicMCPAndPlain
(t *testing.T)
proxy/internal/gateway/pixel_test.go:383
FunctionTestPixelModeSubscriptionDefaultPassesThrough
(t *testing.T)
proxy/internal/gateway/pixel_test.go:195
FunctionTestPixelRenderDefaultsToReflowWithOptOut
(t *testing.T)
engine/cmd/caveman-engine/main_test.go:103
FunctionTestPixelRenderDensityLabelMatchesPixels
The load-bearing invariant: for EVERY flag combination the reported density + layers must match the pixels actually drawn. Walk the matrix {plain, --d
engine/cmd/caveman-engine/pixel_density_test.go:167
FunctionTestPixelRenderDensityPageRatios
End-to-end: the same document rendered at three levels must produce the expected page-count ordering, and each report's resolved `density` field must
engine/cmd/caveman-engine/pixel_density_test.go:65
FunctionTestPixelRenderRejectsUnknownDensity
An invalid --density is a LOUD usage error on the dev surface (the engine's env path stays fail-closed-to-conservative; the flag does not).
engine/cmd/caveman-engine/pixel_density_test.go:245
FunctionTestPixelRenderReportsSummary
The summary line is what `caveman convert` gates on: it must count both sides (text vs. rendered image estimate) and total the per-page estimates exac
engine/cmd/caveman-engine/main_test.go:64
FunctionTestPixelUsageDocumentsNoReflow
(t *testing.T)
engine/cmd/caveman-engine/main_test.go:122
FunctionTestPlainTextPassesThrough
(t *testing.T)
engine/engine_test.go:216
FunctionTestPlaintextOriginalsRefusedOnceAKeyIsConfigured
S10: once a key is configured, plaintext originals are refused unless the operator is migrating, and each refusal is counted.
proxy/internal/middleware/hardening_test.go:335
FunctionTestPlanGptCollapseGatesAndClosedBoundaries
(t *testing.T)
engine/pixel/history_openai_test.go:57
FunctionTestPlanGptCollapsePinsLatestUserRequest
(t *testing.T)
engine/pixel/history_openai_test.go:160
FunctionTestPlanGptCollapseTokenFloorAndAppendOnlyImages
(t *testing.T)
engine/pixel/history_openai_test.go:117
FunctionTestPlanRowsLastTheReplayWindow
A plan row exists for idempotent replay only, so it lasts the replay window, not the whole retention.
proxy/internal/middleware/hardening_test.go:828
FunctionTestPlumbingDispatchWithMockEngine
(t *testing.T)
mcp/server_test.go:178
FunctionTestPoolAnswerHarnessEntryAppliesLikeModels
(t *testing.T)
proxy/internal/cloudlink/pool_test.go:117
FunctionTestPoolAnswerLocalLoginBecomesATarget
(t *testing.T)
proxy/internal/cloudlink/pool_test.go:92
FunctionTestPoolAnswerViaCloudGoesToTheGateway
(t *testing.T)
proxy/internal/cloudlink/pool_test.go:131
FunctionTestPoolAttemptIsOnTheRow
The row says a pool attempt happened: on a fallback, the entry and why it failed; on a served answer, the entry and the host's own answer id.
proxy/internal/gateway/route_pool_test.go:627
FunctionTestPoolClaudeCodeOnTheChatGPTLogin
A Claude Code request runs on the ChatGPT login (Responses only) and reads an Anthropic stream back; a refusal before the first byte falls back.
proxy/internal/gateway/route_pool_test.go:365
FunctionTestPoolCloudFailureBeforeContentFallsBack
A gateway 2xx that fails before content still falls back.
proxy/internal/gateway/route_pool_test.go:474
FunctionTestPoolCloudTargetCarriesTheAnsweredEffort
(t *testing.T)
proxy/internal/gateway/route_pool_test.go:460
FunctionTestPoolCloudTargetSendsTheCallersGrammarToTheGateway
(t *testing.T)
proxy/internal/gateway/route_pool_test.go:287
FunctionTestPoolConstructorsReturnParseErrorsWithoutDialing
(t *testing.T)
shared/platform/postgresconfig/postgresconfig_test.go:220
FunctionTestPoolCutStreamNeverEndsAsACleanEOF
(t *testing.T)
proxy/internal/gateway/route_pool_test.go:343
FunctionTestPoolFailureAfterHeadersBeforeContentFallsBack
A 2xx that fails before any content still falls back to the asked model.
proxy/internal/gateway/route_pool_test.go:422
FunctionTestPoolFailureBeforeTheFirstByteFallsBackToTheAskedModel
(t *testing.T)
proxy/internal/gateway/route_pool_test.go:250
FunctionTestPoolFallbackFitsTheEffortToTheAskedModel
A failed target falls back at the answered effort fitted to the asked model: the word was chosen for the target, and minimal to Claude or max to OpenA
proxy/internal/gateway/route_pool_test.go:506
FunctionTestPoolHostGetsItsOwnReasoningBackOnTheNextTurn
The pool host gets its own reasoning back on the next turn: the harness path's cleaning never touches what the pool host is sent.
proxy/internal/gateway/route_pool_test.go:402
← previousnext →10,101–10,200 of 12,853, ranked by callers