@miraland-labs/conduit-bridge 0.16.140 → 0.16.142

package/dist/attempt-worktree.js

@@ -4,7 +4,7 @@
  */
 import { access, appendFile, mkdir, readFile, rename, rm, writeFile } from "node:fs/promises";
 import { constants } from "node:fs";
-import { join, resolve } from "node:path";
+import { basename, join, resolve } from "node:path";
 import { execFile } from "node:child_process";
 import { promisify } from "node:util";
 import { headCommitOrNull } from "./git.js";
@@ -106,6 +106,16 @@ export async function removeAttemptWorktree(sourceWorkspace, worktreePath) {
             maxBuffer: 1_000_000,
         }).catch(() => undefined);
     }
+    // The attempt branch goes with its worktree: left behind, every attempt added one local branch to
+    // the source checkout for ever. Only the branch Bridge created for this path, and only the local
+    // ref: a delivered branch was pushed before the release and lives on in the forge.
+    const attemptId = basename(worktreePath);
+    if (resolve(worktreePath) === resolve(attemptWorktreePath(sourceWorkspace, attemptId))) {
+        await execFileAsync("git", ["-C", sourceWorkspace, "branch", "-D", attemptBranchName(attemptId)], {
+            timeout: 30_000,
+            maxBuffer: 1_000_000,
+        }).catch(() => undefined);
+    }
     return true;
 }
 /**

package/dist/driver.js

@@ -739,7 +739,9 @@ export const claudeCodeDriver = {
             };
         }
         const maxTurns = resolveMaxTurns(input.maxTurns);
-        const args = ["-p", input.prompt, "--output-format", "json", "--max-turns", String(maxTurns)];
+        // The prompt goes on stdin (`claude -p` reads it there when no prompt argument is given): Linux
+        // caps one argument at 128 KiB (MAX_ARG_STRLEN) and an assignment contract exceeds that.
+        const args = ["-p", "--output-format", "json", "--max-turns", String(maxTurns)];
         args.push(...modelArgs(claudeCodeDriver, input.model));
         if (input.resumeSessionId)
             args.push("--resume", input.resumeSessionId);
@@ -747,7 +749,7 @@ export const claudeCodeDriver = {
         args.push("--disallowedTools", projected.disallowedTools.join(","));
         if (projected.acceptEdits)
             args.push("--permission-mode", "acceptEdits");
-        const { code, stdout, stderr } = await execute(driverExecutable(claudeCodeDriver.name, input.executable), args, input.workspace, agentTurnTimeoutMs(input), fuelSource === "conduit" ? input.fuel : undefined, fuelSource, undefined, input.signal, claudeCodeFuelEnv(fuelSource, input.model));
+        const { code, stdout, stderr } = await execute(driverExecutable(claudeCodeDriver.name, input.executable), args, input.workspace, agentTurnTimeoutMs(input), fuelSource === "conduit" ? input.fuel : undefined, fuelSource, input.prompt, input.signal, claudeCodeFuelEnv(fuelSource, input.model));
         let message = null;
         try {
             message = JSON.parse(stdout);
@@ -1229,9 +1231,9 @@ export const openCodeDriver = {
         if (!agent) {
             return { status: "failed", resultText: null, sessionId: null, error: "No Bridge-mapped OpenCode agent for active grants; refusing to start agent" };
         }
-        const args = openCodeRunArgs(input, agent);
-        args.push(input.prompt);
-        const { code, stdout, stderr } = await withOpenCodePermissions(input.workspace, () => execute(driverExecutable(openCodeDriver.name, input.executable), args, input.workspace, agentTurnTimeoutMs(input), fuelSource === "conduit" ? input.fuel : undefined, fuelSource, undefined, input.signal));
+        // The prompt goes on stdin (`opencode run` reads a piped stdin as the message), as it does for
+        // codex: Linux caps one argument at 128 KiB (MAX_ARG_STRLEN).
+        const { code, stdout, stderr } = await withOpenCodePermissions(input.workspace, () => execute(driverExecutable(openCodeDriver.name, input.executable), openCodeRunArgs(input, agent), input.workspace, agentTurnTimeoutMs(input), fuelSource === "conduit" ? input.fuel : undefined, fuelSource, input.prompt, input.signal));
         const parsed = parseOpenCodeOutput(stdout);
         if (code !== 0 || parsed.isError) {
             return { status: "failed", resultText: parsed.resultText, sessionId: parsed.sessionId, signal: agentExitSignal({ exit_code: code, stderr, stdout, model: input.model }), error: boundedTail(stderr || parsed.resultText || `opencode exited with code ${code}`, 20_000) };
@@ -1313,9 +1315,9 @@ export const kiroDriver = {
                 return { status: "failed", resultText: null, sessionId: null, error: "kiro has no login (set KIRO_API_KEY or run `kiro-cli login`)" };
             }
         }
-        const args = kiroChatArgs(input, trusted);
-        args.push(input.prompt);
-        const { code, stdout, stderr } = await execute(executable, args, input.workspace, agentTurnTimeoutMs(input), undefined, "local", undefined, input.signal);
+        // The prompt goes on stdin (`kiro-cli chat` takes a piped stdin as its input), as it does for
+        // codex: Linux caps one argument at 128 KiB (MAX_ARG_STRLEN).
+        const { code, stdout, stderr } = await execute(executable, kiroChatArgs(input, trusted), input.workspace, agentTurnTimeoutMs(input), undefined, "local", input.prompt, input.signal);
         if (code !== 0) {
             return { status: "failed", resultText: stdout || null, sessionId: null, signal: agentExitSignal({ exit_code: code, stderr, stdout, model: input.model }), error: boundedTail(stderr || stdout || `kiro-cli exited with code ${code}`, 20_000) };
         }
@@ -1638,7 +1640,8 @@ export const antigravityDriver = {
 export function piRunArgs(input) {
     const args = ["--mode", "json", "-p", "--no-session", "--tools", input.tools];
     args.push(...modelArgs(piDriver, input.model));
-    args.push(input.prompt);
+    // The prompt goes on stdin (pi reads a piped stdin as its initial message), as it does for codex:
+    // Linux caps one argument at 128 KiB (MAX_ARG_STRLEN).
     return args;
 }
 /**
@@ -1703,7 +1706,7 @@ export const piDriver = {
         if (version.code !== 0 || !(version.stdout || version.stderr).trim()) {
             return { status: "failed", resultText: null, sessionId: null, error: "pi preflight could not verify the installed CLI version" };
         }
-        const { code, stdout, stderr } = await execute(executable, piRunArgs({ prompt: input.prompt, tools: projected.tools, model: input.model }), input.workspace, agentTurnTimeoutMs(input), undefined, "local", undefined, input.signal);
+        const { code, stdout, stderr } = await execute(executable, piRunArgs({ tools: projected.tools, model: input.model }), input.workspace, agentTurnTimeoutMs(input), undefined, "local", input.prompt, input.signal);
         const parsed = parsePiJsonl(stdout);
         if (code !== 0) {
             return {
@@ -1894,12 +1897,14 @@ const MAX_AGENT_OUTPUT = 8 * 1024 * 1024;
 export const STDIO_DRAIN_MS = 2_000;
 /** How long a stopped child may take to die before the run reports what it already has. */
 const CHILD_STOP_GRACE_MS = 20_000;
-export function execute(executable, args, cwd, timeoutMs, fuel, fuelSource = "conduit", stdinText, signal, extraEnv) {
+export function execute(executable, args, cwd, timeoutMs, fuel, fuelSource = "conduit", stdinText, signal, extraEnv, 
+/** The whole environment, for a run that must see less than an agent does (a probe instrument). */
+env) {
     return new Promise((resolve, reject) => {
         const child = spawn(executable, args, {
             cwd,
             stdio: [stdinText !== undefined ? "pipe" : "ignore", "pipe", "pipe"],
-            env: boundedEnvironment(fuel, fuelSource, extraEnv),
+            env: env ?? boundedEnvironment(fuel, fuelSource, extraEnv),
         });
         let stdout = "";
         let stderr = "";

package/dist/ensure-test-evidence.js

@@ -456,6 +456,18 @@ export function detectGateRanNothing(command, stdout, stderr, code) {
     }
     return null;
 }
+/**
+ * A delivery report that falls short of the delivery contract while the work itself may be
+ * complete: a required evidence kind is missing, a met criterion cites no evidence, test evidence is
+ * a summary, or the pull request or published artifact is missing. The type, not the sentence, is
+ * what tells the control plane this is a report defect for repair.
+ */
+export class DeliveryReportDefectError extends Error {
+    constructor(message) {
+        super(message);
+        this.name = "DeliveryReportDefectError";
+    }
+}
 /**
  * A declared gate that ran and failed. Carries the command as a field so the verification signal
  * can name the gate without reading it back out of this error's text.
@@ -501,7 +513,7 @@ export async function ensureTestEvidence(input) {
             ...named.commands,
         ])];
     if (commands.length === 0) {
-        throw new Error("Agent report is missing required evidence: test (no bounded verification command available for Bridge to run)");
+        throw new DeliveryReportDefectError("Agent report is missing required evidence: test (no bounded verification command available for Bridge to run)");
     }
     const run = input.runCommand ?? runBoundedVerificationCommand;
     const witnessed = [];

package/dist/execution.js

@@ -28,7 +28,7 @@ import { ensureLandCommit } from "./ensure-land-commit.js";
 import { AGENT_NO_LAND_COMMIT_PREFIX, AgentNoLandCommitError, agentNoLandCommitMessage, requiresLandCommit } from "./land-contract.js";
 import { agentClaimsRepositoryWork, claimsVsGitMismatch, isLandTreeEmpty, reportClaimsRepositoryWork, } from "./land-git-reality.js";
 const execFileAsync = promisify(execFile);
-import { boundedTail, captureVerificationFailure, detectGateRanNothing, ensureTestEvidence, needsTestEvidence, runBoundedVerificationCommand, verificationEvidenceCommands, VerificationFailedError, VerificationGateRanNothingError } from "./ensure-test-evidence.js";
+import { boundedTail, captureVerificationFailure, DeliveryReportDefectError, detectGateRanNothing, ensureTestEvidence, needsTestEvidence, runBoundedVerificationCommand, verificationEvidenceCommands, VerificationFailedError, VerificationGateRanNothingError } from "./ensure-test-evidence.js";
 import { ensureChangeEvidence } from "./ensure-change-evidence.js";
 import { ensureDemonstrationEvidence } from "./ensure-demonstration-evidence.js";
 import { ensureNormativeEvidence } from "./ensure-normative-evidence.js";
@@ -2264,11 +2264,16 @@ async function runAttempt(client, config, driver, workspace, brief, taskId, watc
                     .then((stat) => stat?.slice(0, 8_000) ?? null)
                     .catch(() => null)
                 : null;
+            // A report that fell short of the delivery contract is named by its type and travels as the
+            // delivery_report signal, so repair routing never reads this message (K3, debt 18 item 3c).
+            const reportDefect = error instanceof DeliveryReportDefectError;
             const classified = verificationFailed
                 ? { retryable: true, error: `Verification failed after the agent run.\n${message}${diffStat ? `\n\nChanged files:\n${diffStat}` : ""}` }
-                // classifyFinalizeFailure: forge transport → retryable; contract/environment/unknown → fail
-                // closed (unknown prefixed as "Bridge finalize interrupted", not laundered as contract).
-                : classifyFinalizeFailure(message);
+                : reportDefect
+                    ? { retryable: false, error: message }
+                    // classifyFinalizeFailure: forge transport → retryable; contract/environment/unknown → fail
+                    // closed (unknown prefixed as "Bridge finalize interrupted", not laundered as contract).
+                    : classifyFinalizeFailure(message);
             if (verificationFailed)
                 retainAttemptWorktree = true;
             const response = await queueTerminal(client, taskId, {
@@ -2284,6 +2289,7 @@ async function runAttempt(client, config, driver, workspace, brief, taskId, watc
                             command: error.command,
                             ...(diffStat ? { diff_stat: diffStat } : {}),
                         } } : {}),
+                    ...(reportDefect ? { signal: { phase: "delivery_report", witness: "bridge", kind: "contract" } } : {}),
                     // A rework's own PR may already be open (ensureDeliveryPullRequest ran before this
                     // gate). Carry the URL so the control plane can close it instead of littering.
                     ...(report.pull_request_url ? { pull_request_url: report.pull_request_url } : {}),
@@ -2759,7 +2765,7 @@ export function validateDeliveryReport(report, spec, grants = [], actualPaths, b
             }
         });
         if (!published) {
-            throw new Error("Artifact delivery requires a published HTTP(S) URL and sha256 digest");
+            throw new DeliveryReportDefectError("Artifact delivery requires a published HTTP(S) URL and sha256 digest");
         }
     }
     // Invariant 9: grants bound what an attempt may do. Cursor/plan-mode are soft hints the headless
@@ -2773,7 +2779,7 @@ export function validateDeliveryReport(report, spec, grants = [], actualPaths, b
     if (grants.includes("pr_create")
         && report.head_commit
         && !report.pull_request_url) {
-        throw new Error("Repository changes require a pull_request_url when pr_create is granted");
+        throw new DeliveryReportDefectError("Repository changes require a pull_request_url when pr_create is granted");
     }
     const requiredEvidence = spec.required_evidence ?? [];
     const supplied = new Set(report.evidence.map((item) => item.kind));
@@ -2782,20 +2788,20 @@ export function validateDeliveryReport(report, spec, grants = [], actualPaths, b
         // Name the base Bridge diffed against: an empty path list is a fact about that base, not proof
         // that the agent did no work. Without the base the repair brief can only guess the symptom.
         const since = missing.includes("change") && baseCommit ? ` (no paths changed since ${baseCommit})` : "";
-        throw new Error(`Agent report is missing required evidence: ${missing.join(", ")}${since}`);
+        throw new DeliveryReportDefectError(`Agent report is missing required evidence: ${missing.join(", ")}${since}`);
     }
     const unsupported = report.acceptance_results
         .filter((result) => result.status === "met" && !report.evidence.some((item) => item.acceptance_criteria.includes(result.criterion)))
         .map((result) => result.criterion);
     if (unsupported.length)
-        throw new Error(`Met acceptance criteria require mapped evidence: ${unsupported.join("; ")}`);
+        throw new DeliveryReportDefectError(`Met acceptance criteria require mapped evidence: ${unsupported.join("; ")}`);
     // Plan-time normalizeRequiredEvidence puts "test" on verify packages; reject summary-only details.
     if (requiredEvidence.includes("test")) {
         const thin = report.evidence
             .filter((item) => item.kind === "test")
             .filter((item) => item.details.join("\n").trim().length < TEST_EVIDENCE_DETAILS_MIN);
         if (thin.length) {
-            throw new Error("test evidence must include verbatim command output in details (not a summary)");
+            throw new DeliveryReportDefectError("test evidence must include verbatim command output in details (not a summary)");
         }
     }
 }

package/dist/failure-signal.js

@@ -228,9 +228,10 @@ export function contractScopeConflictFailure(detail) {
 function resetClock(resetsAt) {
     return `${resetsAt.slice(0, 16).replace("T", " ")} UTC`;
 }
-/** Signal tails are bounded: one runaway log cannot fill the wire or the card. */
+/** Signal tails are bounded: one runaway log cannot fill the wire or the card. A tail keeps the
+ *  end of the output, because the decisive error is printed last. */
 function bounded2k(text) {
-    return text.length > 2_000 ? `${text.slice(0, 1_999)}…` : text;
+    return text.length > 2_000 ? `…${text.slice(-1_999)}` : text;
 }
 /**
  * Vendor text that names the model as the reason the CLI refused the run. The model name is the one
@@ -402,10 +403,12 @@ export function verificationRedBaseSignal(input) {
 function factValue(value) {
     return value === null ? "none" : value;
 }
-/** The vendor's own reason, kept inside the 2,000-char message bound. */
+/** The vendor's own reason, kept inside the 2,000-char message bound; a stderr tail keeps its end. */
 function quotedVendorText(signal) {
-    const text = signal.vendor_message ?? signal.stderr_tail ?? "";
-    return text.length > 400 ? `${text.slice(0, 399)}…` : text;
+    if (signal.vendor_message)
+        return signal.vendor_message.length > 400 ? `${signal.vendor_message.slice(0, 399)}…` : signal.vendor_message;
+    const tail = signal.stderr_tail ?? "";
+    return tail.length > 400 ? `…${tail.slice(-399)}` : tail;
 }
 /**
  * Classify one typed signal. Null when the signal carries no typed verdict (`kind: "unknown"`):
@@ -614,7 +617,7 @@ export function classifyFailureSignal(signal) {
                 disposition: "rework",
                 responsible_party: "conduit",
                 message: "The delivery report did not satisfy the delivery contract.",
-                next_action: "Bridge's report-only repair turns ran; review the report defect and rework it.",
+                next_action: "Conductor will prepare bounded recovery guidance. You do not need to edit technical constraints.",
                 diagnostic_detail: signal.vendor_message ?? signal.stderr_tail,
             }
             : {

package/dist/investigation.js

@@ -12,11 +12,11 @@ import { createAttemptWorktree, removeAttemptWorktree } from "./attempt-worktree
 import { buildWorkspaceBrief, ensureCommitAvailable, normalizeRepositoryUrl } from "./brief.js";
 import { headCommitOrNull } from "./git.js";
 import { learnDriverFuel, resolveAssignmentModel, changedPathsSince, runDeliveryVerificationGates, discardWorktreeWrites } from "./execution.js";
-import { DRIVERS, extractAgentReportJsonText, parseJsonObjectCandidate } from "./driver.js";
+import { DRIVERS, execute, extractAgentReportJsonText, parseJsonObjectCandidate } from "./driver.js";
 import { pickDriverForClaim, resolveDriverFuel, supportsReadOnlyDiagnosis } from "./drivers.js";
 import { diffStatSince } from "./git-witness.js";
 import { switchManagedWorkspace } from "./on-shift-apply.js";
-import { execFile, spawn } from "node:child_process";
+import { execFile } from "node:child_process";
 import { rm, writeFile } from "node:fs/promises";
 import { createRequire } from "node:module";
 import { join } from "node:path";
@@ -95,15 +95,20 @@ const investigationBriefSchema = z.object({
         source: z.string().max(8_000).optional(),
         checks: z.string().max(500).optional().default(""),
     })).max(20).default([]),
+    /** K12: brief keys the lane names as stating no behaviour, so no probe is owed. Passed through, never run. */
+    no_behaviour: z.array(z.object({
+        key: z.string().trim().min(1).max(40),
+        reason: z.string().trim().min(1).max(500),
+    })).max(40).default([]),
 });
 /** T122: a delivery verification is an investigation whose question carries this header. */
 export const DELIVERY_VERIFICATION_HEADER = "DELIVERY VERIFICATION";
 /** Owner-verbatim: the lane writes probe source; Bridge executes. Shown only when BRIEF KEYS are present. */
-export const DELIVERY_VERIFICATION_PROBE_INSTRUCTION = 'For every BRIEF KEY that states behaviour, add one entry to probes: either {"key":"[sN]","test":"<registered command that asserts it>","checks":"<one line: what it proves>"} or {"key":"[sN]","lang":"ts|py|sh","source":"<a short script run from the repository root that prints PROBE HOLDS as its last line and exits 0 when the key holds at this commit and prints PROBE FAILS as its last line and exits 1 when it does not; any other ending is recorded as broken; it may import repository modules by paths relative to the repository root>","checks":"<one line>"}. You do not run probes; Bridge runs them after you finish. Keys that state no behaviour (wording, scope, docs) get no probe.';
+export const DELIVERY_VERIFICATION_PROBE_INSTRUCTION = 'For every BRIEF KEY that states behaviour, add one entry to probes: either {"key":"[sN]","test":"<registered command that asserts it>","checks":"<one line: what it proves>"} or {"key":"[sN]","lang":"ts|py|sh","source":"<a short script run from the repository root that prints PROBE HOLDS as its last line and exits 0 when the key holds at this commit and prints PROBE FAILS as its last line and exits 1 when it does not; any other ending is recorded as broken; it may import repository modules by paths relative to the repository root>","checks":"<one line>"}. You do not run probes; Bridge runs them after you finish. Every other BRIEF KEY states no behaviour (wording, scope, docs): add it to no_behaviour as {"key":"[sN]","reason":"<one line: why it states no behaviour>"}. Name every BRIEF KEY once, in probes or in no_behaviour.';
 const BRIEF_JSON_EXAMPLE = '{"summary":"one paragraph: what the change does and whether it meets the criteria","findings":["grounded fact with file/symbol"],"verification":["command — result"],"limitations":["what you could not settle"],"criteria":[{"criterion":"c1","assessment":"supported|unsupported|unclear","reason":"what you saw"}],"command_failures":["command that exited non-zero — the decisive output line"],"files_read":0,"bytes_read":0,"commands_run":["command you actually ran"]}';
-const BRIEF_JSON_EXAMPLE_WITH_PROBES = '{"summary":"one paragraph: what the change does and whether it meets the criteria","findings":["grounded fact with file/symbol"],"verification":["command — result"],"limitations":["what you could not settle"],"criteria":[{"criterion":"c1","assessment":"supported|unsupported|unclear","reason":"what you saw"}],"command_failures":["command that exited non-zero — the decisive output line"],"probes":[],"files_read":0,"bytes_read":0,"commands_run":["command you actually ran"]}';
+const BRIEF_JSON_EXAMPLE_WITH_PROBES = '{"summary":"one paragraph: what the change does and whether it meets the criteria","findings":["grounded fact with file/symbol"],"verification":["command — result"],"limitations":["what you could not settle"],"criteria":[{"criterion":"c1","assessment":"supported|unsupported|unclear","reason":"what you saw"}],"command_failures":["command that exited non-zero — the decisive output line"],"probes":[],"no_behaviour":[],"files_read":0,"bytes_read":0,"commands_run":["command you actually ran"]}';
 /** Named in both prompts so the lane knows each list's maximum before it writes. */
-export const BRIEF_LIMITS_SENTENCE = "Limits: findings at most 12, verification at most 8, limitations at most 8, criteria at most 40, command_failures at most 10, commands_run at most 10, probes at most 20; each text entry at most 1000 characters, commands_run entries at most 200 characters. Keep the most decisive entries.";
+export const BRIEF_LIMITS_SENTENCE = "Limits: findings at most 12, verification at most 8, limitations at most 8, criteria at most 40, command_failures at most 10, commands_run at most 10, probes at most 20, no_behaviour at most 40; each text entry at most 1000 characters, commands_run entries at most 200 characters. Keep the most decisive entries.";
 export function isDeliveryVerification(assignment) {
     return assignment.question.trimStart().startsWith(DELIVERY_VERIFICATION_HEADER);
 }
@@ -232,6 +237,7 @@ const BRIEF_LIST_MAX = {
     command_failures: 10,
     commands_run: 10,
     probes: 20,
+    no_behaviour: 40,
 };
 function clipBriefText(value, max) {
     return value.length > max ? value.slice(0, max) : value;
@@ -281,6 +287,9 @@ export function fitBriefToContract(candidate) {
     fitNamedList("command_failures", (items) => fitBriefStrings(items, BRIEF_ITEM_MAX));
     fitNamedList("commands_run", (items) => fitBriefStrings(items, BRIEF_COMMAND_RUN_MAX));
     fitNamedList("probes", (items) => items);
+    fitNamedList("no_behaviour", (items) => items.map((item) => item && typeof item === "object" && typeof item.reason === "string"
+        ? { ...item, reason: clipBriefText(item.reason, 500) }
+        : item));
     clipBriefStringField(brief, "summary", BRIEF_SUMMARY_MAX);
     if (typeof brief.files_read === "number" && brief.files_read > BRIEF_FILES_READ_MAX) {
         brief.files_read = BRIEF_FILES_READ_MAX;
@@ -625,34 +634,17 @@ function toStepWitness(witness) {
         instrument: witness.instrument === "delivered" ? "delivered" : "running",
     };
 }
+/**
+ * The delivered instrument runs through the agent runs' own process helper (T90): bounded output,
+ * an answer on the child's exit even when a process it left behind holds the pipes, and a stop
+ * that escalates to SIGKILL. Its environment stays the probe's own, PATH and HOME only.
+ */
 async function execDeliveredProbeRun(argv, stdin, workspace, timeoutMs) {
     const [file, ...args] = argv;
     if (!file)
         throw new Error("delivered probe command is empty");
-    return await new Promise((resolve, reject) => {
-        const child = spawn(file, args, {
-            cwd: workspace,
-            env: probeEnv(),
-            stdio: ["pipe", "pipe", "pipe"],
-        });
-        let stdout = "";
-        let stderr = "";
-        const timer = setTimeout(() => {
-            child.kill("SIGKILL");
-        }, timeoutMs);
-        child.stdout?.on("data", (chunk) => { stdout += String(chunk); });
-        child.stderr?.on("data", (chunk) => { stderr += String(chunk); });
-        child.on("error", (error) => {
-            clearTimeout(timer);
-            reject(error);
-        });
-        child.on("close", (code) => {
-            clearTimeout(timer);
-            resolve({ code: code ?? -1, stdout, stderr });
-        });
-        child.stdin?.on("error", () => undefined);
-        child.stdin?.end(stdin);
-    });
+    const result = await execute(file, args, workspace, timeoutMs, undefined, "conduit", stdin, undefined, undefined, probeEnv());
+    return { code: result.code ?? -1, stdout: result.stdout, stderr: result.stderr };
 }
 export async function runDeliveredProbeInstrument(input) {
     const entry = join(input.workspace, PROBE_RUN_ENTRY);
@@ -715,10 +707,6 @@ export const INVESTIGATION_RENEW_INTERVAL_MS = 2 * 60_000;
 export function investigationRunId(investigationId) {
     return `${investigationId}-${crypto.randomUUID().slice(0, 8)}`;
 }
-async function deleteRunBranch(workspace, runId) {
-    await execFileAsync("git", ["-C", workspace, "branch", "-D", `conduit/attempt/${runId}`], { timeout: 30_000, maxBuffer: 1_000_000 })
-        .catch(() => undefined);
-}
 export async function executeNextInvestigation(client, config, workspace, brief, timeoutMs, options = {}) {
     const data = await client.request("/runner/v1/investigations");
     const assignments = z.array(investigationAssignmentSchema).parse(data.investigations ?? []);
@@ -916,6 +904,7 @@ export async function executeNextInvestigation(client, config, workspace, brief,
                     ? gatesRun.filter((gate) => gate.exitCode !== 0).map(gateFailureFact).slice(0, 10)
                     : observed.command_failures,
                 step_witness: stepWitness,
+                no_behaviour: deliveryVerification ? observed.no_behaviour : [],
                 witness: {
                     repository_fingerprint: assignment.repository_fingerprint ?? liveRepository ?? "unknown",
                     commit,
@@ -949,7 +938,6 @@ export async function executeNextInvestigation(client, config, workspace, brief,
         if (attemptWorkspace) {
             await removeAttemptWorktree(workspace, attemptWorkspace)
                 .catch((error) => console.error(`Investigation worktree release failed: ${redactSecrets(error instanceof Error ? error.message : "unknown")}`));
-            await deleteRunBranch(workspace, runId);
         }
     }
     return true;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@miraland-labs/conduit-bridge",
-  "version": "0.16.140",
+  "version": "0.16.142",
   "description": "Conduit Bridge CLI \u2014 join, connect, disconnect, multi-driver lanes, and run Claude Code / Codex / Cursor / OpenCode / Pi / Kiro / Antigravity / Grok Build agents for a Conduit organization",
   "type": "module",
   "bin": {