feat(kernel): static-first reviewer gate (role-reliability §5)

Run operator-configured static-analysis commands (compiler/detekt/formatters) as a deterministic harness step on the producing stage, before the LLM reviewer. A stage declares `static_analysis = [commands]`; after it produces its artifact, each command runs in the workspace root and the results are recorded in a StaticAnalysisCompletedEvent (invariant #9 env observation — recorded live, folded on replay, never re-run). A non-clean command fails the stage retryably with its output fed back verbatim (§2: compiler/test output is ground truth), so the implementer fixes it and only static-clean code ever reaches the reviewer. This makes §5's "reviewer context excludes static findings" mechanical — the findings are resolved upstream, so the reviewer can't waste inference on what detekt catches for free; no output-parsing, no context plumbing. - StaticAnalysisCompletedEvent + StaticAnalysisFinding (core:events) + registration. - StaticAnalysisRunner seam + ProcessStaticAnalysisRunner (whitespace-split argv, stderr merged, timeout→nonzero exit) in core:kernel; wired (nullable) into OrchestratorEngines and the server. Null runner / no workspace root → logged no-op, never blocks the run. - StageConfig.staticAnalysis + `static_analysis` TOML field; runStaticAnalysis folded into a runPostStageGates sequence (produces → grounding → echo → static analysis). - Documented `static_analysis` on the role_pipeline implementer stage (commented; commands are workspace-specific). The reviewer prompt half shipped in 3467826.
2026-06-14 23:51:49 +04:00
parent 0d8df4a130
commit 365eb8a279
13 changed files with 354 additions and 9 deletions
@@ -34,6 +34,7 @@ import com.correx.core.kernel.orchestration.OrchestrationConfig
 import com.correx.core.kernel.orchestration.OrchestrationProjector
 import com.correx.core.kernel.orchestration.OrchestrationRepository
 import com.correx.core.kernel.orchestration.OrchestratorEngines
+import com.correx.core.kernel.orchestration.ProcessStaticAnalysisRunner
 import com.correx.core.kernel.orchestration.OrchestratorRepositories
 import com.correx.core.kernel.orchestration.WorkspaceContext
 import com.correx.core.kernel.orchestration.WorkspaceTools
@@ -277,6 +278,7 @@ fun main() {
        toolCallAssessor = toolCallAssessor,
        workspacePolicy = workspacePolicy,
        workspaceToolRegistryProvider = wsToolRegistryProvider,
+        staticAnalysisRunner = ProcessStaticAnalysisRunner(),
    )
    val decisionJournalRepository = InfrastructureModule.createDecisionJournalRepository(eventStore)
    val defaultOrchestrationConfig = OrchestrationConfig(
@@ -0,0 +1,40 @@
+package com.correx.core.events.events
+
+import com.correx.core.events.types.SessionId
+import com.correx.core.events.types.StageId
+import kotlinx.serialization.SerialName
+import kotlinx.serialization.Serializable
+
+/** The outcome of one operator-configured static-analysis command run against a stage's output. */
+@Serializable
+data class StaticAnalysisFinding(
+    val command: String,
+    val exitCode: Int,
+    /**
+     * Truncated tail of the command's combined stdout+stderr. The full output is fed back to the
+     * model verbatim at runtime but is not retained in the log — the tail carries the error summary
+     * and keeps the event bounded.
+     */
+    val summary: String,
+) {
+    val clean: Boolean get() = exitCode == 0
+}
+
+/**
+ * Environment observation (invariant #9): the static-analysis commands (compiler / detekt /
+ * formatters) a stage ran against its just-produced output, each with its exit code and a truncated
+ * output tail, captured at the moment the stage completed. Replay reads these recorded facts and
+ * never re-runs the commands.
+ *
+ * Realizes role-reliability §5 "static-first ordering": deterministic tools run before the LLM
+ * reviewer. A non-clean finding fails the producing stage retryably (the output is fed back so the
+ * implementer fixes it), so only static-clean output ever reaches the reviewer — its context thereby
+ * excludes anything static tools already catch, with no parsing or context plumbing.
+ */
+@Serializable
+@SerialName("StaticAnalysisCompleted")
+data class StaticAnalysisCompletedEvent(
+    val sessionId: SessionId,
+    val stageId: StageId,
+    val findings: List<StaticAnalysisFinding>,
+) : EventPayload
@@ -10,6 +10,7 @@ import com.correx.core.events.events.ArtifactValidatedEvent
 import com.correx.core.events.events.ArtifactValidatingEvent
 import com.correx.core.events.events.BriefEchoMismatchEvent
 import com.correx.core.events.events.BriefGroundingCheckedEvent
+import com.correx.core.events.events.StaticAnalysisCompletedEvent
 import com.correx.core.events.events.ChatSessionStartedEvent
 import com.correx.core.events.events.ClarificationAnsweredEvent
 import com.correx.core.events.events.ClarificationRequestedEvent
@@ -110,6 +111,7 @@ val eventModule = SerializersModule {
        subclass(RepoKnowledgeRetrievedEvent::class)
        subclass(BriefGroundingCheckedEvent::class)
        subclass(BriefEchoMismatchEvent::class)
+        subclass(StaticAnalysisCompletedEvent::class)
        subclass(RiskAssessedEvent::class)
        subclass(ChatSessionStartedEvent::class)
        subclass(ChatTurnEvent::class)
@@ -0,0 +1,51 @@
+package com.correx.core.events.serialization
+
+import com.correx.core.events.events.EventPayload
+import com.correx.core.events.events.StaticAnalysisCompletedEvent
+import com.correx.core.events.events.StaticAnalysisFinding
+import com.correx.core.events.types.SessionId
+import com.correx.core.events.types.StageId
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertFalse
+import kotlin.test.assertTrue
+
+class StaticAnalysisCompletedEventSerializationTest {
+
+    private val sample = StaticAnalysisCompletedEvent(
+        sessionId = SessionId("sess-1"),
+        stageId = StageId("implementer"),
+        findings = listOf(
+            StaticAnalysisFinding(command = "./gradlew compileKotlin", exitCode = 0, summary = "BUILD SUCCESSFUL"),
+            StaticAnalysisFinding(command = "./gradlew detekt", exitCode = 1, summary = "MagicNumber at Foo.kt:42"),
+        ),
+    )
+
+    @Test
+    fun `round-trips as polymorphic EventPayload`() {
+        val encoded = eventJson.encodeToString(EventPayload.serializer(), sample)
+        assertTrue(encoded.contains("\"type\":\"StaticAnalysisCompleted\""), "SerialName must be present: $encoded")
+        val decoded = eventJson.decodeFromString(EventPayload.serializer(), encoded)
+        assertEquals(sample, decoded)
+    }
+
+    @Test
+    fun `decodes hand-written StaticAnalysisCompleted JSON`() {
+        val json = """
+            {"type":"StaticAnalysisCompleted","sessionId":"s","stageId":"implementer",
+             "findings":[{"command":"./gradlew detekt","exitCode":2,"summary":"3 violations"}]}
+        """.trimIndent()
+        val decoded = eventJson.decodeFromString(EventPayload.serializer(), json)
+        assertTrue(decoded is StaticAnalysisCompletedEvent)
+        val finding = (decoded as StaticAnalysisCompletedEvent).findings.single()
+        assertEquals("./gradlew detekt", finding.command)
+        assertEquals(2, finding.exitCode)
+        assertFalse(finding.clean)
+    }
+
+    @Test
+    fun `clean is derived from exit code zero`() {
+        assertTrue(StaticAnalysisFinding("c", 0, "").clean)
+        assertFalse(StaticAnalysisFinding("c", 1, "").clean)
+    }
+}
@@ -28,4 +28,7 @@ data class OrchestratorEngines(
    val workspacePolicy: WorkspacePolicy? = null,
    val worldProbe: WorldProbe = FileSystemWorldProbe(),
    val workspaceToolRegistryProvider: WorkspaceToolRegistryProvider? = null,
+    // Runs operator-configured static-analysis commands as a harness step (role-reliability §5).
+    // Null = no static-first step; a stage declaring `static_analysis` then no-ops with a warning.
+    val staticAnalysisRunner: StaticAnalysisRunner? = null,
 )
@@ -0,0 +1,55 @@
+package com.correx.core.kernel.orchestration
+
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.TimeoutCancellationException
+import kotlinx.coroutines.async
+import kotlinx.coroutines.withContext
+import kotlinx.coroutines.withTimeout
+import java.nio.file.Path
+
+/**
+ * Process-spawning [StaticAnalysisRunner]: splits each operator-configured command on whitespace
+ * into argv (no shell), runs it in the workspace root with stderr merged into stdout, and returns
+ * its exit code with the combined output. A spawn failure or timeout is reported as a non-zero exit
+ * so the gate treats a hung or unrunnable check as a failure rather than silently passing.
+ *
+ * Commands come from the operator's workflow config (trusted, like prompt files), not from the LLM,
+ * so they bypass the tool-approval pipeline by design. Mirrors the streaming/timeout shape of
+ * ShellTool to avoid pipe-buffer deadlock on large output.
+ */
+class ProcessStaticAnalysisRunner(
+    private val timeoutMs: Long = DEFAULT_TIMEOUT_MS,
+) : StaticAnalysisRunner {
+
+    override suspend fun run(workingDir: Path, command: String): StaticAnalysisRunResult =
+        withContext(Dispatchers.IO) {
+            val argv = command.trim().split(WHITESPACE).filter { it.isNotEmpty() }
+            if (argv.isEmpty()) return@withContext StaticAnalysisRunResult(EXIT_NOT_RUN, "empty command")
+            val process = runCatching {
+                ProcessBuilder(argv).directory(workingDir.toFile()).redirectErrorStream(true).start()
+            }.getOrElse {
+                return@withContext StaticAnalysisRunResult(EXIT_NOT_RUN, "failed to start '$command': ${it.message}")
+            }
+            runCatching {
+                withTimeout(timeoutMs) {
+                    val outputDeferred = async { process.inputStream.bufferedReader().use { it.readText() } }
+                    val exit = process.waitFor()
+                    StaticAnalysisRunResult(exit, outputDeferred.await())
+                }
+            }.getOrElse {
+                process.destroyForcibly()
+                val reason = if (it is TimeoutCancellationException) {
+                    "static analysis '$command' timed out after ${timeoutMs}ms"
+                } else {
+                    it.message ?: "error running '$command'"
+                }
+                StaticAnalysisRunResult(EXIT_NOT_RUN, reason)
+            }
+        }
+
+    private companion object {
+        const val DEFAULT_TIMEOUT_MS = 300_000L
+        const val EXIT_NOT_RUN = -1
+        val WHITESPACE = Regex("\\s+")
+    }
+}
@@ -30,6 +30,8 @@ import com.correx.core.events.events.ArtifactContentStoredEvent
 import com.correx.core.events.events.ArtifactCreatedEvent
 import com.correx.core.events.events.BriefEchoMismatchEvent
 import com.correx.core.events.events.BriefGroundingCheckedEvent
+import com.correx.core.events.events.StaticAnalysisCompletedEvent
+import com.correx.core.events.events.StaticAnalysisFinding
 import com.correx.core.events.events.ContextTruncatedEvent
 import com.correx.core.events.events.ClarificationAnswer
 import com.correx.core.events.events.ClarificationAnsweredEvent
@@ -151,6 +153,11 @@ private const val HTTP_TOO_MANY_REQUESTS = 429
 // Cap on clarification rounds per stage, so a stage that keeps re-asking eventually proceeds.
 private const val MAX_CLARIFICATION_ROUNDS = 3

+// Static-analysis output caps: the tail retained in the recorded event vs. the (larger) slice fed
+// back to the model so it can fix the failure. Tails, because the error summary sits at the end.
+private const val STATIC_ANALYSIS_SUMMARY_CAP = 2_000
+private const val STATIC_ANALYSIS_FEEDBACK_CAP = 6_000
+
@SuppressWarnings(
    "ForbiddenComment",
    "UnusedParameter",
@@ -182,6 +189,7 @@ abstract class SessionOrchestrator(
    private val toolCallAssessor: ToolCallAssessor? = engines.toolCallAssessor
    private val workspacePolicy: WorkspacePolicy? = engines.workspacePolicy
    private val worldProbe: WorldProbe = engines.worldProbe
+    private val staticAnalysisRunner: StaticAnalysisRunner? = engines.staticAnalysisRunner
    private val workspaceToolRegistryProvider: WorkspaceToolRegistryProvider? = engines.workspaceToolRegistryProvider
    private val inferenceRepository: InferenceRepository = repositories.inferenceRepository
    internal val orchestrationRepository: OrchestrationRepository = repositories.orchestrationRepository
@@ -549,15 +557,7 @@ abstract class SessionOrchestrator(
                    is StageExecutionResult.Success -> {
                        emitToolArtifacts(sessionId, stageId, stageConfig)
                        emitLlmArtifacts(sessionId, stageId, stageConfig)
-                        when (val produced = verifyProduces(sessionId, stageId, stageConfig)) {
-                            is StageExecutionResult.Success ->
-                                when (val grounded = groundBriefReferences(sessionId, stageId, stageConfig, effectives)) {
-                                    is StageExecutionResult.Success ->
-                                        checkBriefEcho(sessionId, stageId, stageConfig)
-                                    is StageExecutionResult.Failure -> grounded
-                                }
-                            is StageExecutionResult.Failure -> produced
-                        }
+                        runPostStageGates(sessionId, stageId, stageConfig, effectives)
                    }

                    is StageExecutionResult.Failure -> outcome
@@ -1468,6 +1468,88 @@ abstract class SessionOrchestrator(
        }
    }

+    /**
+     * Post-validation stage gates, run in order; the first failure short-circuits and is returned.
+     * Cheap deterministic checks run first (produces presence, brief grounding, brief echo), then
+     * the heavier process-spawning static-analysis step last, so it only runs once a stage is
+     * otherwise sound.
+     */
+    private suspend fun runPostStageGates(
+        sessionId: SessionId,
+        stageId: StageId,
+        stageConfig: StageConfig,
+        effectives: RunEffectives,
+    ): StageExecutionResult {
+        val gates: List<suspend () -> StageExecutionResult> = listOf(
+            { verifyProduces(sessionId, stageId, stageConfig) },
+            { groundBriefReferences(sessionId, stageId, stageConfig, effectives) },
+            { checkBriefEcho(sessionId, stageId, stageConfig) },
+            { runStaticAnalysis(sessionId, stageId, stageConfig, effectives) },
+        )
+        for (gate in gates) {
+            val result = gate()
+            if (result is StageExecutionResult.Failure) return result
+        }
+        return StageExecutionResult.Success(emptyList())
+    }
+
+    /**
+     * Static-first reviewer gate (role-reliability §5): for a stage that declares `static_analysis`
+     * commands, run them (compiler / detekt / formatters) against its just-produced output in the
+     * workspace root. Every result is recorded in a [StaticAnalysisCompletedEvent] — an environment
+     * observation captured at run time (invariant #9), so replay reads it back and never re-runs.
+     *
+     * A non-clean command fails the stage retryably with its output fed back verbatim (§2: compiler/
+     * test output is ground truth, not LLM noise), so the producing stage fixes the issue and only
+     * static-clean output ever reaches the downstream LLM reviewer — its context thereby excludes
+     * anything static tools already catch, with no parsing or context plumbing.
+     */
+    private suspend fun runStaticAnalysis(
+        sessionId: SessionId,
+        stageId: StageId,
+        stageConfig: StageConfig,
+        effectives: RunEffectives,
+    ): StageExecutionResult {
+        if (stageConfig.staticAnalysis.isEmpty()) return StageExecutionResult.Success(emptyList())
+        val runner = staticAnalysisRunner
+        val workspaceRoot = effectives.policy?.workspaceRoot
+        if (runner == null || workspaceRoot == null) {
+            log.warn(
+                "[Orchestrator] stage declares static_analysis but no {} is wired — skipping " +
+                    "session={} stage={}",
+                if (runner == null) "runner" else "workspace root",
+                sessionId.value, stageId.value,
+            )
+            return StageExecutionResult.Success(emptyList())
+        }
+
+        val results = stageConfig.staticAnalysis.map { command ->
+            val run = runner.run(workspaceRoot, command)
+            StaticAnalysisFinding(
+                command = command,
+                exitCode = run.exitCode,
+                summary = run.output.takeLast(STATIC_ANALYSIS_SUMMARY_CAP),
+            ) to run.output
+        }
+        emit(sessionId, StaticAnalysisCompletedEvent(sessionId, stageId, results.map { it.first }))
+
+        val failures = results.filterNot { it.first.clean }
+        if (failures.isEmpty()) return StageExecutionResult.Success(emptyList())
+
+        log.warn(
+            "[Orchestrator] static analysis failed session={} stage={} commands={}",
+            sessionId.value, stageId.value,
+            failures.joinToString(", ") { it.first.command },
+        )
+        val detail = failures.joinToString("\n\n") { (finding, output) ->
+            "$ ${finding.command}  (exit ${finding.exitCode})\n${output.takeLast(STATIC_ANALYSIS_FEEDBACK_CAP)}"
+        }
+        return StageExecutionResult.Failure(
+            "stage ${stageId.value} did not pass static analysis. Fix these before review:\n\n$detail",
+            retryable = true,
+        )
+    }
+
    private suspend fun emitProcessResultEvents(
        sessionId: SessionId,
        stageId: StageId,
@@ -0,0 +1,22 @@
+package com.correx.core.kernel.orchestration
+
+import java.nio.file.Path
+
+/** Result of running one static-analysis command: its exit code and combined stdout+stderr. */
+data class StaticAnalysisRunResult(
+    val exitCode: Int,
+    val output: String,
+)
+
+/**
+ * Seam for running operator-configured static-analysis commands (compiler / detekt / formatters)
+ * as a deterministic harness step (role-reliability §5 static-first ordering). Implemented by a
+ * process-spawning adapter in infrastructure and injected, so the deterministic core never spawns
+ * processes itself and the gate stays unit-testable with a fake.
+ *
+ * The orchestrator records each result as a [com.correx.core.events.events.StaticAnalysisCompletedEvent]
+ * (invariant #9 environment observation) — replay reads the recorded facts and never re-runs the command.
+ */
+fun interface StaticAnalysisRunner {
+    suspend fun run(workingDir: Path, command: String): StaticAnalysisRunResult
+}
@@ -0,0 +1,54 @@
+package com.correx.core.kernel.orchestration
+
+import kotlinx.coroutines.runBlocking
+import org.junit.jupiter.api.Assertions.assertEquals
+import org.junit.jupiter.api.Assertions.assertTrue
+import org.junit.jupiter.api.Test
+import java.nio.file.Path
+import kotlin.io.path.createTempDirectory
+import kotlin.io.path.writeText
+
+class ProcessStaticAnalysisRunnerTest {
+
+    private val runner = ProcessStaticAnalysisRunner()
+
+    private fun script(dir: Path, name: String, body: String): String {
+        val file = dir.resolve(name)
+        file.writeText("#!/bin/sh\n$body\n")
+        file.toFile().setExecutable(true)
+        return "./$name"
+    }
+
+    @Test
+    fun `clean command reports exit zero and captures output`() = runBlocking {
+        val dir = createTempDirectory("sa-ok")
+        val cmd = script(dir, "ok.sh", "echo all-good")
+        val result = runner.run(dir, cmd)
+        assertEquals(0, result.exitCode)
+        assertTrue(result.output.contains("all-good"), "output: ${result.output}")
+    }
+
+    @Test
+    fun `failing command reports nonzero exit and captures stderr`() = runBlocking {
+        val dir = createTempDirectory("sa-fail")
+        // stderr is merged into the captured output (redirectErrorStream).
+        val cmd = script(dir, "bad.sh", "echo boom 1>&2\nexit 2")
+        val result = runner.run(dir, cmd)
+        assertEquals(2, result.exitCode)
+        assertTrue(result.output.contains("boom"), "output: ${result.output}")
+    }
+
+    @Test
+    fun `unrunnable command is a failure, not a silent pass`() = runBlocking {
+        val dir = createTempDirectory("sa-missing")
+        val result = runner.run(dir, "./does-not-exist-xyz")
+        assertTrue(result.exitCode != 0, "exit: ${result.exitCode}")
+    }
+
+    @Test
+    fun `blank command is a failure`() = runBlocking {
+        val dir = createTempDirectory("sa-blank")
+        val result = runner.run(dir, "   ")
+        assertTrue(result.exitCode != 0, "exit: ${result.exitCode}")
+    }
+}
@@ -23,5 +23,10 @@ data class StageConfig(
    // the declared set is scope creep — the dominant local-implementer failure — and is
    // blocked by the plane-2 ManifestContainmentRule.
    val writeManifest: List<String> = emptyList(),
+    // Operator-configured static-analysis commands (compiler/detekt/formatters) run as a
+    // deterministic harness step after this stage produces its output (role-reliability §5).
+    // A non-clean command fails the stage retryably with its output fed back, so only
+    // static-clean output reaches the downstream reviewer. Empty = no static-first step.
+    val staticAnalysis: List<String> = emptyList(),
    val metadata: Map<String, String> = emptyMap(),
 )
@@ -75,12 +75,20 @@ max_retries = 2
 #    stage may write — a FILE_WRITE outside it is blocked as scope creep. Left open
 #    here because the targets are task-specific; a task-scoped workflow would set e.g.
 #      writes = ["core/sessions/**", "testing/sessions/**"]
+#    Optional: `static_analysis` runs deterministic tools (compiler/detekt/formatters)
+#    against the patch BEFORE the reviewer (role-reliability §5). A non-clean command
+#    fails this stage retryably with its output fed back verbatim, so only static-clean
+#    code reaches the reviewer — the reviewer never spends inference on what tools catch
+#    for free. Commands are workspace-specific, so left commented; a Kotlin/Gradle repo
+#    would set e.g.
+#      static_analysis = ["./gradlew compileKotlin -q", "./gradlew detekt -q"]
 [[stages]]
 id = "implementer"
 prompt = "prompts/implementer.md"
 needs = ["impl_plan"]
 produces = [{ name = "patch", kind = "file_written" }]
 allowed_tools = ["file_read", "file_write", "file_edit", "ShellTool"]
+# static_analysis = ["./gradlew compileKotlin -q", "./gradlew detekt -q"]
 token_budget = 32768
 max_retries = 3

@@ -42,6 +42,7 @@ private data class StageSection(
    val needs: List<String> = emptyList(),
    @param:JsonProperty("allowed_tools") val allowedTools: List<String> = emptyList(),
    val writes: List<String> = emptyList(),
+    @param:JsonProperty("static_analysis") val staticAnalysis: List<String> = emptyList(),
    @param:JsonProperty("token_budget") val tokenBudget: Int = 4096,
    @param:JsonProperty("max_retries") val maxRetries: Int = 3,
    @param:JsonProperty("requires_approval") val requiresApproval: Boolean = false,
@@ -106,6 +107,7 @@ class TomlWorkflowLoader(
                needs = s.needs.map { ArtifactId(it) }.toSet(),
                allowedTools = s.allowedTools.toSet(),
                writeManifest = s.writes,
+                staticAnalysis = s.staticAnalysis,
                tokenBudget = s.tokenBudget,
                // Propagate the declared token budget to the inference completion cap.
                // Without this the StageConfig default (maxTokens=2048) is used, truncating
@@ -156,6 +156,25 @@ class TomlWorkflowLoaderTest {
        assertEquals(emptyList(), collectStage.writeManifest)
    }

+    @Test
+    fun `static_analysis maps to the stage static-analysis commands`() {
+        val toml = validToml.replace(
+            "allowed_tools = [\"ShellTool\"]",
+            "allowed_tools = [\"ShellTool\"]\nstatic_analysis = [\"./gradlew compileKotlin\", \"./gradlew detekt\"]",
+        )
+        val path = Files.createTempFile("workflow", ".toml").also { it.writeText(toml) }
+        val graph = loader.load(path)
+        val collectStage = graph.stages[graph.start]!!
+        assertEquals(listOf("./gradlew compileKotlin", "./gradlew detekt"), collectStage.staticAnalysis)
+    }
+
+    @Test
+    fun `static_analysis absent means no commands`() {
+        val path = Files.createTempFile("workflow", ".toml").also { it.writeText(validToml) }
+        val graph = loader.load(path)
+        assertEquals(emptyList(), graph.stages[graph.start]!!.staticAnalysis)
+    }
+
    @Test
    fun `ground_references true maps to metadata key`() {
        val toml = validToml.replace(