Harden marketplace task creation

docs/audit/AUDIT_ROADMAP.md

@@ -15,6 +15,19 @@ The goal of this roadmap is to make the program correct, predictable, and
 ready for an on-chain audit before mainnet deployment or SDK integration.
 
 
+## Current hardening delta
+
+The marketplace job specification flow is part of the active hardening scope.
+Runtime claims for marketplace tasks must verify the content-addressed job spec
+before signing a claim transaction, and the coordination program exposes a
+`claim_task_with_job_spec` path that requires the matching `task_job_spec` PDA.
+
+This is an allowed freeze-period change because it narrows the execution surface:
+workers should not claim marketplace work from prompt text or remote payloads
+that have not been matched to the on-chain job spec hash and URI. Protocol,
+SDK, and runtime changes for this surface must be validated together.
+
+
 ## Guiding principle
 
 No new features are added until correctness is proven. This phase is about reducing risk, not increasing surface area. The program should become boring to read and impossible to surprise.

runtime/package.json

@@ -41,7 +41,7 @@
     "check:executor-baseline": "node scripts/check-executor-baseline.mjs",
     "check-idl-drift": "tsx scripts/check-idl-drift.ts",
     "pretest": "npm run build --workspace=@tetsuo-ai/desktop-tool-contracts && npm run build --workspace=@tetsuo-ai/plugin-kit-channel-fixture",
-    "test": "vitest run --exclude tests/integration.test.ts --exclude tests/eval-replay.integration.test.ts --exclude tests/benchmark-runner.integration.test.ts",
+    "test": "vitest run",
     "validate:required": "npx tsx scripts/run-required-validation.ts",
     "test:marketplace-integration": "vitest run tests/marketplace-cli.integration.test.ts",
     "test:cross-repo-integration": "vitest run tests/integration.test.ts tests/eval-replay.integration.test.ts tests/benchmark-runner.integration.test.ts tests/marketplace-cli.integration.test.ts",

runtime/src/channels/discord/plugin.ts

@@ -292,7 +292,9 @@ export class DiscordChannel extends BaseChannelPlugin {
     });
 
     client.on("interactionCreate", (interaction: unknown) => {
-      this.handleInteraction(interaction as DiscordInteraction).catch((err) => {
+      return this.handleInteraction(
+        interaction as DiscordInteraction,
+      ).catch((err) => {
         this.context.logger.error(
           `Error handling interactionCreate: ${errorMessage(err)}`,
         );

runtime/src/channels/webchat/handlers.ts

@@ -1245,7 +1245,9 @@ async function handleTasksCreate(
       }
     }
     const { program } = await createProgramContext(deps);
-    const tool = createCreateTaskTool(program, silentLogger);
+    const tool = createCreateTaskTool(program, silentLogger, {
+      allowRawTaskCreation: true,
+    });
     const result = await tool.execute(createArgs);
     if (result.isError) {
       send({ type: 'error', error: `Failed to create task: ${parseToolError(result)}`, id });

runtime/src/channels/webchat/plugin.test.ts

@@ -1998,8 +1998,12 @@ describe("WebChatChannel", () => {
           ),
           send,
         );
-        await new Promise((resolve) => setTimeout(resolve, 0));
-        expect(findResponse(send, "watch.cockpit", "req-watch-cockpit")?.payload).toEqual(
+        const cockpitResponse = await waitForResponse(
+          send,
+          "watch.cockpit",
+          "req-watch-cockpit",
+        );
+        expect(cockpitResponse.payload).toEqual(
           expect.objectContaining({
             session: expect.objectContaining({
               sessionId: "session-continuity",
@@ -2092,18 +2096,20 @@ describe("WebChatChannel", () => {
           ),
           send,
         );
-        await new Promise((resolve) => setTimeout(resolve, 0));
-
-        const forkResponse = findResponse(send, "chat.session.fork", "req-fork");
-        expect(forkResponse?.payload).toEqual(
+        const forkResponse = await waitForResponse(
+          send,
+          "chat.session.fork",
+          "req-fork",
+        );
+        expect(forkResponse.payload).toEqual(
           expect.objectContaining({
             sourceSessionId: "session-source",
             forkSource: "runtime_state",
             targetSessionId: expect.any(String),
           }),
         );
 
-        const targetSessionId = (forkResponse?.payload as Record<string, unknown>)
+        const targetSessionId = (forkResponse.payload as Record<string, unknown>)
           .targetSessionId as string;
         expect(await store.loadSession(targetSessionId)).toMatchObject({
           metadata: {

runtime/src/cli/index.test.ts

@@ -783,6 +783,8 @@ describe("runtime root CLI", () => {
         "Task111111111111111111111111111111111111111",
         "--worker-agent-pda",
         "Agent11111111111111111111111111111111111111",
+        "--job-spec-store-dir",
+        "/tmp/agenc-job-specs",
         "--output",
         "json",
       ],
@@ -797,6 +799,7 @@ describe("runtime root CLI", () => {
       expect.objectContaining({
         taskPda: "Task111111111111111111111111111111111111111",
         workerAgentPda: "Agent11111111111111111111111111111111111111",
+        jobSpecStoreDir: "/tmp/agenc-job-specs",
       }),
     );
   });

runtime/src/cli/index.ts

@@ -672,7 +672,7 @@ const MARKET_COMMAND_OPTIONS: Record<MarketCommand, Set<string>> = {
   ]),
   "tasks.detail": new Set(["job-spec-store-dir"]),
   "tasks.cancel": new Set(),
-  "tasks.claim": new Set(["worker-agent-pda"]),
+  "tasks.claim": new Set(["worker-agent-pda", "job-spec-store-dir"]),
   "tasks.complete": new Set(["proof-hash", "result-data", "worker-agent-pda"]),
   "tasks.dispute": new Set([
     "evidence",
@@ -2954,6 +2954,9 @@ function normalizeAndValidateMarketCommand(
         ...base,
         taskPda,
         workerAgentPda: parseOptionalStringFlag(parsed.flags["worker-agent-pda"]),
+        jobSpecStoreDir: parseOptionalStringFlag(
+          parsed.flags["job-spec-store-dir"],
+        ),
       } as MarketTaskClaimOptions;
       break;
     }

runtime/src/cli/marketplace-cli.ts

@@ -1073,9 +1073,10 @@ export async function runMarketTaskCreateCommand(
 
   try {
     const { program } = await createSignerProgramContext(options);
-    const createTaskOptions = options.jobSpecStoreDir
-      ? { jobSpecStoreDir: options.jobSpecStoreDir }
-      : undefined;
+    const createTaskOptions = {
+      ...(options.jobSpecStoreDir ? { jobSpecStoreDir: options.jobSpecStoreDir } : {}),
+      allowRawTaskCreation: true,
+    };
     const tool = createCreateTaskTool(program, silentLogger, createTaskOptions);
     const result = await tool.execute({
       description: options.description,
@@ -1233,7 +1234,16 @@ export async function runMarketTaskClaimCommand(
 
   try {
     const { program } = await createSignerProgramContext(options);
-    const tool = createClaimTaskTool(program, silentLogger);
+    const tool = createClaimTaskTool(
+      program,
+      silentLogger,
+      options.jobSpecStoreDir
+        ? {
+            jobSpecStoreDir: options.jobSpecStoreDir,
+            claimJobSpecVerification: "required",
+          }
+        : {},
+    );
     const result = await tool.execute({
       taskPda: options.taskPda,
       workerAgentPda: options.workerAgentPda,

runtime/src/eval/pipeline-http-repro.ts

@@ -142,7 +142,12 @@ export async function runPipelineHttpRepro(
   });
 
   const step2 = await runBash(
-    `cd ${workspace} && python3 -m http.server ${port} >/tmp/agenc-http.log 2>&1 & echo $!`,
+    [
+      "set -euo pipefail",
+      `cd ${workspace}`,
+      `python3 -m http.server ${port} >/tmp/agenc-http.log 2>&1 &`,
+      "echo $!",
+    ].join("\n"),
   );
   if (step2.ok) {
     serverPid = step2.stdout.split(/\s+/)[0]?.trim();
@@ -184,35 +189,35 @@ export async function runPipelineHttpRepro(
     preview: preview(step5),
   });
 
+  const serverPidArg = serverPid && /^\d+$/.test(serverPid) ? serverPid : undefined;
   const step6 = await runBash(
     [
-      "if command -v lsof >/dev/null 2>&1; then",
-      `  SERVER_PIDS=$(lsof -nP -iTCP:${port} -sTCP:LISTEN -t 2>/dev/null || true)`,
-      '  for pid in $SERVER_PIDS; do',
-      '    kill "$pid" >/dev/null 2>&1 || true',
-      "  done",
-      "elif command -v fuser >/dev/null 2>&1; then",
+      "set +e",
+      ...(serverPidArg
+        ? [`kill ${serverPidArg} >/dev/null 2>&1 || true`]
+        : []),
+      "for _ in 1 2 3 4 5; do",
+      `  if ! curl -fsS --max-time 1 http://127.0.0.1:${port} >/dev/null 2>&1; then echo 0; exit 0; fi`,
+      "  sleep 0.2",
+      "done",
+      "if command -v fuser >/dev/null 2>&1; then",
       `  fuser -k ${port}/tcp >/dev/null 2>&1 || true`,
-      "else",
-      `  pkill -f 'http.server ${port}' >/dev/null 2>&1 || true`,
       "fi",
-      "sleep 1",
-      `if command -v ss >/dev/null 2>&1; then`,
-      `  ss -ltn '( sport = :${port} )' | tail -n +2 | wc -l`,
-      "elif command -v lsof >/dev/null 2>&1; then",
-      `  lsof -nP -iTCP:${port} -sTCP:LISTEN 2>/dev/null | tail -n +2 | wc -l`,
-      "else",
-      `  if curl -fsS --max-time 1 http://127.0.0.1:${port} >/dev/null 2>&1; then echo 1; else echo 0; fi`,
+      "if command -v pkill >/dev/null 2>&1; then",
+      `  pkill -f 'http.server ${port}' >/dev/null 2>&1 || true`,
       "fi",
+      "sleep 0.5",
+      `if curl -fsS --max-time 1 http://127.0.0.1:${port} >/dev/null 2>&1; then echo 1; else echo 0; fi`,
     ].join("\n"),
   );
-  const portListeners = Number(step6.stdout.trim() || "0");
+  const listenerStatus = step6.stdout.trim().split(/\s+/).at(-1) ?? "1";
+  const portListeners = Number(listenerStatus);
   steps.push({
     step: 6,
     tool: "system.bash",
     ok: step6.ok && Number.isFinite(portListeners) && portListeners === 0,
     preview: step6.ok
-      ? `listeners_on_${port}=${step6.stdout.trim()}`
+      ? `listeners_on_${port}=${listenerStatus}`
       : preview(step6),
   });
 

runtime/src/gateway/approvals.test.ts

@@ -178,9 +178,9 @@ describe('ApprovalEngine', () => {
       expect(engine.requiresApproval('wallet.transfer', {})).toBeNull();
     });
 
-    it('matches agenc.createTask with reward > 1 SOL (lamports)', () => {
-      expect(engine.requiresApproval('agenc.createTask', { reward: 1_000_000_001 })).not.toBeNull();
-      expect(engine.requiresApproval('agenc.createTask', { reward: 1_000_000_000 })).toBeNull();
+    it('always requires approval for raw agenc.createTask', () => {
+      expect(engine.requiresApproval('agenc.createTask', { reward: 1 })).not.toBeNull();
+      expect(engine.requiresApproval('agenc.createTask', {})).not.toBeNull();
     });
 
     it('always requires approval for agenc.registerAgent', () => {
@@ -1063,9 +1063,9 @@ describe('ApprovalEngine', () => {
       expect(rule!.conditions!.minAmount).toBe(0.1);
     });
 
-    it('agenc.createTask has minAmount 1 SOL in lamports', () => {
+    it('agenc.createTask has no threshold condition', () => {
       const rule = DEFAULT_APPROVAL_RULES.find((r) => r.tool === 'agenc.createTask');
-      expect(rule!.conditions!.minAmount).toBe(1_000_000_000);
+      expect(rule!.conditions).toBeUndefined();
     });
 
     it('agenc.stakeReputation has minAmount 0.1 SOL in lamports', () => {

runtime/src/gateway/approvals.ts

@@ -303,8 +303,7 @@ export const DEFAULT_APPROVAL_RULES: readonly ApprovalRule[] = [
   },
   {
     tool: "agenc.createTask",
-    conditions: { minAmount: 1_000_000_000 },
-    description: "Task creation with reward exceeding 1 SOL",
+    description: "Raw marketplace task creation",
   },
   {
     tool: 'agenc.registerAgent',

runtime/src/gateway/system-prompt-routing.test.ts

@@ -61,7 +61,7 @@ You have broad access to this machine via the system.bash tool.`;
 describe("system prompt routing filter", () => {
   it("detects when routed tools still require protocol context", () => {
     expect(
-      hasProtocolToolRouting(["system.bash", "agenc.createTask"]),
+      hasProtocolToolRouting(["system.bash", "agenc.createTaskFromTemplate"]),
     ).toBe(true);
     expect(
       hasProtocolToolRouting(["system.bash", "social.sendMessage"]),
@@ -89,7 +89,7 @@ describe("system prompt routing filter", () => {
   it("preserves protocol sections when routed tools include protocol families", () => {
     const filtered = filterSystemPromptForToolRouting({
       systemPrompt: SYSTEM_PROMPT,
-      routedToolNames: ["system.bash", "agenc.createTask"],
+      routedToolNames: ["system.bash", "agenc.createTaskFromTemplate"],
     });
 
     expect(filtered).toContain("Solana:");

runtime/src/gateway/worktree-isolation.ts

@@ -25,6 +25,34 @@ function cloneExecutionLocation(
   return JSON.parse(JSON.stringify(location)) as RuntimeExecutionLocation;
 }
 
+function getEquivalentPathRoots(path: string): readonly string[] {
+  const normalizedPath = resolvePath(path);
+  if (process.platform !== "darwin") {
+    return [normalizedPath];
+  }
+  if (normalizedPath.startsWith("/private/var/")) {
+    return [normalizedPath, normalizedPath.slice("/private".length)];
+  }
+  if (normalizedPath.startsWith("/var/")) {
+    return [normalizedPath, `/private${normalizedPath}`];
+  }
+  return [normalizedPath];
+}
+
+function relativeToEquivalentRoot(
+  path: string,
+  root: string,
+): string | undefined {
+  for (const rootCandidate of getEquivalentPathRoots(root)) {
+    for (const pathCandidate of getEquivalentPathRoots(path)) {
+      if (isPathWithinRoot(pathCandidate, rootCandidate)) {
+        return relative(rootCandidate, pathCandidate);
+      }
+    }
+  }
+  return undefined;
+}
+
 function translatePathForWorktree(
   path: string | undefined,
   location: RuntimeExecutionLocation,
@@ -39,12 +67,16 @@ function translatePathForWorktree(
   }
   const normalizedPath = resolvePath(path);
   const normalizedGitRoot = resolvePath(location.gitRoot);
-  if (!isPathWithinRoot(normalizedPath, normalizedGitRoot)) {
+  const relativePath = relativeToEquivalentRoot(
+    normalizedPath,
+    normalizedGitRoot,
+  );
+  if (relativePath === undefined) {
     return normalizedPath;
   }
   return resolvePath(
     location.worktreePath,
-    relative(normalizedGitRoot, normalizedPath),
+    relativePath,
   );
 }