0xray 3.1.1 → 3.3.0

package/dist/inference/inference-cycle.js

@@ -2,31 +2,18 @@ import * as fs from "fs";
 import * as path from "path";
 import { shouldTriggerCycle, accumulateCorpus } from "./inference-accumulator.js";
 import { DeployVerifier } from "./deploy-verifier.js";
-import { VotingCoordinator } from "../delegation/voting-coordinator.js";
-import { XrayStateManager } from "../state/state-manager.js";
 import { frameworkLogger } from "../core/framework-logger.js";
-import { getGovernanceIntegration } from "../integrations/governance/index.js";
 import { featuresConfigLoader } from "../core/features-config.js";
 import { mcpClientManager } from "../mcps/mcp-client.js";
 import { getConfigDir } from "../core/config-paths.js";
-import { invokeViaOpencode as invokeOpencodeFromEngine } from "../execution/opencode-cli-invoker.js";
-import { ProposalApplier } from "../execution/proposal-applier.js";
+import { generateProposals } from "./inference-proposal-generator.js";
+import { applyProposals as applyProposalsEx } from "./inference-applier.js";
 const CYCLE_STATE_FILE = "inference-cycle-state.json";
 const CYCLE_HISTORY_FILE = "inference-cycle-history.json";
-const GOVERNANCE_AGENTS = {
-    // Real individual knowledge-skill MCP servers (have analyze_proposal handlers)
-    fix: ["code-review", "security-audit", "researcher"],
-    refactor: ["code-review", "security-audit", "researcher"],
-    guard: ["code-review", "security-audit", "researcher"],
-    automate: ["code-review", "security-audit", "researcher"],
-    codify: ["code-review", "security-audit", "researcher"],
-};
 export class InferenceCycle {
-    // Singleton registry + state management to prevent recursive agent spawning
     static instances = new Map();
     static reEntryLock = false;
     static governedProposalIds = new Set();
-    static votingCoordinator = null;
     static getInstance(projectRoot, options) {
         const root = path.resolve(projectRoot || process.cwd());
         if (!InferenceCycle.instances.has(root)) {
@@ -38,7 +25,6 @@ export class InferenceCycle {
         InferenceCycle.instances.clear();
         InferenceCycle.reEntryLock = false;
         InferenceCycle.governedProposalIds.clear();
-        InferenceCycle.votingCoordinator = null;
     }
     inferenceDir;
     stateDir;
@@ -55,7 +41,6 @@ export class InferenceCycle {
         this.options = options ?? {};
     }
     async governExternalProposals(proposals) {
-        // GATE: prevent re-entry — same as maybeRunCycle
         if (InferenceCycle.reEntryLock) {
             const blockedId = `blocked-${Date.now()}`;
             frameworkLogger.log("inference-cycle", "external-re-entry-blocked", "warning", {});
@@ -93,12 +78,10 @@ export class InferenceCycle {
                 p.status = "approved";
             for (const p of proposals.filter((p) => p.status !== "approved"))
                 p.status = "rejected";
-            // Track governed proposals for dedup across cycles
             for (const p of proposals)
                 InferenceCycle.governedProposalIds.add(p.id);
             this.setPhase("complete");
             this.saveCycleState(cycleId);
-            this.saveGovernanceState(this.getCoordinator());
             const result = this.buildResult(cycleId, true, "external proposals", startTime, undefined, proposals, votes);
             this.appendHistory(result);
             frameworkLogger.log("inference-cycle", "external-governance-complete", "info", {
@@ -113,7 +96,6 @@ export class InferenceCycle {
         }
     }
     async maybeRunCycle() {
-        // GATE: prevent recursive re-entry while a cycle is in progress
         if (InferenceCycle.reEntryLock) {
             const blockedId = `blocked-${Date.now()}`;
             frameworkLogger.log("inference-cycle", "re-entry-blocked", "warning", {});
@@ -136,14 +118,30 @@ export class InferenceCycle {
             });
             const corpus = accumulateCorpus(this.inferenceDir);
             this.setPhase("proposing");
-            const proposals = this.generateProposals(corpus);
-            this.adjustFromHistory(proposals);
+            const proposals = generateProposals(corpus, this.loadHistory());
             if (proposals.length === 0) {
                 this.setPhase("complete");
                 return this.buildResult(cycleId, true, threshold.reason, startTime, corpus, proposals);
             }
             this.setPhase("governing");
-            const votes = await this.governProposals(proposals);
+            let votes = [];
+            try {
+                votes = await this.governProposals(proposals);
+            }
+            catch (e) {
+                frameworkLogger.log("inference-cycle", "govern-proposals-error", "error", {
+                    error: e.message,
+                    stack: e.stack?.substring(0, 500),
+                });
+                for (const proposal of proposals) {
+                    votes.push({
+                        proposalId: proposal.id,
+                        decision: "reject",
+                        confidence: 0,
+                        details: [`rejected: governance error: ${e.message}`],
+                    });
+                }
+            }
             // Track governed proposals for dedup across cycles
             for (const p of proposals)
                 InferenceCycle.governedProposalIds.add(p.id);
@@ -157,7 +155,7 @@ export class InferenceCycle {
             if (approved.length > 0) {
                 if (!this.options.skipApply) {
                     this.setPhase("applying");
-                    await this.applyProposals(approved);
+                    await applyProposalsEx(approved, this.projectRoot, this.agentInvoker, this.options);
                 }
                 let deployResult;
                 if (!this.options.skipDeployVerify) {
@@ -188,7 +186,6 @@ export class InferenceCycle {
                 }
                 this.setPhase("complete");
                 this.saveCycleState(cycleId);
-                this.saveGovernanceState(this.getCoordinator());
                 this.appendHistory(this.buildResult(cycleId, true, threshold.reason, startTime, corpus, proposals, votes, deployResult));
                 return this.buildResult(cycleId, true, threshold.reason, startTime, corpus, proposals, votes, deployResult);
             }
@@ -201,311 +198,38 @@ export class InferenceCycle {
             InferenceCycle.reEntryLock = false;
         }
     }
-    generateProposals(corpus) {
-        const proposals = [];
-        for (const problem of corpus.recurringProblems) {
-            proposals.push({
-                id: `prop-${Date.now()}-${proposals.length}`,
-                type: this.classifyProposalType(problem.pattern),
-                title: this.generateTitle(problem),
-                description: `Recurring across ${problem.occurrences} sessions: ${problem.pattern}`,
-                evidence: problem.sessions.map((s) => `Seen in session ${s}`),
-                confidence: Math.min(0.95, 0.5 + problem.occurrences * 0.15),
-                source: "recurring_problem",
-                status: "pending",
-            });
-        }
-        const seenPatterns = new Set(corpus.recurringProblems.map((p) => p.pattern));
-        const allProblems = corpus.sessions.flatMap((s) => s.problems.map((p) => ({ problem: p, session: s.sessionId })));
-        for (const { problem, session } of allProblems) {
-            const normalized = problem.replace(/\([a-f0-9]{7}\)/g, "").trim();
-            if (seenPatterns.has(normalized))
-                continue;
-            if (proposals.length >= 5)
-                break;
-            proposals.push({
-                id: `prop-${Date.now()}-${proposals.length}`,
-                type: this.classifyProposalType(problem),
-                title: `Investigate: ${problem.substring(0, 80)}`,
-                description: `Observed in session ${session}: ${problem}`,
-                evidence: [problem],
-                confidence: 0.4,
-                source: "recurring_problem",
-                status: "pending",
-            });
-            seenPatterns.add(normalized);
-        }
-        for (const pattern of corpus.recurringPatterns) {
-            if (pattern.occurrences < 2)
-                continue;
-            const type = this.patternToProposalType(pattern);
-            proposals.push({
-                id: `prop-${Date.now()}-${proposals.length}`,
-                type,
-                title: `Codify ${pattern.name} pattern`,
-                description: `${pattern.name} detected across ${pattern.occurrences} sessions (avg confidence: ${Math.round(pattern.avgConfidence * 100)}%). ${pattern.description}`,
-                evidence: pattern.evidence,
-                confidence: pattern.avgConfidence,
-                source: "recurring_pattern",
-                status: "pending",
-            });
-        }
-        const wrongTurns = corpus.allWrongTurns.slice(0, 2);
-        for (const wt of wrongTurns) {
-            const summary = wt.length > 50 ? `${wt.substring(0, 47)}...` : wt;
-            proposals.push({
-                id: `prop-${Date.now()}-${proposals.length}`,
-                type: "guard",
-                title: `Guard against: ${summary}`,
-                description: `Recurring wrong turn detected: ${wt}. Add a guard or validation to prevent this pattern.`,
-                evidence: [wt],
-                confidence: 0.7,
-                source: "wrong_turn",
-                status: "pending",
-            });
-        }
-        return proposals.sort((a, b) => b.confidence - a.confidence).slice(0, 3);
-    }
-    adjustFromHistory(proposals) {
-        this.adjustConfidenceFromHistory(proposals);
-        proposals.sort((a, b) => b.confidence - a.confidence);
-    }
-    getCoordinator() {
-        if (!InferenceCycle.votingCoordinator) {
-            const stateManager = this.getGovernanceStateManager();
-            InferenceCycle.votingCoordinator = new VotingCoordinator(stateManager);
-        }
-        return InferenceCycle.votingCoordinator;
-    }
-    async applyProposals(proposals) {
-        const applier = new ProposalApplier(this.projectRoot, async (p) => this.applyProposalWork(p), async (p, prUrl) => {
-            if (this.options.skipResearcherReview)
-                return "go";
-            return this.researcherReview(p, prUrl);
-        });
-        const results = await applier.applyProposals(proposals);
-        for (const r of results) {
-            const p = proposals.find(pr => pr.id === r.proposalId);
-            if (p)
-                p.status = r.success ? "applied" : "failed";
-        }
-    }
-    async applyProposalWork(p) {
-        let filesChanged = false;
-        if (p.type === "fix" || p.type === "refactor") {
-            filesChanged = await this.applyCodeChange(p);
-        }
-        else if (p.type === "guard") {
-            filesChanged = await this.applyGuard(p);
-        }
-        else if (p.type === "automate") {
-            filesChanged = await this.applyAutomation(p);
-        }
-        return filesChanged;
-    }
-    async applyCodeChange(p) {
-        const targetFiles = this.extractTargetFiles(p.evidence);
-        const prompt = [
-            `Apply approved inference proposal`,
-            ``,
-            `Type: ${p.type}`,
-            `Title: ${p.title}`,
-            `Description: ${p.description}`,
-            targetFiles.length > 0 ? `Target files: ${targetFiles.join(", ")}` : "No specific target files identified",
-            `Evidence: ${p.evidence.slice(0, 5).join("; ")}`,
-            `Confidence: ${(p.confidence * 100).toFixed(0)}%`,
-            ``,
-            `1. Read the relevant source files`,
-            `2. Apply the ${p.type} described above`,
-            `3. Make minimal, surgical changes`,
-            `4. If the change is unsafe or unclear, skip and explain`,
-        ].join("\n");
-        frameworkLogger.log("inference-cycle", "apply-invoking-agent", "info", {
-            proposalId: p.id,
-            proposalType: p.type,
-            targetFiles,
-        });
-        try {
-            let agentName = p.type === "refactor" ? "refactorer" : "code-reviewer";
-            // In pure MCP mode, use real skill server names so the orchestrator dispatches to actual MCP tools
-            if (process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true') {
-                agentName = p.type === "refactor" ? "refactoring-strategies" : "code-review";
-            }
-            await this.invokeAgentInternal(agentName, prompt);
-            return true;
-        }
-        catch (err) {
-            frameworkLogger.log("inference-cycle", "apply-agent-failed", "warning", {
-                proposalId: p.id,
-                error: String(err),
-            });
-            return false;
-        }
-    }
-    async applyGuard(p) {
-        const prompt = [
-            `Add guard/validation for the following issue:`,
-            ``,
-            `Title: ${p.title}`,
-            `Description: ${p.description}`,
-            `Evidence: ${p.evidence.join("; ")}`,
-            `Confidence: ${(p.confidence * 100).toFixed(0)}%`,
-            ``,
-            `1. Read the relevant source files`,
-            `2. Add the missing guard, validation, or edge case handling`,
-            `3. If this is a codex rule, add the term to .xray/codex.json`,
-            `4. Make minimal, surgical changes`,
-        ].join("\n");
-        try {
-            const agentName = process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true'
-                ? "code-review"
-                : "code-reviewer";
-            await this.invokeAgentInternal(agentName, prompt);
-            return true;
-        }
-        catch (err) {
-            frameworkLogger.log("inference-cycle", "apply-guard-failed", "warning", {
-                proposalId: p.id,
-                error: String(err),
-            });
-            const guardPath = path.join(this.projectRoot, "docs", "guards", `${p.title.replace(/[^a-z0-9]/gi, "-")}.md`);
-            fs.mkdirSync(path.dirname(guardPath), { recursive: true });
-            fs.writeFileSync(guardPath, `# Guard: ${p.title}\n\n${p.description}\n\n## Evidence\n${p.evidence.join("\n")}`);
-            return true;
-        }
-    }
-    async applyAutomation(p) {
-        const prompt = [
-            `Design automation for the following manual process:`,
-            ``,
-            `Title: ${p.title}`,
-            `Description: ${p.description}`,
-            `Evidence: ${p.evidence.join("; ")}`,
-            `Confidence: ${(p.confidence * 100).toFixed(0)}%`,
-            ``,
-            `1. Read the relevant source files`,
-            `2. Design the automation (script, processor, or config change)`,
-            `3. Implement it if straightforward, otherwise describe the design`,
-        ].join("\n");
-        try {
-            const agentName = process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true'
-                ? "architecture-patterns"
-                : "architect";
-            await this.invokeAgentInternal(agentName, prompt);
-            return true;
-        }
-        catch (err) {
-            frameworkLogger.log("inference-cycle", "apply-automation-failed", "warning", {
-                proposalId: p.id,
-                error: String(err),
-            });
-            const automationPath = path.join(this.projectRoot, "docs", "automation-proposals.md");
-            const entry = `\n## ${p.title}\n\n${p.description}\n\n**Evidence:** ${p.evidence.join(", ")}\n`;
-            fs.mkdirSync(path.dirname(automationPath), { recursive: true });
-            fs.appendFileSync(automationPath, entry);
-            return true;
-        }
-    }
-    extractTargetFiles(evidence) {
-        const filePattern = /[a-zA-Z0-9/_-]+\.(ts|js|mjs|json|yml|yaml)/g;
-        const files = new Set();
-        for (const item of evidence) {
-            const matches = item.matchAll(filePattern);
-            for (const match of matches) {
-                const f = match[0];
-                if (f.startsWith("src/") || f.startsWith("dist/") || f.startsWith(".opencode/")) {
-                    files.add(f);
-                }
-            }
-        }
-        return [...files];
-    }
-    async researcherReview(p, prUrl) {
-        const prompt = `You are a researcher agent reviewing a PR for proposal: "${p.title}".
-
-PR URL: ${prUrl}
-
-Proposal Type: ${p.type}
-Description: ${p.description}
-Evidence: ${p.evidence.slice(0, 5).join("; ")}
-
-Your job: Review the actual codebase to verify this proposal makes sense. Search the code to confirm:
-1. Does the problem described actually exist in the codebase?
-2. Is the proposed solution appropriate?
-3. Are there any missed edge cases?
-
-Respond with EXACTLY one of:
-- GO (proposal is valid, approve)
-- NO-GO (proposal is invalid, reject)
-- MODIFY: <specific changes needed>`;
-        try {
-            const result = await this.invokeAgentInternal("researcher", prompt);
-            const output = result.toLowerCase();
-            if (output.includes("no-go"))
-                return "no-go";
-            if (output.includes("modify"))
-                return "modify";
-            return "go";
-        }
-        catch (err) {
-            frameworkLogger.log("inference-cycle", "researcher-review-failed", "warning", { error: String(err) });
-            return "go";
-        }
-    }
     async governProposals(proposals) {
-        // Primary path: Use the first-class Governance MCP (real skill servers + required Dynamo)
-        // This is the clean, centralized path (governance.server.ts + GovernanceService)
+        // Governance MCP is the sole governance path (governance.server.ts + GovernanceService).
+        // It calls individual skill MCPs (code-review, security-audit, researcher) internally
+        // and integrates with external Dynamo Solar SSOT.
         const useGovernanceMcp = process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true' ||
             this.isGovernanceMcpPreferred();
-        if (useGovernanceMcp) {
-            try {
-                const result = await Promise.race([
-                    mcpClientManager.callServerTool("governance", "govern_proposals", {
-                        proposals: proposals.map(p => ({
-                            id: p.id,
-                            type: p.type,
-                            title: p.title,
-                            description: p.description,
-                            evidence: p.evidence || [],
-                            source: p.source || "inference",
-                            confidence: p.confidence || 0.8,
-                        })),
-                        context: { source: "inference-cycle" },
-                        options: { require_external: true },
-                    }),
-                    new Promise((_, reject) => setTimeout(() => reject(new Error("Governance MCP timed out after 8s")), 8000)),
-                ]);
-                const text = result?.content?.[0]?.text || "";
-                const parsed = this.parseGovernanceMcpResponse(text, proposals);
-                frameworkLogger.log("inference-cycle", "governance-mcp-primary-path", "info", {
-                    proposalCount: proposals.length,
-                    overall: parsed.overallDecision,
-                });
-                return parsed.votes;
-            }
-            catch (err) {
-                frameworkLogger.log("inference-cycle", "governance-mcp-failed", "error", {
-                    error: err instanceof Error ? err.message : String(err),
-                });
-                // In forced pure MCP mode we must not silently fall back
-                if (process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true') {
-                    throw err;
-                }
-                // In normal mode, fall back to legacy path with deprecation warning.
-                // The legacy path will be removed once all consumers have migrated to the Governance MCP + Dynamo Solar SSOT model.
-                frameworkLogger.log("inference-cycle", "governance-legacy-fallback", "warning", {
-                    message: "Falling back to legacy governance path. This path is deprecated and will be removed in a future version.",
-                });
-            }
-        }
-        // Legacy internal path (deprecated)
-        const internalVotes = await this.governProposalsInternal(proposals);
-        const governanceIntegration = getGovernanceIntegration();
-        if (governanceIntegration?.isAvailable()) {
-            const externalVotes = await this.governProposalsExternal(proposals);
-            return this.mergeGovernanceVotes(internalVotes, externalVotes, proposals);
-        }
-        return internalVotes;
+        if (!useGovernanceMcp) {
+            throw new Error("Governance MCP is required but not available");
+        }
+        const result = await Promise.race([
+            mcpClientManager.callServerTool("governance", "govern_proposals", {
+                proposals: proposals.map(p => ({
+                    id: p.id,
+                    type: p.type,
+                    title: p.title,
+                    description: p.description,
+                    evidence: p.evidence || [],
+                    source: p.source || "inference",
+                    confidence: p.confidence || 0.8,
+                })),
+                context: { source: "inference-cycle" },
+                options: { require_external: true },
+            }),
+            new Promise((_, reject) => setTimeout(() => reject(new Error("Governance MCP timed out after 8s")), 8000)),
+        ]);
+        const text = result?.content?.[0]?.text || "";
+        const parsed = this.parseGovernanceMcpResponse(text, proposals);
+        frameworkLogger.log("inference-cycle", "governance-mcp-primary-path", "info", {
+            proposalCount: proposals.length,
+            overall: parsed.overallDecision,
+        });
+        return parsed.votes;
     }
     isGovernanceMcpPreferred() {
         try {
@@ -547,415 +271,6 @@ Respond with EXACTLY one of:
             return { votes, overallDecision: "needs_revision" };
         }
     }
-    /**
-     * Oscillator 1: Internal VotingCoordinator-based governance
-     */
-    async governProposalsInternal(proposals) {
-        // Defensive: if pure MCP mode is forced, use individual skill servers only
-        if (process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true') {
-            return this.governProposalsWithIndividualSkills(proposals);
-        }
-        const coordinator = this.getCoordinator();
-        const sessionId = `inference-governance-${Date.now()}`;
-        const results = [];
-        const todoPrompt = [
-            `Vote on ${proposals.length} inference proposals. Output EXACTLY one PROPOSAL block per proposal.`,
-            ``,
-            `FORMAT (exact, no extra text, no markdown):`,
-            `PROPOSAL: <number>`,
-            `  AGENT: architect`,
-            `  DECISION: approve|reject|abstain`,
-            `  CONFIDENCE: 0.XX`,
-            `  REASONING: <brief reason>`,
-            ``,
-            `Example:`,
-            `PROPOSAL: 1`,
-            `  AGENT: architect`,
-            `  DECISION: approve`,
-            `  CONFIDENCE: 0.85`,
-            `  REASONING: Cleanup reduces maintenance burden`,
-            ``,
-            `PROPOSALS:`,
-        ];
-        for (let i = 0; i < proposals.length; i++) {
-            const p = proposals[i];
-            todoPrompt.push(`${i + 1}. [${p.type}] "${p.title}"`);
-        }
-        try {
-            const jsonOutput = await this.invokeAgentInternal("architect", todoPrompt.join("\n"));
-            const allVotes = this.parseSubagentVotes(jsonOutput, proposals);
-            for (const proposal of proposals) {
-                const agents = GOVERNANCE_AGENTS[proposal.type] ?? ["code-reviewer"];
-                const participants = ["architect", ...agents];
-                const voteId = await coordinator.initiateVoting(sessionId, proposal.title, proposal.description, participants, {
-                    complexity: Math.min(50, 10 + proposal.evidence.length * 5 + (proposal.confidence > 0.8 ? 10 : 0)),
-                    riskLevel: proposal.type === "fix" ? "low" : proposal.type === "guard" ? "high" : "medium",
-                    hasSecurityConcerns: proposal.type === "guard" || proposal.type === "fix",
-                    hasArchitecturalImpact: proposal.type === "codify" || proposal.type === "automate",
-                    participantCount: participants.length,
-                });
-                const proposalVotes = allVotes.filter((v) => v.proposalId === proposal.id);
-                for (const v of proposalVotes) {
-                    coordinator.submitVote(voteId, v.agentName, v.decision, v.confidence, v.reasoning);
-                }
-                const resolved = coordinator.resolveVoting(voteId);
-                if (resolved) {
-                    results.push({
-                        proposalId: proposal.id,
-                        decision: resolved.decision === "approve" ? "approve" : "reject",
-                        confidence: resolved.confidence,
-                        details: resolved.details?.map((d) => `${d.agentName}: vote=${d.vote}, weight=${d.weight.toFixed(2)}`) || [],
-                    });
-                    if (resolved.details) {
-                        for (const detail of resolved.details) {
-                            this.getCoordinator().getAggregator().updateAgentPerformance(detail.agentName, resolved.decision, detail.vote, detail.vote === resolved.decision, proposal.confidence);
-                        }
-                    }
-                }
-                else {
-                    results.push(this.rejectNoQuorum(proposal, "agents did not vote (no quorum)"));
-                }
-            }
-        }
-        catch (error) {
-            frameworkLogger.log("inference-cycle", "architect-governance-failed", "error", { error: String(error) });
-            for (const proposal of proposals) {
-                results.push(this.rejectNoQuorum(proposal, `agent invocation failed: ${error instanceof Error ? error.message : String(error)}`));
-            }
-        }
-        const metrics = coordinator.getMetrics();
-        frameworkLogger.log("inference-cycle", "governance-metrics", "info", {
-            totalVotes: metrics.totalVotes,
-            avgConfidence: metrics.averageConfidence.toFixed(2),
-            strategyUsage: metrics.strategyUsage,
-        });
-        return results;
-    }
-    /**
-     * Pure individual knowledge-skill MCP path for governance.
-     * Used when XRAY_FORCE_MCP_GOVERNANCE=true.
-     * Each proposal is evaluated directly by the relevant skill servers using analyze_proposal.
-     */
-    async governProposalsWithIndividualSkills(proposals) {
-        const results = [];
-        const GOVERNANCE_AGENTS = {
-            fix: ["code-review", "security-audit", "researcher"],
-            refactor: ["code-review", "security-audit", "researcher"],
-            guard: ["code-review", "security-audit", "researcher"],
-            automate: ["code-review", "security-audit", "researcher"],
-            codify: ["code-review", "security-audit", "researcher"],
-        };
-        for (const proposal of proposals) {
-            const agents = GOVERNANCE_AGENTS[proposal.type] ?? ["code-review", "security-audit"];
-            const skillVotes = [];
-            for (const agent of agents) {
-                try {
-                    const skillResult = await mcpClientManager.callServerTool(agent, "analyze_proposal", {
-                        proposalTitle: proposal.title,
-                        proposalDescription: proposal.description,
-                        evidence: proposal.evidence,
-                        proposalType: proposal.type,
-                    });
-                    let structured = "";
-                    const contents = skillResult?.content || [];
-                    for (const c of contents) {
-                        if (c?.text && c.text.includes("DECISION:")) {
-                            structured = c.text.trim();
-                            break;
-                        }
-                    }
-                    if (!structured) {
-                        const full = JSON.stringify(skillResult);
-                        const m = full.match(/DECISION:\s*(approve|reject|abstain)[\s\S]{0,300}?REASONING:[^\n"]*/i);
-                        if (m)
-                            structured = m[0].trim();
-                    }
-                    skillVotes.push({
-                        agent,
-                        toolUsed: "analyze_proposal",
-                        rawResponse: structured || JSON.stringify(skillResult),
-                        structuredVote: structured || null,
-                    });
-                }
-                catch (err) {
-                    skillVotes.push({
-                        agent,
-                        toolUsed: "analyze_proposal",
-                        rawResponse: `error: ${err}`,
-                        structuredVote: null,
-                    });
-                }
-            }
-            const approves = skillVotes.filter((v) => v.structuredVote && v.structuredVote.includes("DECISION: approve")).length;
-            const rejects = skillVotes.filter((v) => v.structuredVote && v.structuredVote.includes("DECISION: reject")).length;
-            const decision = approves > rejects ? "approve" : (rejects > approves ? "reject" : "abstain");
-            let avgConf = 0.75;
-            const confMatches = skillVotes.map((v) => {
-                if (!v.structuredVote)
-                    return 0.75;
-                const m = v.structuredVote.match(/CONFIDENCE:\s*([0-9.]+)/);
-                return m ? parseFloat(m[1]) : 0.75;
-            });
-            if (confMatches.length > 0)
-                avgConf = confMatches.reduce((a, b) => a + b, 0) / confMatches.length;
-            results.push({
-                proposalId: proposal.id,
-                decision: decision,
-                confidence: Math.round(avgConf * 100) / 100,
-                details: skillVotes.map((v) => `${v.agent}: ${v.structuredVote?.split('\n')[0] || 'no structured vote'}`),
-            });
-        }
-        return results;
-    }
-    /**
-     * Merge internal and external governance votes.
-     * A proposal passes both oscillators — must be approved by internal AND external.
-     */
-    mergeGovernanceVotes(internalVotes, externalVotes, proposals) {
-        const merged = [];
-        for (const proposal of proposals) {
-            const internal = internalVotes.find((v) => v.proposalId === proposal.id);
-            const external = externalVotes.find((v) => v.proposalId === proposal.id);
-            if (!internal) {
-                merged.push(this.rejectNoQuorum(proposal, "internal governance vote missing"));
-                continue;
-            }
-            const internalApproved = internal.decision === "approve";
-            const externalApproved = external?.decision === "approve";
-            const bothApproved = internalApproved && (!external || externalApproved);
-            merged.push({
-                proposalId: proposal.id,
-                decision: bothApproved ? "approve" : "reject",
-                confidence: internal.confidence * (external ? external.confidence : 1.0),
-                details: [
-                    ...internal.details,
-                    ...(external ? [`External governance: ${external.decision} (conf: ${external.confidence.toFixed(2)})`, ...external.details] : []),
-                ],
-            });
-        }
-        return merged;
-    }
-    // Removed: buildOrchestratorGovernancePrompt (replaced by inline TODO prompt in governProposals)
-    /**
-     * Govern proposals using external chrono-warp-drive Dynamo endpoint
-     */
-    async governProposalsExternal(proposals) {
-        const governanceIntegration = getGovernanceIntegration();
-        if (!governanceIntegration) {
-            throw new Error("Governance integration not available");
-        }
-        // Build agent reviews from proposal evidence
-        const agentReviews = proposals.flatMap((p) => p.evidence.slice(0, 2).map((e) => `[${p.type}] ${e}`));
-        try {
-            const batchResult = await governanceIntegration.checkProposals(proposals, agentReviews, []);
-            const votes = batchResult.results.map((result) => ({
-                proposalId: result.governanceResponse.proposalId,
-                decision: result.vote.toLowerCase() === "yes" ? "approve" : "reject",
-                confidence: result.governanceResponse.confidence,
-                details: [
-                    `Governance: ${result.governanceResponse.recommendation}`,
-                    `Isotope: ${result.governanceResponse.governanceIsotopeId}`,
-                    ...result.governanceResponse.reasons.slice(0, 2),
-                ],
-            }));
-            frameworkLogger.log("inference-cycle", "external-governance-complete", "info", {
-                proposalCount: proposals.length,
-                passedCount: batchResult.results.filter((r) => r.passed).length,
-            });
-            return votes;
-        }
-        catch (error) {
-            frameworkLogger.log("inference-cycle", "external-governance-failed", "error", {
-                error: error instanceof Error ? error.message : String(error),
-            });
-            return proposals.map((p) => this.rejectNoQuorum(p, "external governance endpoint failed"));
-        }
-    }
-    async invokeAgentInternal(agentName, prompt) {
-        frameworkLogger.log("inference-cycle", "invoke-agent-internal", "info", {
-            agentName,
-            promptLength: prompt.length,
-        });
-        try {
-            const { mcpClientManager } = await import("../mcps/mcp-client.js");
-            const MCP_TIMEOUT_MS = 8000;
-            const result = await Promise.race([
-                mcpClientManager.callServerTool("orchestrator", "orchestrate-task", {
-                    description: prompt,
-                    tasks: [{
-                            id: `task-${Date.now()}`,
-                            description: prompt,
-                            type: agentName,
-                            priority: "high",
-                        }],
-                    executionMode: "sequential",
-                }),
-                new Promise((_, reject) => setTimeout(() => reject(new Error(`Orchestrator MCP timed out after ${MCP_TIMEOUT_MS}ms`)), MCP_TIMEOUT_MS)),
-            ]);
-            const content = result.content;
-            let responseText = "";
-            if (content && Array.isArray(content)) {
-                responseText = content.map((c) => c.text ?? "").join("");
-            }
-            else {
-                responseText = JSON.stringify(result);
-            }
-            // In pure MCP governance mode, trust the orchestrator response (it now does real work)
-            const isPureMcp = process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true';
-            const hasRealContent = /PROPOSAL:\s*\d+/i.test(responseText) ||
-                /DECISION:\s*(approve|reject|abstain)/i.test(responseText) ||
-                /Agent Outputs \(real MCP responses\):/i.test(responseText);
-            if (hasRealContent || isPureMcp) {
-                if (isPureMcp && !hasRealContent) {
-                    frameworkLogger.log("inference-cycle", "pure-mcp-orchestrator-response", "info", {
-                        agentName,
-                        responsePreview: responseText.substring(0, 300),
-                    });
-                }
-                return responseText;
-            }
-            frameworkLogger.log("inference-cycle", "mcp-no-useful-content", "info", {
-                agentName,
-                responsePreview: responseText.substring(0, 200),
-            });
-        }
-        catch (mcpError) {
-            frameworkLogger.log("inference-cycle", "mcp-invocation-failed", "info", {
-                agentName,
-                error: String(mcpError),
-            });
-        }
-        if (this.agentInvoker) {
-            frameworkLogger.log("inference-cycle", "invoke-via-callback", "info", { agentName });
-            return this.agentInvoker(agentName, prompt);
-        }
-        // Only fall back to OpenCode if not in forced pure MCP mode
-        if (process.env.XRAY_FORCE_MCP_GOVERNANCE === 'true') {
-            throw new Error(`[PURE MCP] Orchestrator returned no usable response for agent "${agentName}" and OpenCode fallback is disabled`);
-        }
-        return this.invokeViaOpencode(agentName, prompt);
-    }
-    async invokeViaOpencode(agentName, prompt) {
-        return invokeOpencodeFromEngine(agentName, prompt, this.projectRoot);
-    }
-    parseSubagentVotes(jsonOutput, proposals) {
-        const votes = [];
-        const lines = jsonOutput.split("\n").filter((l) => l.trim());
-        for (const line of lines) {
-            try {
-                const obj = JSON.parse(line);
-                if (obj.type === "tool_use" && obj.part?.type === "tool" && obj.part.tool === "task") {
-                    const agentName = obj.part.state?.input?.subagent_type || "";
-                    const output = obj.part.state?.output || "";
-                    // Each task output may contain multiple PROPOSAL blocks
-                    const blocks = output.split(/PROPOSAL:\s*/).filter((b) => b.trim().length > 0);
-                    for (const block of blocks) {
-                        const numMatch = block.match(/^(\d+)/);
-                        if (!numMatch)
-                            continue;
-                        const proposalIdx = parseInt(numMatch[1], 10) - 1;
-                        if (proposalIdx < 0 || proposalIdx >= proposals.length)
-                            continue;
-                        const decisionMatch = block.match(/DECISION:\s*(\w+)/i);
-                        const confMatch = block.match(/CONFIDENCE:\s*(0?\.\d+|1\.0|1|0)/i);
-                        const reasonMatch = block.match(/REASONING:\s*(.+)/i);
-                        if (decisionMatch) {
-                            const proposal = proposals[proposalIdx];
-                            if (proposal) {
-                                votes.push({
-                                    proposalId: proposal.id,
-                                    agentName,
-                                    decision: decisionMatch[1].toLowerCase(),
-                                    confidence: confMatch ? Math.min(1, Math.max(0, parseFloat(confMatch[1]))) : 0.5,
-                                    reasoning: reasonMatch ? reasonMatch[1].trim() : "",
-                                });
-                            }
-                        }
-                    }
-                }
-            }
-            catch {
-                // Not JSON, skip
-            }
-        }
-        // Fallback: if no votes found from JSON format, try parsing plain-text
-        // PROPOSAL/DECISION blocks directly (e.g. from opencode CLI fallback path)
-        if (votes.length === 0 && /PROPOSAL:\s*\d+/i.test(jsonOutput)) {
-            const blocks = jsonOutput.split(/PROPOSAL:\s*/).filter((b) => b.trim().length > 0);
-            for (const block of blocks) {
-                const numMatch = block.match(/^(\d+)/);
-                if (!numMatch)
-                    continue;
-                const numStr = numMatch[1];
-                if (!numStr)
-                    continue;
-                const proposalIdx = parseInt(numStr, 10) - 1;
-                if (proposalIdx < 0 || proposalIdx >= proposals.length)
-                    continue;
-                const agentMatch = block.match(/AGENT:\s*(\w[\w-]*)/i);
-                const decisionMatch = block.match(/DECISION:\s*(\w+)/i);
-                const confMatch = block.match(/CONFIDENCE:\s*(0?\.\d+|1\.0|1|0)/i);
-                const reasonMatch = block.match(/REASONING:\s*(.+)/i);
-                if (decisionMatch) {
-                    const proposal = proposals[proposalIdx];
-                    if (proposal) {
-                        let decision = decisionMatch[1].toLowerCase();
-                        if (decision === "yes")
-                            decision = "approve";
-                        if (decision === "no")
-                            decision = "reject";
-                        votes.push({
-                            proposalId: proposal.id,
-                            agentName: agentMatch?.[1] ?? "architect",
-                            decision,
-                            confidence: confMatch ? Math.min(1, Math.max(0, parseFloat(confMatch[1]))) : 0.5,
-                            reasoning: reasonMatch ? reasonMatch[1].trim() : "",
-                        });
-                    }
-                }
-            }
-        }
-        return votes;
-    }
-    getGovernanceStateManager() {
-        if (!fs.existsSync(this.stateDir)) {
-            fs.mkdirSync(this.stateDir, { recursive: true });
-        }
-        const stateFile = path.join(this.stateDir, "governance-state.json");
-        const stateManager = new XrayStateManager();
-        if (fs.existsSync(stateFile)) {
-            try {
-                const data = JSON.parse(fs.readFileSync(stateFile, "utf-8"));
-                for (const [key, value] of Object.entries(data)) {
-                    stateManager.set(key, value);
-                }
-            }
-            catch {
-                frameworkLogger.log("inference-cycle", "governance-state-load-failed", "warning", {});
-            }
-        }
-        return stateManager;
-    }
-    saveGovernanceState(coordinator) {
-        const stateFile = path.join(this.stateDir, "governance-state.json");
-        try {
-            // Export the coordinator's internal state (voting history, metrics, etc.)
-            const history = coordinator.getVotingHistory();
-            const metrics = coordinator.getMetrics();
-            const exportData = {
-                votingHistory: history,
-                metrics,
-                exportedAt: new Date().toISOString(),
-            };
-            if (!fs.existsSync(this.stateDir)) {
-                fs.mkdirSync(this.stateDir, { recursive: true });
-            }
-            fs.writeFileSync(stateFile, JSON.stringify(exportData, null, 2));
-        }
-        catch (error) {
-            frameworkLogger.log("inference-cycle", "governance-state-save-failed", "warning", { error: String(error) });
-        }
-    }
     extractTextFromNdjson(output) {
         const texts = [];
         for (const line of output.split("\n")) {
@@ -975,13 +290,10 @@ Respond with EXACTLY one of:
         return texts.join("\n").trim();
     }
     resolveOpencodeRoot() {
-        // Use the provider-agnostic config path resolver (prefers .xray/)
         const configDir = getConfigDir(this.projectRoot);
-        // If we resolved to a .xray or custom dir, use its parent as the "root"
         if (configDir.includes(".xray") || configDir.includes("xray")) {
             return path.dirname(configDir);
         }
-        // Legacy fallback
         let dir = this.projectRoot;
         for (let i = 0; i < 10; i++) {
             if (fs.existsSync(path.join(dir, ".opencode")))
@@ -996,61 +308,6 @@ Respond with EXACTLY one of:
             return cwd;
         return this.projectRoot;
     }
-    rejectNoQuorum(proposal, reason) {
-        return {
-            proposalId: proposal.id,
-            decision: "reject",
-            confidence: 0,
-            details: [`rejected: ${reason}`],
-        };
-    }
-    classifyProposalType(problemPattern) {
-        const lower = problemPattern.toLowerCase();
-        if (lower.includes("bug") || lower.includes("fix") || lower.includes("stability"))
-            return "fix";
-        if (lower.includes("dead code") || lower.includes("remove") || lower.includes("health"))
-            return "refactor";
-        if (lower.includes("manual") || lower.includes("automate"))
-            return "automate";
-        if (lower.includes("guard") || lower.includes("path") || lower.includes("timing") || lower.includes("edge case"))
-            return "guard";
-        return "codify";
-    }
-    patternToProposalType(pattern) {
-        const name = pattern.name.toLowerCase();
-        if (name.includes("dead code"))
-            return "refactor";
-        if (name.includes("extract"))
-            return "refactor";
-        if (name.includes("registry") || name.includes("facade"))
-            return "codify";
-        if (name.includes("test"))
-            return "guard";
-        if (name.includes("stability"))
-            return "fix";
-        return "codify";
-    }
-    generateTitle(problem) {
-        const pattern = problem.pattern;
-        const ACTION_MAP = [
-            [/^Bug fix$/i, "Fix recurring bug pattern"],
-            [/^Code health cleanup$/i, "Clean up code health issues"],
-            [/^Accumulated dead code$/i, "Remove accumulated dead code"],
-            [/^Incomplete implementation$/i, "Complete partial implementation"],
-            [/^Technical debt/i, "Address technical debt"],
-            [/^Missing guard/i, "Add missing guard"],
-            [/^Manual step/i, "Automate manual step"],
-        ];
-        for (const [re, title] of ACTION_MAP) {
-            if (re.test(pattern)) {
-                return `${title} (${problem.occurrences}x across ${problem.sessions.length} sessions)`;
-            }
-        }
-        if (pattern.length > 80) {
-            return `Address: ${pattern.substring(0, 77)}... (${problem.occurrences}x)`;
-        }
-        return `Address: ${pattern} (${problem.occurrences}x)`;
-    }
     setPhase(phase) {
         this.phase = phase;
         frameworkLogger.log("inference-cycle", "phase-change", "info", { phase });
@@ -1098,39 +355,6 @@ Respond with EXACTLY one of:
             return [];
         }
     }
-    adjustConfidenceFromHistory(proposals) {
-        const history = this.loadHistory();
-        if (history.length === 0)
-            return;
-        const recentVotes = history.slice(-10).flatMap((h) => h.votes);
-        const approvedIds = new Set(recentVotes.filter((v) => v.decision === "approve").map((v) => v.proposalId));
-        const approvedTypes = new Map();
-        const rejectedTypes = new Map();
-        for (const h of history.slice(-10)) {
-            for (const p of h.proposals) {
-                if (p.status === "approved" || p.status === "applied") {
-                    approvedTypes.set(p.type, (approvedTypes.get(p.type) ?? 0) + 1);
-                }
-                else if (p.status === "rejected" || p.status === "failed") {
-                    rejectedTypes.set(p.type, (rejectedTypes.get(p.type) ?? 0) + 1);
-                }
-            }
-        }
-        for (const proposal of proposals) {
-            const approved = approvedTypes.get(proposal.type) ?? 0;
-            const rejected = rejectedTypes.get(proposal.type) ?? 0;
-            const total = approved + rejected;
-            if (total >= 3) {
-                const successRate = approved / total;
-                if (successRate < 0.3) {
-                    proposal.confidence *= 0.8;
-                }
-                else if (successRate > 0.7) {
-                    proposal.confidence = Math.min(0.95, proposal.confidence * 1.05);
-                }
-            }
-        }
-    }
     buildBlockedResult(cycleId, reason) {
         return {
             cycleId,