@caupulican/pi-adaptative 0.81.38 → 0.81.39

package/dist/core/reflection-controller.js

@@ -12,11 +12,20 @@
 import { existsSync, mkdirSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 import { runAgentLoop, } from "@caupulican/pi-agent-core";
+import { computeLaneAffinityKey, DEFAULT_ISOLATED_LANE_KIND, } from "./agent-session.js";
 import { AUTONOMY_TELEMETRY_EVENT_TYPES } from "./autonomy/telemetry-events.js";
 import { APPLY_WRITE_REFUSED_REASON_CODE, appendLearningAuditSnapshot, contradictionsForReflectionWrite, getLearningAuditSnapshots, proposalFromReflectionWrite, rollbackPlanForReflectionWrite, } from "./learning/learning-audit.js";
 import { evaluateLearningDecision } from "./learning/learning-gate.js";
 import { ObservationStore, observationKey } from "./learning/observation-store.js";
 import { decideDemand, ReflectionEngine, } from "./learning/reflection-engine.js";
+import { registerInFlightWork } from "./reload-blockers.js";
+import { runSkillAudit } from "./tools/skill-audit.js";
+// reasonCode when an automatic skill promotion is blocked by an overlap with an existing skill
+// (skill_audit's similarity threshold) and routed to a consolidation proposal instead of a blind write.
+export const SKILL_OVERLAP_CONSOLIDATION_REASON_CODE = "skill_overlap_consolidation_proposed";
+// reasonCode when the overlap audit itself fails — the promotion is held (not written unaudited)
+// rather than silently skipping the check.
+export const SKILL_AUDIT_UNAVAILABLE_REASON_CODE = "skill_audit_unavailable";
 export class ReflectionController {
     deps;
     constructor(deps) {
@@ -27,9 +36,11 @@ export class ReflectionController {
      * reflection and bounded child lanes (adaptive-agent design §6c/§7).
      *
      * Isolation invariants (audited by codex): builds fresh context (no main history), defaults to no
-     * tools, and passes **no `sessionId`**. A tool-enabled call receives only caller-owned tools and
-     * hooks, and is turn-bounded. It cannot mutate `agent.state.messages`, append session entries, or
-     * touch the foreground tool registry. Mirrors `generateSummary()`'s one-shot mechanics otherwise.
+     * tools, and passes **no real `sessionId`** — only a deterministic SYNTHETIC cache-affinity key
+     * (see {@link computeLaneAffinityKey}) derived from `(laneKind, model, systemPrompt)`, which can never
+     * equal or embed the real session id. A tool-enabled call receives only caller-owned tools and hooks,
+     * and is turn-bounded. It cannot mutate `agent.state.messages`, append session entries, or touch the
+     * foreground tool registry. Mirrors `generateSummary()`'s one-shot mechanics otherwise.
      *
      * Returns the result even on an error/aborted stop reason (callers — e.g. a background reflection
      * microtask — decide whether to act); it does not throw on a model-level error.
@@ -41,136 +52,156 @@ export class ReflectionController {
         }
         await this.deps.ensureModelReady(model);
         const thinkingLevel = opts.thinkingLevel ?? "off";
-        // Fresh, isolated context: explicit messages, caller-owned tools only, nothing from the main session.
-        const context = {
-            systemPrompt: opts.systemPrompt,
-            messages: opts.messages,
-            tools: opts.tools ?? [],
-        };
-        // Isolate the prompt cache and DELIBERATELY omit sessionId so no session-aware caching/routing
-        // can entangle this call with the main session.
-        const options = {
-            maxTokens: opts.maxTokens,
-            signal: opts.signal,
-            cacheRetention: opts.cacheRetention ?? "none",
-            reasoning: thinkingLevel,
-        };
-        // When streamFn is the raw streamSimple (e.g. in tests), auth must be injected explicitly.
-        // Throw only when auth genuinely fails — providers that authenticate without an API key
-        // (OAuth, local no-key) legitimately return ok with an undefined apiKey.
-        if (this.deps.isRawStreamSimple()) {
-            const auth = await this.deps.getModelRegistry().getApiKeyAndHeaders(model);
-            if (!auth.ok) {
-                throw new Error(auth.error);
-            }
-            options.apiKey = auth.apiKey;
-            options.headers = auth.headers;
-        }
-        if (opts.tools && opts.tools.length > 0) {
-            const agent = this.deps.getAgent();
-            const foregroundModel = agent.state.model;
-            const usesForegroundModel = foregroundModel?.provider === model.provider && foregroundModel.id === model.id;
-            const textToolCallProtocol = this.deps.resolveTextToolCallProtocol(model);
-            const requestedMaxTurns = typeof opts.maxTurns === "number" && Number.isFinite(opts.maxTurns) ? Math.floor(opts.maxTurns) : 6;
-            const maxTurns = Math.max(1, Math.min(12, requestedMaxTurns));
-            let completedTurns = 0;
-            const childContext = {
+        // Registered for the full isolated-completion call (one-shot or child-loop) so the
+        // reload gate waits it out — this is the SINGLE choke point every isolated completion in the
+        // codebase runs through (reflection, research/worker/fitness lanes, context-pipeline curation,
+        // model-router judge calls). registerInFlightWork is a pure sync map op (cannot throw), so
+        // placing it as the last statement before `try` still guarantees the matching finally always
+        // runs, on every return path below and on any thrown error.
+        const deregisterInFlight = registerInFlightWork(this.deps.getAgentDir(), "isolated-completion", opts.laneKind ?? DEFAULT_ISOLATED_LANE_KIND);
+        try {
+            // Fresh, isolated context: explicit messages, caller-owned tools only, nothing from the main session.
+            const context = {
                 systemPrompt: opts.systemPrompt,
-                messages: [],
-                tools: [...opts.tools],
+                messages: opts.messages,
+                tools: opts.tools ?? [],
             };
-            const loopConfig = {
-                model,
+            // Isolate the prompt cache from the main session by DELIBERATELY never sending the real
+            // sessionId. In its place, a deterministic SYNTHETIC affinity key lets providers with
+            // session-affinity headers / prompt_cache_key route repeat calls from the SAME lane (same
+            // laneKind+model+systemPrompt) to the same cache-warm backend, without entangling this call
+            // with — or leaking any identity of — the main session.
+            const affinityKey = computeLaneAffinityKey(opts.laneKind ?? DEFAULT_ISOLATED_LANE_KIND, model, opts.systemPrompt);
+            const options = {
                 maxTokens: opts.maxTokens,
-                cacheRetention: opts.cacheRetention ?? "none",
+                signal: opts.signal,
+                cacheRetention: opts.cacheRetention,
                 reasoning: thinkingLevel,
-                ...(options.apiKey !== undefined ? { apiKey: options.apiKey } : {}),
-                ...(options.headers !== undefined ? { headers: options.headers } : {}),
-                temperature: textToolCallProtocol ? 0 : undefined,
-                textToolCallProtocol,
-                onTextToolProtocolParse: usesForegroundModel ? agent.onTextToolProtocolParse : undefined,
-                transport: agent.transport,
-                thinkingBudgets: agent.thinkingBudgets,
-                maxRetryDelayMs: agent.maxRetryDelayMs,
-                maxStallTurns: Math.max(2, Math.min(maxTurns, agent.maxStallTurns ?? maxTurns)),
-                toolExecution: "sequential",
-                toolArgumentTeachEnabled: agent.toolArgumentTeachEnabled,
-                onToolArgumentValidation: usesForegroundModel ? agent.onToolArgumentValidation : undefined,
-                toolValidationEscalationThreshold: agent.toolValidationEscalationThreshold,
-                onToolValidationEscalation: usesForegroundModel ? agent.onToolValidationEscalation : undefined,
-                beforeToolCall: opts.beforeToolCall,
-                afterToolCall: opts.afterToolCall,
-                shouldStopAfterTurn: () => {
-                    completedTurns += 1;
-                    return completedTurns >= maxTurns;
-                },
-                convertToLlm: agent.convertToLlm,
+                sessionId: affinityKey,
             };
-            const messages = await runAgentLoop(opts.messages, childContext, loopConfig, () => { }, opts.signal, agent.streamFn);
-            const assistantMessages = messages.filter((message) => message.role === "assistant");
-            let finalAssistant = assistantMessages.at(-1);
-            if (!finalAssistant) {
-                throw new Error("runIsolatedCompletion: child loop produced no assistant message");
+            // When streamFn is the raw streamSimple (e.g. in tests), auth must be injected explicitly.
+            // Throw only when auth genuinely fails — providers that authenticate without an API key
+            // (OAuth, local no-key) legitimately return ok with an undefined apiKey.
+            if (this.deps.isRawStreamSimple()) {
+                const auth = await this.deps.getModelRegistry().getApiKeyAndHeaders(model);
+                if (!auth.ok) {
+                    throw new Error(auth.error);
+                }
+                options.apiKey = auth.apiKey;
+                options.headers = auth.headers;
             }
-            const hasFinalText = finalAssistant.content.some((content) => content.type === "text" && content.text.trim().length > 0);
-            const endedOnToolCall = finalAssistant.content.some((content) => content.type === "toolCall");
-            if (opts.finalTextPrompt && !hasFinalText && endedOnToolCall && !opts.signal?.aborted) {
-                // A hard turn bound may stop immediately after successful tool execution. Preserve that bound,
-                // then allow exactly one tool-free synthesis call so the gathered work is not thrown away.
-                const finalizationMessages = await agent.convertToLlm(messages);
-                const finalizationStream = await agent.streamFn(model, {
+            if (opts.tools && opts.tools.length > 0) {
+                const agent = this.deps.getAgent();
+                const foregroundModel = agent.state.model;
+                const usesForegroundModel = foregroundModel?.provider === model.provider && foregroundModel.id === model.id;
+                const textToolCallProtocol = this.deps.resolveTextToolCallProtocol(model);
+                const requestedMaxTurns = typeof opts.maxTurns === "number" && Number.isFinite(opts.maxTurns) ? Math.floor(opts.maxTurns) : 6;
+                const maxTurns = Math.max(1, Math.min(12, requestedMaxTurns));
+                let completedTurns = 0;
+                const childContext = {
                     systemPrompt: opts.systemPrompt,
-                    messages: [
-                        ...finalizationMessages,
-                        { role: "user", content: opts.finalTextPrompt, timestamp: Date.now() },
-                    ],
-                    tools: [],
-                }, options);
-                finalAssistant = await finalizationStream.result();
-                assistantMessages.push(finalAssistant);
+                    messages: [],
+                    tools: [...opts.tools],
+                };
+                const loopConfig = {
+                    model,
+                    maxTokens: opts.maxTokens,
+                    cacheRetention: opts.cacheRetention,
+                    reasoning: thinkingLevel,
+                    // Same synthetic per-lane affinity key as the one-shot path above — never the real sessionId.
+                    sessionId: affinityKey,
+                    ...(options.apiKey !== undefined ? { apiKey: options.apiKey } : {}),
+                    ...(options.headers !== undefined ? { headers: options.headers } : {}),
+                    temperature: textToolCallProtocol ? 0 : undefined,
+                    textToolCallProtocol,
+                    onTextToolProtocolParse: usesForegroundModel ? agent.onTextToolProtocolParse : undefined,
+                    transport: agent.transport,
+                    thinkingBudgets: agent.thinkingBudgets,
+                    maxRetryDelayMs: agent.maxRetryDelayMs,
+                    maxStallTurns: Math.max(2, Math.min(maxTurns, agent.maxStallTurns ?? maxTurns)),
+                    onRunawayStop: usesForegroundModel ? agent.onRunawayStop : undefined,
+                    toolExecution: "sequential",
+                    toolArgumentTeachEnabled: agent.toolArgumentTeachEnabled,
+                    onToolArgumentValidation: usesForegroundModel ? agent.onToolArgumentValidation : undefined,
+                    toolValidationEscalationThreshold: agent.toolValidationEscalationThreshold,
+                    onToolValidationEscalation: usesForegroundModel ? agent.onToolValidationEscalation : undefined,
+                    beforeToolCall: opts.beforeToolCall,
+                    afterToolCall: opts.afterToolCall,
+                    shouldStopAfterTurn: () => {
+                        completedTurns += 1;
+                        return completedTurns >= maxTurns;
+                    },
+                    convertToLlm: agent.convertToLlm,
+                };
+                const messages = await runAgentLoop(opts.messages, childContext, loopConfig, () => { }, opts.signal, agent.streamFn);
+                const assistantMessages = messages.filter((message) => message.role === "assistant");
+                let finalAssistant = assistantMessages.at(-1);
+                if (!finalAssistant) {
+                    throw new Error("runIsolatedCompletion: child loop produced no assistant message");
+                }
+                const hasFinalText = finalAssistant.content.some((content) => content.type === "text" && content.text.trim().length > 0);
+                const endedOnToolCall = finalAssistant.content.some((content) => content.type === "toolCall");
+                if (opts.finalTextPrompt && !hasFinalText && endedOnToolCall && !opts.signal?.aborted) {
+                    // A hard turn bound may stop immediately after successful tool execution. Preserve that bound,
+                    // then allow exactly one tool-free synthesis call so the gathered work is not thrown away.
+                    const finalizationMessages = await agent.convertToLlm(messages);
+                    const finalizationStream = await agent.streamFn(model, {
+                        systemPrompt: opts.systemPrompt,
+                        messages: [
+                            ...finalizationMessages,
+                            { role: "user", content: opts.finalTextPrompt, timestamp: Date.now() },
+                        ],
+                        tools: [],
+                    }, options);
+                    finalAssistant = await finalizationStream.result();
+                    assistantMessages.push(finalAssistant);
+                }
+                const usage = assistantMessages.reduce((total, message) => ({
+                    input: total.input + message.usage.input,
+                    output: total.output + message.usage.output,
+                    cacheRead: total.cacheRead + message.usage.cacheRead,
+                    cacheWrite: total.cacheWrite + message.usage.cacheWrite,
+                    totalTokens: total.totalTokens + message.usage.totalTokens,
+                    cost: {
+                        input: total.cost.input + message.usage.cost.input,
+                        output: total.cost.output + message.usage.cost.output,
+                        cacheRead: total.cost.cacheRead + message.usage.cost.cacheRead,
+                        cacheWrite: total.cost.cacheWrite + message.usage.cost.cacheWrite,
+                        total: total.cost.total + message.usage.cost.total,
+                    },
+                }), {
+                    input: 0,
+                    output: 0,
+                    cacheRead: 0,
+                    cacheWrite: 0,
+                    totalTokens: 0,
+                    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+                });
+                const text = finalAssistant.content
+                    .filter((content) => content.type === "text")
+                    .map((content) => content.text)
+                    .join("");
+                return { text, usage, stopReason: finalAssistant.stopReason };
             }
-            const usage = assistantMessages.reduce((total, message) => ({
-                input: total.input + message.usage.input,
-                output: total.output + message.usage.output,
-                cacheRead: total.cacheRead + message.usage.cacheRead,
-                cacheWrite: total.cacheWrite + message.usage.cacheWrite,
-                totalTokens: total.totalTokens + message.usage.totalTokens,
-                cost: {
-                    input: total.cost.input + message.usage.cost.input,
-                    output: total.cost.output + message.usage.cost.output,
-                    cacheRead: total.cost.cacheRead + message.usage.cost.cacheRead,
-                    cacheWrite: total.cost.cacheWrite + message.usage.cost.cacheWrite,
-                    total: total.cost.total + message.usage.cost.total,
-                },
-            }), {
+            const stream = await this.deps.getAgent().streamFn(model, context, options);
+            const result = await stream.result();
+            const text = result.content
+                .filter((c) => c.type === "text")
+                .map((c) => c.text)
+                .join("");
+            const usage = result.usage ?? {
                 input: 0,
                 output: 0,
                 cacheRead: 0,
                 cacheWrite: 0,
                 totalTokens: 0,
                 cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
-            });
-            const text = finalAssistant.content
-                .filter((content) => content.type === "text")
-                .map((content) => content.text)
-                .join("");
-            return { text, usage, stopReason: finalAssistant.stopReason };
+            };
+            return { text, usage, stopReason: result.stopReason };
+        }
+        finally {
+            deregisterInFlight();
         }
-        const stream = await this.deps.getAgent().streamFn(model, context, options);
-        const result = await stream.result();
-        const text = result.content
-            .filter((c) => c.type === "text")
-            .map((c) => c.text)
-            .join("");
-        const usage = result.usage ?? {
-            input: 0,
-            output: 0,
-            cacheRead: 0,
-            cacheWrite: 0,
-            totalTokens: 0,
-            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
-        };
-        return { text, usage, stopReason: result.stopReason };
     }
     /**
      * Native end-of-loop reflection pass (R2). Demand-gates (zero-I/O), and when warranted runs the
@@ -202,6 +233,7 @@ export class ReflectionController {
             // The reflection system prompt is static (#33) — let the provider cache the prefix so
             // repeated passes only pay for the variable tail.
             cacheRetention: "short",
+            laneKind: "reflection",
         });
         const result = await new ReflectionEngine().reflect({
             recentTurnText: input.recentTurnText,
@@ -296,25 +328,43 @@ export class ReflectionController {
                     },
                 });
             }
+            // An automatic skill promotion the gate would otherwise apply must still clear the same
+            // skill_audit overlap check the model-invoked `skillify` tool enforces — otherwise reflection
+            // can silently write a near-duplicate SKILL.md. A non-overlapping skill is unaffected
+            // (skillPromotionBlock stays undefined) and promotes exactly as before.
+            const skillPromotionBlock = write.kind === "promote_skill" && decision.kind === "apply"
+                ? this._checkSkillPromotionOverlap(write)
+                : undefined;
             // The gate's decision and the write's actual outcome are two different questions: the memory
             // tool can refuse a write (budget exceeded, drift, threat) via details.success:false without
             // throwing. Capture that outcome instead of assuming "decision.kind === apply" means it landed
             // — otherwise a refused write leaves a phantom "apply" audit whose rollback later fails
             // not-found (or, worse, misfires against whatever now occupies that text).
-            const applied = decision.kind === "apply" ? await this._applyReflectionWrite(write, signal) : false;
-            const writeFailed = decision.kind === "apply" && !applied;
+            const applied = decision.kind === "apply" && !skillPromotionBlock ? await this._applyReflectionWrite(write, signal) : false;
+            const writeFailed = decision.kind === "apply" && !skillPromotionBlock && !applied;
             if (decision.kind !== "no-op") {
                 auditSequence += 1;
                 appendLearningAuditSnapshot(this.deps.getSessionManager(), {
                     id: `audit-${auditSequence}`,
                     proposalId,
                     layer: proposal.layer,
-                    action: writeFailed ? "apply_failed" : decision.kind === "apply" ? "apply" : "propose",
-                    summary: proposal.summary,
-                    reasonCode: writeFailed ? APPLY_WRITE_REFUSED_REASON_CODE : decision.reasonCode,
+                    action: skillPromotionBlock
+                        ? "propose"
+                        : writeFailed
+                            ? "apply_failed"
+                            : decision.kind === "apply"
+                                ? "apply"
+                                : "propose",
+                    summary: skillPromotionBlock ? `${proposal.summary} — ${skillPromotionBlock.note}` : proposal.summary,
+                    reasonCode: skillPromotionBlock
+                        ? skillPromotionBlock.reasonCode
+                        : writeFailed
+                            ? APPLY_WRITE_REFUSED_REASON_CODE
+                            : decision.reasonCode,
                     decision,
-                    // No rollback plan on a failed apply — nothing durable landed, so there is nothing to undo.
-                    rollback: writeFailed ? undefined : rollback,
+                    // No rollback plan on a failed apply or a held/proposed promotion — nothing durable
+                    // landed in either case, so there is nothing to undo.
+                    rollback: skillPromotionBlock || writeFailed ? undefined : rollback,
                     createdAt: new Date().toISOString(),
                 });
             }
@@ -466,9 +516,48 @@ export class ReflectionController {
         }
         return false;
     }
+    /**
+     * The same skill_audit overlap check the model-invoked `skillify` tool enforces
+     * (tools/skillify.ts → tools/skill-audit.ts's `runSkillAudit`), run before an AUTOMATIC promotion
+     * so the reflection engine can never silently write a near-duplicate SKILL.md. Reuses the audit
+     * seam unchanged (never duplicated) against the same already-loaded skills universe skillify
+     * compares a draft against. Returns a block reason when the draft overlaps an existing skill above
+     * the audit's own similarity threshold, or when the audit itself fails (the promotion is held, not
+     * written unaudited); undefined when clear to promote. Never throws — mirrors every other
+     * best-effort check in this file.
+     */
+    _checkSkillPromotionOverlap(write) {
+        try {
+            const cwd = this.deps.getCwd?.() ?? process.cwd();
+            const audit = runSkillAudit(cwd, write);
+            const overlap = audit.nearDuplicates.find((d) => d.a === "[draft]" || d.b === "[draft]");
+            if (!overlap)
+                return undefined;
+            const otherPath = overlap.a === "[draft]" ? overlap.b : overlap.a;
+            // Every skill file is named "SKILL.md" — unlike skillify's own display formatter (which takes
+            // the path's last segment and always shows that literal filename), look up the declared
+            // frontmatter name so the proposal actually names the skill it overlaps with.
+            const otherName = audit.skills.find((s) => s.path === otherPath)?.name ?? otherPath;
+            return {
+                reasonCode: SKILL_OVERLAP_CONSOLIDATION_REASON_CODE,
+                note: `overlaps existing skill "${otherName}" (${(overlap.similarity * 100).toFixed(0)}% similar) — consolidation proposed instead of a duplicate write`,
+            };
+        }
+        catch {
+            // Bounded, observable degradation via the promote path's existing error-reporting shape: an
+            // audit failure must never throw into the reflection pass, but it also must not silently
+            // fall through to an unaudited write — hold the promotion and record why.
+            return {
+                reasonCode: SKILL_AUDIT_UNAVAILABLE_REASON_CODE,
+                note: "skill overlap audit failed; promotion held rather than written unaudited",
+            };
+        }
+    }
     /**
      * R7: write a reflection-promoted skill as `<agentDir>/skills/<name>/SKILL.md` so it loads like any
-     * user skill. Best-effort; never clobbers an existing (hand-authored) skill of the same name.
+     * user skill. Best-effort; never clobbers an existing (hand-authored) skill of the same name. The
+     * overlap audit runs at the call site in {@link runReflectionPass} — a write only reaches here
+     * once the draft has already cleared it.
      */
     _promoteReflectionSkill(rawName, description, body) {
         const name = rawName