gossipcat 0.4.30 → 0.4.31

package/README.md

@@ -7,7 +7,7 @@
 </p>
 
 <p align="center">
-  <a href="https://www.npmjs.com/package/gossipcat"><img src="https://img.shields.io/npm/v/gossipcat?color=0ea5e9" alt="npm version" /></a>
+  <a href="https://www.npmjs.com/package/gossipcat"><img src="https://img.shields.io/npm/v/gossipcat?color=0ea5e9&v=0430" alt="npm version" /></a>
   <a href="https://www.npmjs.com/package/gossipcat"><img src="https://img.shields.io/npm/dw/gossipcat?color=0ea5e9" alt="npm weekly downloads" /></a>
   <a href="https://github.com/gossipcat-ai/gossipcat-ai/blob/master/LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue" alt="MIT License" /></a>
   <a href="#quickstart"><img src="https://img.shields.io/badge/node-22%2B-green" alt="Node 22+" /></a>
@@ -85,15 +85,14 @@ The core difference: gossipcat verifies findings against actual `file:line` cita
 
 ## Real-world session
 
-What a typical gossipcat session looks like in practice (2026-04-29):
+What a typical gossipcat session looks like in practice (2026-05-22, v0.4.30 ship):
 
-- **2 PRs shipped** — #317 (env-scrub fix, e180d41) and #318 (test hardening, cac57db), both through full consensus before merge
-- **1 consensus round** on PR #317 — round `591af14b-3f674c9c`, 9 confirmed / 0 disputed findings
-- **1 stale backlog item correctly identified** — write-time insight filter was already shipped in a prior session; the `verify-the-premise` skill caught it before sonnet-implementer started redundant implementation work
-- **1 spec correctly deferred** — age-based archive pruning: full scan found 0 candidates meeting the threshold; design locked in docs, not built (right call)
-- **1 hallucination caught** on haiku-researcher — extrapolated from a 50-row sample when the full dataset was needed; `hallucination_caught` signal recorded, accuracy score updated, no fix shipped on bad data
+- **1 feature shipped end-to-end** — consensus auto-verify (PR #448, master commit `4b28a1c`, 1255+ LOC, 50/50 new tests, zero regressions). Full design ↔ ship arc through gossipcat itself: 6 consensus rounds on the spec before any code was written.
+- **6 consensus rounds on the spec** caught **21+ HIGH-severity defects** across rev-1 → rev-6 — including a double-dispatch bug on the `run()` path (rev-3, opus-implementer), a phantom `AgentTeam` type the implementer would have invented if shipped (rev-5, sonnet-reviewer grep-grounded against live `AgentConfig`), and a `metadata` field that didn't exist on `ConsensusSignal` at all (rev-4, sonnet — `metadata` lives only on `MetaSignal` / `PipelineSignal`). Each round produced a measurable rev: rev-1 had 4 HIGHs, rev-6 had 0.
+- **1 post-merge bug caught by a pre-existing drift test** — `signal-allowlist-drift.test.ts:108` (which exists precisely to catch this — same failure mode as PR #329's silent-drop of `transport_failure`) flagged that the implementer added the 2 new signals to `KNOWN_SIGNALS` + the type union + `OPERATIONAL_SIGNAL_NAMES` but missed `VALID_CONSENSUS_SIGNALS` in `performance-writer.ts`. 7-line fix landed in the same PR before merge.
+- **Methodology lesson recorded** — sonnet-reviewer's habit of grepping cited file:line against live code (rather than trusting prose claims) is what ultimately closed the spec. Recorded as a `citation_grounding` agreement signal so the agent's pattern compounds across sessions.
 
-The signal pipeline ran the whole session. Nothing landed without cross-review. One agent's score dropped for the sample-extrapolation error.
+Nothing landed without cross-review. Two agents got `+/-` score adjustments based on what they caught vs. what they missed. The spec is now usable as a worked example of what 6 rounds of multi-agent design review looks like — `docs/superpowers/specs/2026-05-21-consensus-auto-verify-design.md`.
 
 <br/>
 
@@ -149,6 +148,12 @@ Real-time view of tasks, consensus reports, agent scores, and activity feed. Ter
 Per-agent cognitive memory persists across sessions. Agents remember past findings, patterns, and project context.
 </td>
 </tr>
+<tr>
+<td align="center" colspan="3">
+<h3>Auto-Verify (v0.4.30)</h3>
+Opt-in. Every UNVERIFIED finding gets <code>file_read</code>-checked by a verifier agent before the report is returned. <code>tag</code> stays <code>'unverified'</code> — auto-verify is metadata, not state transition. Flag: <code>GOSSIP_CONSENSUS_AUTO_VERIFY_UNVERIFIED=1</code>.
+</td>
+</tr>
 </table>
 
 <br/>

package/dist-mcp/mcp-server.js

@@ -50040,10 +50040,16 @@ function stampConcurrencyTaint(map2) {
   }
   return true;
 }
-function elidePromptIfRequested(projectRoot, taskId, agentPrompt, promptFormat, warmCached = false) {
+function elidePromptIfRequested(projectRoot, taskId, agentPrompt, promptFormat, warmCached = false, writeMode) {
   if (promptFormat !== "elided") return { elided: false };
-  const bytes = Buffer.byteLength(agentPrompt, "utf8");
-  const promptPath = writeDispatchPrompt(projectRoot, taskId, agentPrompt);
+  const body = writeMode === "worktree" ? `// GOSSIP_ISOLATION: worktree
+// This task was dispatched with write_mode: "worktree".
+// The orchestrator MUST invoke Agent() with isolation: "worktree".
+// Do not paraphrase this requirement.
+
+` + agentPrompt : agentPrompt;
+  const bytes = Buffer.byteLength(body, "utf8");
+  const promptPath = writeDispatchPrompt(projectRoot, taskId, body);
   const warmSuffix = warmCached ? " \u2014 warm-cached (skills) + live task" : "";
   const marker = `[skills section elided: see ${promptPath}, ${bytes} bytes${warmSuffix} \u2014 READ this file and pass its CONTENTS verbatim as the Agent(prompt: ...) value. Do NOT pass the path string.]`;
   return { elided: true, promptPath, marker, bytes };
@@ -50415,17 +50421,31 @@ Task: ${task}`;
       worktreePath: void 0,
       timestamp: Date.now()
     });
-    const elision = elidePromptIfRequested(process.cwd(), taskId, agentPrompt, prompt_format, singleWarm);
+    const elision = elidePromptIfRequested(
+      process.cwd(),
+      taskId,
+      agentPrompt,
+      prompt_format,
+      singleWarm,
+      useWorktree ? "worktree" : void 0
+    );
     if (elision.elided) {
       const info = ctx.nativeTaskMap.get(taskId);
       if (info) info.promptPath = elision.promptPath;
     }
     persistNativeTaskMap();
+    const promptRef = elision.elided ? "<file contents>" : `<AGENT_PROMPT:${taskId} below>`;
+    const agentCall = useWorktree ? `Agent(
+  model: "${nativeConfig.model}",
+  prompt: ${promptRef},
+  isolation: "worktree",           // REQUIRED \u2014 do not omit
+  run_in_background: true
+)` : `Agent(model: "${nativeConfig.model}", prompt: ${promptRef}, run_in_background: true)`;
     const promptInstruction = elision.elided ? `Step 1 \u2014 ${elision.marker}
-Agent(model: "${nativeConfig.model}", prompt: <file contents>${useWorktree ? ', isolation: "worktree"' : ""}, run_in_background: true)
+${agentCall}
 
 ` : `Step 1 \u2014 Pass the AGENT_PROMPT:${taskId} content item below verbatim to Agent(prompt: ...):
-Agent(model: "${nativeConfig.model}", prompt: <AGENT_PROMPT:${taskId} below>${useWorktree ? ', isolation: "worktree"' : ""}, run_in_background: true)
+${agentCall}
 
 `;
     return { content: [
@@ -50437,8 +50457,10 @@ NATIVE_DISPATCH: Execute this via Claude Code Agent tool, then relay the result.
 Task ID: ${taskId}
 Agent: ${agent_id}
 Model: ${nativeConfig.model}
+` + (useWorktree ? `Worktree isolation: REQUIRED \u2014 Agent() MUST be invoked with isolation: "worktree"
 
-` + promptInstruction + `Step 2 \u2014 REQUIRED after agent completes:
+` : `
+`) + promptInstruction + `Step 2 \u2014 REQUIRED after agent completes:
 gossip_relay(task_id: "${taskId}", relay_token: "${relayToken}", result: "<agent output>")
 (VERBATIM \u2014 pass the agent's raw output; do NOT paraphrase or summarize, or <agent_finding> tags will be lost)
 
@@ -50680,15 +50702,24 @@ Task: ${def.task}`;
       worktreePath: void 0,
       timestamp: Date.now()
     });
-    const parallelElision = elidePromptIfRequested(process.cwd(), taskId, agentPrompt, prompt_format, parallelWarm);
+    const parallelElision = elidePromptIfRequested(process.cwd(), taskId, agentPrompt, prompt_format, parallelWarm, def.write_mode);
     if (parallelElision.elided) {
       const info = ctx.nativeTaskMap.get(taskId);
       if (info) info.promptPath = parallelElision.promptPath;
     }
     const parallelPromptRef = parallelElision.elided ? parallelElision.marker : `<AGENT_PROMPT:${taskId} below>`;
+    const parallelUseWorktree = def.write_mode === "worktree";
+    const parallelWorktreeBanner = parallelUseWorktree ? `
+  Worktree isolation: REQUIRED \u2014 Agent() MUST be invoked with isolation: "worktree"` : "";
+    const parallelAgentCall = parallelUseWorktree ? `Agent(
+  model: "${nativeConfig.model}",
+  prompt: ${parallelPromptRef},
+  isolation: "worktree",           // REQUIRED \u2014 do not omit
+  run_in_background: true
+)` : `Agent(model: "${nativeConfig.model}", prompt: ${parallelPromptRef}, run_in_background: true)`;
     lines.push(`  ${taskId} \u2192 ${def.agent_id} (native \u2014 dispatch via Agent tool)`);
     nativeInstructions.push(
-      `[${taskId}] Agent(model: "${nativeConfig.model}", prompt: ${parallelPromptRef}${def.write_mode === "worktree" ? ', isolation: "worktree"' : ""}, run_in_background: true)
+      `[${taskId}] ${parallelAgentCall}${parallelWorktreeBanner}
   \u2192 then: gossip_relay(task_id: "${taskId}", relay_token: "${relayToken}", result: "<output>")`
     );
     if (!parallelElision.elided) {
@@ -50898,16 +50929,25 @@ Task: ${def.task}`;
       }
     }
     agentPrompt = maybeApplyUnverifiedNote(agentPrompt, def.task, def.agent_id);
-    const consensusElision = elidePromptIfRequested(process.cwd(), taskId, agentPrompt, prompt_format, consensusWarm);
+    const consensusElision = elidePromptIfRequested(process.cwd(), taskId, agentPrompt, prompt_format, consensusWarm, def.write_mode);
     if (consensusElision.elided) {
       const info = ctx.nativeTaskMap.get(taskId);
       if (info) info.promptPath = consensusElision.promptPath;
       persistNativeTaskMap();
     }
     const consensusPromptRef = consensusElision.elided ? consensusElision.marker : `<AGENT_PROMPT:${taskId} below>`;
+    const consensusUseWorktree = def.write_mode === "worktree";
+    const consensusWorktreeBanner = consensusUseWorktree ? `
+  Worktree isolation: REQUIRED \u2014 Agent() MUST be invoked with isolation: "worktree"` : "";
+    const consensusAgentCall = consensusUseWorktree ? `Agent(
+  model: "${nativeConfig.model}",
+  prompt: ${consensusPromptRef},
+  isolation: "worktree",           // REQUIRED \u2014 do not omit
+  run_in_background: true
+)` : `Agent(model: "${nativeConfig.model}", prompt: ${consensusPromptRef}, run_in_background: true)`;
     lines.push(`  ${taskId} \u2192 ${def.agent_id} (native \u2014 dispatch via Agent tool)`);
     nativeInstructions.push(
-      `[${taskId}] Agent(model: "${nativeConfig.model}", prompt: ${consensusPromptRef}, run_in_background: true)
+      `[${taskId}] ${consensusAgentCall}${consensusWorktreeBanner}
   \u2192 then: gossip_relay(task_id: "${taskId}", relay_token: "${relayToken}", result: "<output>")`
     );
     if (!consensusElision.elided) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gossipcat",
-  "version": "0.4.30",
+  "version": "0.4.31",
   "description": "Multi-agent orchestration for Claude Code — parallel review, consensus, adaptive dispatch",
   "mcpName": "io.github.ataberk-xyz/gossipcat",
   "repository": {