codeloop-mcp-server 0.1.76 → 0.1.78

package/dist/evidence/interaction_evidence.d.ts

@@ -0,0 +1,46 @@
+export interface InteractionEvidence {
+    generated_at: string;
+    /** What drove it: the deep-E2E executor. */
+    source: "run_journey";
+    target: string;
+    /** Deterministic steps the executor drove. */
+    driven_steps: number;
+    /** Screenshots captured during the journey. */
+    screenshots: number;
+    /** Mobile interaction engine used, when target was mobile. */
+    mobile_engine?: "maestro" | "coordinate";
+    /** Maestro flow pass/fail, when applicable. */
+    mobile_flows_passed?: number;
+    mobile_flows_total?: number;
+}
+/** Persist a run_journey interaction-evidence record into its run dir. */
+export declare function writeInteractionEvidence(runDir: string, ev: InteractionEvidence): void;
+/**
+ * Find the newest run_journey interaction_evidence.json across the project's
+ * artifacts/runs/* dirs. Returns the record + its mtime (ms) or null.
+ */
+export declare function loadLatestInteractionEvidence(artifactsRunsDir: string): {
+    evidence: InteractionEvidence;
+    mtimeMs: number;
+} | null;
+export interface InteractionGateInput {
+    /** A fresh run_journey artifact exists (and is not stale vs source). */
+    journeyEvidence: {
+        evidence: InteractionEvidence;
+        mtimeMs: number;
+    } | null;
+    journeyStale: boolean;
+    /** The manual recording flow produced interaction logs at some point. */
+    hasManualInteractionLogs: boolean;
+}
+export interface GateVerdict {
+    passed: boolean;
+    reason: string;
+}
+/**
+ * Verdict for the interaction_evidence gate. Passes when the app was driven by
+ * EITHER source; fails only when a UI project shows NO interactive testing at
+ * all, steering the agent to the one-call codeloop_run_journey path.
+ */
+export declare function evaluateInteractionEvidenceGate(input: InteractionGateInput): GateVerdict;
+//# sourceMappingURL=interaction_evidence.d.ts.map

package/dist/evidence/interaction_evidence.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"interaction_evidence.d.ts","sourceRoot":"","sources":["../../src/evidence/interaction_evidence.ts"],"names":[],"mappings":"AAkBA,MAAM,WAAW,mBAAmB;IAClC,YAAY,EAAE,MAAM,CAAC;IACrB,4CAA4C;IAC5C,MAAM,EAAE,aAAa,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,8CAA8C;IAC9C,YAAY,EAAE,MAAM,CAAC;IACrB,+CAA+C;IAC/C,WAAW,EAAE,MAAM,CAAC;IACpB,8DAA8D;IAC9D,aAAa,CAAC,EAAE,SAAS,GAAG,YAAY,CAAC;IACzC,+CAA+C;IAC/C,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAID,0EAA0E;AAC1E,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,EAAE,mBAAmB,GAAG,IAAI,CAMtF;AAED;;;GAGG;AACH,wBAAgB,6BAA6B,CAC3C,gBAAgB,EAAE,MAAM,GACvB;IAAE,QAAQ,EAAE,mBAAmB,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAAG,IAAI,CAoB3D;AAED,MAAM,WAAW,oBAAoB;IACnC,wEAAwE;IACxE,eAAe,EAAE;QAAE,QAAQ,EAAE,mBAAmB,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC3E,YAAY,EAAE,OAAO,CAAC;IACtB,yEAAyE;IACzE,wBAAwB,EAAE,OAAO,CAAC;CACnC;AAED,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,OAAO,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED;;;;GAIG;AACH,wBAAgB,+BAA+B,CAAC,KAAK,EAAE,oBAAoB,GAAG,WAAW,CA4BxF"}

package/dist/evidence/interaction_evidence.js

@@ -0,0 +1,91 @@
+/**
+ * interaction_evidence — proof that the running app was actually DRIVEN this
+ * project (typed into, tapped, submitted), by EITHER the hands-free executor
+ * (codeloop_run_journey, which writes interaction_evidence.json) OR the manual
+ * record -> interact -> replay flow (which writes logs/interaction_log.jsonl).
+ *
+ * The `interaction_evidence` gate is the umbrella that makes "you must drive
+ * the UI" a checked condition for UI projects, and — crucially — points the
+ * agent at the ONE-CALL path (codeloop_run_journey) when no interaction
+ * happened at all. It is deliberately satisfied by either evidence source so
+ * it never regresses teams already using the manual recording flow.
+ *
+ * Applicable-or-n/a: the caller only adds it for UI projects and treats audit
+ * (read-only) mode as n/a.
+ */
+import { existsSync, readFileSync, writeFileSync, readdirSync, statSync } from "fs";
+import { join } from "path";
+const FILE = "interaction_evidence.json";
+/** Persist a run_journey interaction-evidence record into its run dir. */
+export function writeInteractionEvidence(runDir, ev) {
+    try {
+        writeFileSync(join(runDir, FILE), JSON.stringify(ev, null, 2));
+    }
+    catch {
+        /* best-effort */
+    }
+}
+/**
+ * Find the newest run_journey interaction_evidence.json across the project's
+ * artifacts/runs/* dirs. Returns the record + its mtime (ms) or null.
+ */
+export function loadLatestInteractionEvidence(artifactsRunsDir) {
+    let best = null;
+    let runDirs = [];
+    try {
+        runDirs = readdirSync(artifactsRunsDir);
+    }
+    catch {
+        return null;
+    }
+    for (const d of runDirs) {
+        const p = join(artifactsRunsDir, d, FILE);
+        if (!existsSync(p))
+            continue;
+        try {
+            const mtimeMs = statSync(p).mtimeMs;
+            const evidence = JSON.parse(readFileSync(p, "utf-8"));
+            if (!best || mtimeMs > best.mtimeMs)
+                best = { evidence, mtimeMs };
+        }
+        catch {
+            /* skip unreadable */
+        }
+    }
+    return best;
+}
+/**
+ * Verdict for the interaction_evidence gate. Passes when the app was driven by
+ * EITHER source; fails only when a UI project shows NO interactive testing at
+ * all, steering the agent to the one-call codeloop_run_journey path.
+ */
+export function evaluateInteractionEvidenceGate(input) {
+    const { journeyEvidence, journeyStale, hasManualInteractionLogs } = input;
+    if (journeyEvidence && !journeyStale) {
+        const e = journeyEvidence.evidence;
+        const mobile = e.mobile_engine
+            ? ` (mobile engine: ${e.mobile_engine}${e.mobile_flows_total != null ? `, ${e.mobile_flows_passed ?? 0}/${e.mobile_flows_total} flow(s)` : ""})`
+            : "";
+        return {
+            passed: true,
+            reason: `codeloop_run_journey drove ${e.driven_steps} step(s) on ${e.target}${mobile}, ${e.screenshots} screenshot(s).`,
+        };
+    }
+    if (hasManualInteractionLogs) {
+        return {
+            passed: true,
+            reason: "Interactions recorded via the manual codeloop_interact flow (interaction_log.jsonl present).",
+        };
+    }
+    if (journeyEvidence && journeyStale) {
+        return {
+            passed: false,
+            reason: "The only run_journey interaction evidence is STALE (source code changed after it ran). Re-run codeloop_run_journey.",
+        };
+    }
+    return {
+        passed: false,
+        reason: "No interactive testing found. Drive the app once with codeloop_run_journey (one call: launches the app, types/taps/submits, screenshots).",
+    };
+}
+//# sourceMappingURL=interaction_evidence.js.map

package/dist/evidence/interaction_evidence.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"interaction_evidence.js","sourceRoot":"","sources":["../../src/evidence/interaction_evidence.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AACH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,aAAa,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,IAAI,CAAC;AACpF,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAkB5B,MAAM,IAAI,GAAG,2BAA2B,CAAC;AAEzC,0EAA0E;AAC1E,MAAM,UAAU,wBAAwB,CAAC,MAAc,EAAE,EAAuB;IAC9E,IAAI,CAAC;QACH,aAAa,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACjE,CAAC;IAAC,MAAM,CAAC;QACP,iBAAiB;IACnB,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,6BAA6B,CAC3C,gBAAwB;IAExB,IAAI,IAAI,GAA8D,IAAI,CAAC;IAC3E,IAAI,OAAO,GAAa,EAAE,CAAC;IAC3B,IAAI,CAAC;QACH,OAAO,GAAG,WAAW,CAAC,gBAAgB,CAAC,CAAC;IAC1C,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;IACD,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,CAAC,GAAG,IAAI,CAAC,gBAAgB,EAAE,CAAC,EAAE,IAAI,CAAC,CAAC;QAC1C,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;YAAE,SAAS;QAC7B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;YACpC,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,EAAE,OAAO,CAAC,CAAwB,CAAC;YAC7E,IAAI,CAAC,IAAI,IAAI,OAAO,GAAG,IAAI,CAAC,OAAO;gBAAE,IAAI,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC;QACpE,CAAC;QAAC,MAAM,CAAC;YACP,qBAAqB;QACvB,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAeD;;;;GAIG;AACH,MAAM,UAAU,+BAA+B,CAAC,KAA2B;IACzE,MAAM,EAAE,eAAe,EAAE,YAAY,EAAE,wBAAwB,EAAE,GAAG,KAAK,CAAC;IAC1E,IAAI,eAAe,IAAI,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,CAAC,GAAG,eAAe,CAAC,QAAQ,CAAC;QACnC,MAAM,MAAM,GAAG,CAAC,CAAC,aAAa;YAC5B,CAAC,CAAC,oBAAoB,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,kBAAkB,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,mBAAmB,IAAI,CAAC,IAAI,CAAC,CAAC,kBAAkB,UAAU,CAAC,CAAC,CAAC,EAAE,GAAG;YAChJ,CAAC,CAAC,EAAE,CAAC;QACP,OAAO;YACL,MAAM,EAAE,IAAI;YACZ,MAAM,EAAE,8BAA8B,CAAC,CAAC,YAAY,eAAe,CAAC,CAAC,MAAM,GAAG,MAAM,KAAK,CAAC,CAAC,WAAW,iBAAiB;SACxH,CAAC;IACJ,CAAC;IACD,IAAI,wBAAwB,EAAE,CAAC;QAC7B,OAAO;YACL,MAAM,EAAE,IAAI;YACZ,MAAM,EAAE,8FAA8F;SACvG,CAAC;IACJ,CAAC;IACD,IAAI,eAAe,IAAI,YAAY,EAAE,CAAC;QACpC,OAAO;YACL,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,qHAAqH;SAC9H,CAAC;IACJ,CAAC;IACD,OAAO;QACL,MAAM,EAAE,KAAK;QACb,MAAM,EAAE,2IAA2I;KACpJ,CAAC;AACJ,CAAC"}

package/dist/index.js

@@ -2684,6 +2684,17 @@ Returns: checklist of completed and pending verification steps.`, {
         }
         const gateIsPassing = hasGateCheck && latestMeta?.gate_result === "passed";
         const gateConfidence = latestMeta?.confidence ?? 0;
+        // Deep-E2E journey evidence — has codeloop_run_journey driven the app?
+        // (Audit/read-only mode never drives the app, so the step is n/a there.)
+        let hasJourneyEvidence = false;
+        let workflowAuditMode = false;
+        try {
+            const { loadLatestInteractionEvidence } = await import("./evidence/interaction_evidence.js");
+            hasJourneyEvidence = loadLatestInteractionEvidence(join(baseDir, "runs")) != null;
+            const { resolveAgentMode } = await import("./evidence/agent_mode.js");
+            workflowAuditMode = resolveAgentMode({ cwd, configMode: config.agent_mode }) === "audit";
+        }
+        catch { /* best-effort */ }
         // Interaction coverage: compare interaction_log selectors/URLs against discover_screens
         let interactionCount = 0;
         const coveredRoutes = new Set();
@@ -2803,6 +2814,23 @@ Returns: checklist of completed and pending verification steps.`, {
                         ? `${videoCount} video(s) recorded`
                         : "No video recordings found. Call codeloop_start_recording → interact with ALL elements → codeloop_stop_recording → codeloop_interaction_replay.",
             },
+            {
+                step: "3b. Deep-E2E journey (run_journey)",
+                status: !isUIProject || workflowAuditMode
+                    ? "n/a"
+                    : hasJourneyEvidence || interactionCount > 0
+                        ? "done"
+                        : "PENDING",
+                detail: !isUIProject
+                    ? "Not a UI project — deep-E2E journey not required"
+                    : workflowAuditMode
+                        ? "Audit/read-only mode — the app is not launched or driven. Switch to fix mode for full interaction testing."
+                        : hasJourneyEvidence
+                            ? "codeloop_run_journey drove the app (interaction_evidence.json present)."
+                            : interactionCount > 0
+                                ? "App driven via the manual codeloop_interact flow."
+                                : "App NOT driven yet. Call codeloop_run_journey ONCE — it launches the app / boots the emulator+simulator, types/taps/submits by label (Maestro on mobile), screenshots every screen, and records video. This satisfies the interaction_evidence gate.",
+            },
             {
                 step: "4. Gate check",
                 status: !hasGateCheck ? "PENDING" : gateIsPassing ? "done" : "NEEDS_RERUN",