codeloop-mcp-server 0.1.78 → 0.1.82

package/dist/evidence/agent_mode.d.ts

@@ -1,7 +1,14 @@
 export type AgentMode = "fix" | "audit";
+/**
+ * How long a PERSISTED audit mode stays in effect after it was last set.
+ * Each audit tool call rewrites the file (refreshing this window), so a
+ * continuous audit session never lapses; only an idle/abandoned audit does.
+ * Kept short enough that a later, unrelated request defaults back to fix.
+ */
+export declare const AUDIT_TTL_MS: number;
 /** Normalize a free-text mode value to the canonical enum. */
 export declare function normalizeAgentMode(value?: string | null): AgentMode | undefined;
-export declare function readPersistedAgentMode(cwd: string): AgentMode | undefined;
+export declare function readPersistedAgentMode(cwd: string, now?: number): AgentMode | undefined;
 export declare function persistAgentMode(cwd: string, mode: AgentMode): void;
 /**
  * Resolve the effective mode for a tool call and persist an explicit param so
@@ -15,8 +22,11 @@ export declare function resolveAgentMode(opts: {
 /** Shared schema description so every tool's `mode` param reads identically. */
 export declare const MODE_PARAM_DESCRIPTION: string;
 /**
- * The read-only directive appended to verify / diagnose / gate_check responses
- * when audit mode is active. Replaces the aggressive auto-fix prose.
+ * The report-only directive appended to verify / diagnose / gate_check responses
+ * when audit tone is active. It softens ONLY the auto-fix prose — it never tells
+ * the agent to skip verification. CodeLoop still runs/expects the full suite
+ * (including codeloop_run_journey to launch + drive the app); audit just means
+ * "don't edit the user's code yet."
  */
 export declare function buildAuditDirective(tool: "verify" | "diagnose" | "gate_check"): string;
 //# sourceMappingURL=agent_mode.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"agent_mode.d.ts","sourceRoot":"","sources":["../../src/evidence/agent_mode.ts"],"names":[],"mappings":"AAuBA,MAAM,MAAM,SAAS,GAAG,KAAK,GAAG,OAAO,CAAC;AAIxC,8DAA8D;AAC9D,wBAAgB,kBAAkB,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,GAAG,SAAS,CAc/E;AAED,wBAAgB,sBAAsB,CAAC,GAAG,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS,CASzE;AAED,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,SAAS,GAAG,IAAI,CAQnE;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE;IACrC,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,SAAS,CAWZ;AAED,gFAAgF;AAChF,eAAO,MAAM,sBAAsB,QAOyC,CAAC;AAE7E;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,QAAQ,GAAG,UAAU,GAAG,YAAY,GAAG,MAAM,CAyBtF"}
+{"version":3,"file":"agent_mode.d.ts","sourceRoot":"","sources":["../../src/evidence/agent_mode.ts"],"names":[],"mappings":"AAwCA,MAAM,MAAM,SAAS,GAAG,KAAK,GAAG,OAAO,CAAC;AAIxC;;;;;GAKG;AACH,eAAO,MAAM,YAAY,QAAiB,CAAC;AAE3C,8DAA8D;AAC9D,wBAAgB,kBAAkB,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,GAAG,SAAS,CAc/E;AAED,wBAAgB,sBAAsB,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,GAAE,MAAmB,GAAG,SAAS,GAAG,SAAS,CAgBnG;AAED,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,SAAS,GAAG,IAAI,CAQnE;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE;IACrC,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,SAAS,CAWZ;AAED,gFAAgF;AAChF,eAAO,MAAM,sBAAsB,QAawC,CAAC;AAE5E;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,QAAQ,GAAG,UAAU,GAAG,YAAY,GAAG,MAAM,CAyBtF"}

package/dist/evidence/agent_mode.js

@@ -1,12 +1,18 @@
 /**
- * Agent mode — "fix" (default) vs "audit" (read-only).
+ * Agent mode — "fix" (default) vs "audit" (report-only TONE).
  *
- * CodeLoop's tool responses normally PUSH the agent into a mandatory
- * verify → diagnose → fix → re-verify → gate auto-fix loop. That is the right
- * default, but it directly contradicts a user who says "do NOT modify my code,
- * just run the checks and list every problem you find." In that case CodeLoop
- * must obey the user: still run all verifications and produce the full findings
- * list, but STOP pushing the agent to edit files or loop.
+ * IMPORTANT: audit is NOT a CodeLoop verification mode — it does not change what
+ * CodeLoop verifies. CodeLoop ALWAYS runs the full suite (build, tests, launches
+ * + DRIVES the app via codeloop_run_journey, screenshots, gate, confidence) and
+ * lists every issue it finds, in every mode. Driving the app is verification,
+ * not a code modification.
+ *
+ * "audit" only adjusts the response TONE for the one case where a user says
+ * "do NOT modify my code, just run the checks and list every problem you find."
+ * Then CodeLoop still verifies everything and produces the full findings list,
+ * but its prose does not COMMAND the agent to edit files or enter the auto-fix
+ * loop — the agent presents the report and lets the user decide what to fix.
+ * The agent's source-code edits are paused; nothing about CodeLoop's checks is.
  *
  * Mode resolution precedence (highest first):
  *   1. explicit `mode` param on the current tool call
@@ -17,10 +23,28 @@
  * When a tool receives an explicit `mode`, it persists it so the rest of the
  * session inherits the same mode without the agent having to repeat it on
  * every call.
+ *
+ * READ-ONLY IS OPT-IN, NOT STICKY. audit must only apply when the user has
+ * actually asked for it ("don't modify my code, just list the problems").
+ * Persisted audit therefore EXPIRES: it is honored only for AUDIT_TTL_MS after
+ * it was last set, and every audit tool call refreshes that timestamp. The
+ * moment the agent stops passing `mode:"audit"` (because the user is no longer
+ * restricting edits), the persisted audit goes stale and CodeLoop reverts to
+ * the active fix default — verifying, checking the gate/confidence, and pushing
+ * the auto-fix loop. Persisted "fix" never expires (it IS the default). An
+ * explicit config.agent_mode:"audit" is a deliberate user opt-in and is honored
+ * with no TTL.
  */
 import { existsSync, readFileSync, writeFileSync, mkdirSync } from "fs";
 import { join, dirname } from "path";
 const MODE_FILE_REL = join(".codeloop", "agent_mode.json");
+/**
+ * How long a PERSISTED audit mode stays in effect after it was last set.
+ * Each audit tool call rewrites the file (refreshing this window), so a
+ * continuous audit session never lapses; only an idle/abandoned audit does.
+ * Kept short enough that a later, unrelated request defaults back to fix.
+ */
+export const AUDIT_TTL_MS = 30 * 60 * 1000; // 30 minutes
 /** Normalize a free-text mode value to the canonical enum. */
 export function normalizeAgentMode(value) {
     if (!value)
@@ -36,13 +60,21 @@ export function normalizeAgentMode(value) {
     }
     return undefined;
 }
-export function readPersistedAgentMode(cwd) {
+export function readPersistedAgentMode(cwd, now = Date.now()) {
     try {
         const p = join(cwd, MODE_FILE_REL);
         if (!existsSync(p))
             return undefined;
         const data = JSON.parse(readFileSync(p, "utf-8"));
-        return normalizeAgentMode(data.mode);
+        const mode = normalizeAgentMode(data.mode);
+        // Persisted audit is OPT-IN and time-boxed: ignore it once stale so CodeLoop
+        // reverts to the active fix default unless the user keeps requesting audit.
+        if (mode === "audit") {
+            const setAt = data.set_at ? Date.parse(data.set_at) : NaN;
+            if (!Number.isFinite(setAt) || now - setAt > AUDIT_TTL_MS)
+                return undefined;
+        }
+        return mode;
     }
     catch {
         return undefined;
@@ -73,29 +105,38 @@ export function resolveAgentMode(opts) {
         "fix");
 }
 /** Shared schema description so every tool's `mode` param reads identically. */
-export const MODE_PARAM_DESCRIPTION = "Agent workflow mode. 'fix' (default) = the standard auto-fix loop: after " +
-    "verify/gate you should diagnose and modify code until gates pass. " +
-    "'audit' = READ-ONLY: set this when the user asked you to ONLY run the " +
-    "checks and LIST the problems WITHOUT modifying their code. In audit mode " +
-    "CodeLoop still runs every verification and returns the full findings, but " +
-    "its responses will NOT push you to edit files or loop — you present the " +
-    "report to the user and stop. The mode persists for the session once set.";
+export const MODE_PARAM_DESCRIPTION = "Response TONE only — it does NOT change what CodeLoop verifies. LEAVE UNSET " +
+    "for the normal behavior: 'fix' (default) = the standard auto-fix loop where, " +
+    "after verify/gate, you actively diagnose and modify code, re-verify, and drive " +
+    "the gate/confidence up until it passes. ONLY set 'audit' when the user " +
+    "EXPLICITLY asked you to run the checks and LIST the problems WITHOUT modifying " +
+    "their code (e.g. 'don't change anything, just tell me what's wrong'). Do NOT " +
+    "infer audit on your own. In EITHER mode CodeLoop runs the FULL suite — build, " +
+    "tests, launching + DRIVING the app (codeloop_run_journey), screenshots, gate, " +
+    "confidence — and returns every issue; driving the app is verification, not a " +
+    "code edit. 'audit' only stops CodeLoop's prose from commanding you to edit " +
+    "files or loop, so you present the report and let the user decide. Audit is " +
+    "OPT-IN and time-boxed: it auto-expires back to the fix default; pass mode:'fix' " +
+    "(or omit it) the moment the user is ready for you to apply fixes again.";
 /**
- * The read-only directive appended to verify / diagnose / gate_check responses
- * when audit mode is active. Replaces the aggressive auto-fix prose.
+ * The report-only directive appended to verify / diagnose / gate_check responses
+ * when audit tone is active. It softens ONLY the auto-fix prose — it never tells
+ * the agent to skip verification. CodeLoop still runs/expects the full suite
+ * (including codeloop_run_journey to launch + drive the app); audit just means
+ * "don't edit the user's code yet."
  */
 export function buildAuditDirective(tool) {
     const head = [
         "",
         "",
-        "📋 AUDIT MODE (read-only) — the user asked you to LIST problems, not fix them.",
-        "Do NOT modify any source files. Do NOT enter the auto-fix loop. Do NOT call codeloop_diagnose-then-edit unless the user explicitly asks.",
+        "📋 AUDIT MODE (report-only) — the user asked you to LIST problems, not fix them YET.",
+        "Do NOT modify any source files and do NOT enter the auto-fix loop. This pauses CODE EDITS only — it does NOT pause verification: still run CodeLoop's full read-only checks, including codeloop_run_journey (launch + drive the app + screenshots) and codeloop_gate_check, so your findings list is complete.",
     ];
     const tail = {
         verify: [
             "Present the failing checks above to the user as a findings list (file, what failed, why). " +
-                "You MAY call codeloop_diagnose to get a richer, structured breakdown to include in the report — diagnose only classifies, it does not edit code. " +
-                "After reporting, STOP and wait for the user to decide what to fix.",
+                "You MAY call codeloop_diagnose for a richer structured breakdown and SHOULD still call codeloop_run_journey to drive the app — both only gather evidence, neither edits code. " +
+                "After producing the complete report, STOP and wait for the user to decide what to fix.",
         ],
         diagnose: [
             "The repair_tasks above are a DIAGNOSIS for the REPORT, not a TODO list to execute. " +

package/dist/evidence/agent_mode.js.map

@@ -1 +1 @@
-{"version":3,"file":"agent_mode.js","sourceRoot":"","sources":["../../src/evidence/agent_mode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AACH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,IAAI,CAAC;AACxE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAIrC,MAAM,aAAa,GAAG,IAAI,CAAC,WAAW,EAAE,iBAAiB,CAAC,CAAC;AAE3D,8DAA8D;AAC9D,MAAM,UAAU,kBAAkB,CAAC,KAAqB;IACtD,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,IAAI,EAAE,CAAC;IAC7C,IACE,CAAC,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAE,QAAQ,EAAE,aAAa;QACrE,aAAa,EAAE,WAAW,EAAE,WAAW,EAAE,eAAe,EAAE,YAAY;QACtE,QAAQ,EAAE,QAAQ,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EACjC,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;QACnE,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,GAAW;IAChD,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;QACnC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;YAAE,OAAO,SAAS,CAAC;QACrC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,EAAE,OAAO,CAAC,CAAsB,CAAC;QACvE,OAAO,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACvC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,GAAW,EAAE,IAAe;IAC3D,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;QACnC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3C,aAAa,CAAC,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACxF,CAAC;IAAC,MAAM,CAAC;QACP,kEAAkE;IACpE,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,IAIhC;IACC,MAAM,SAAS,GAAG,kBAAkB,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACrD,IAAI,SAAS,EAAE,CAAC;QACd,gBAAgB,CAAC,IAAI,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;QACtC,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,OAAO,CACL,sBAAsB,CAAC,IAAI,CAAC,GAAG,CAAC;QAChC,kBAAkB,CAAC,IAAI,CAAC,UAAU,CAAC;QACnC,KAAK,CACN,CAAC;AACJ,CAAC;AAED,gFAAgF;AAChF,MAAM,CAAC,MAAM,sBAAsB,GACjC,2EAA2E;IAC3E,oEAAoE;IACpE,wEAAwE;IACxE,2EAA2E;IAC3E,4EAA4E;IAC5E,0EAA0E;IAC1E,0EAA0E,CAAC;AAE7E;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAA0C;IAC5E,MAAM,IAAI,GAAG;QACX,EAAE;QACF,EAAE;QACF,gFAAgF;QAChF,0IAA0I;KAC3I,CAAC;IACF,MAAM,IAAI,GAAkC;QAC1C,MAAM,EAAE;YACN,4FAA4F;gBAC1F,mJAAmJ;gBACnJ,oEAAoE;SACvE;QACD,QAAQ,EAAE;YACR,qFAAqF;gBACnF,2FAA2F;gBAC3F,sGAAsG;SACzG;QACD,UAAU,EAAE;YACV,8DAA8D;gBAC5D,sFAAsF;gBACtF,+FAA+F;SAClG;KACF,CAAC;IACF,OAAO,CAAC,GAAG,IAAI,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC7C,CAAC"}
+{"version":3,"file":"agent_mode.js","sourceRoot":"","sources":["../../src/evidence/agent_mode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AACH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,IAAI,CAAC;AACxE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAIrC,MAAM,aAAa,GAAG,IAAI,CAAC,WAAW,EAAE,iBAAiB,CAAC,CAAC;AAE3D;;;;;GAKG;AACH,MAAM,CAAC,MAAM,YAAY,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,aAAa;AAEzD,8DAA8D;AAC9D,MAAM,UAAU,kBAAkB,CAAC,KAAqB;IACtD,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,IAAI,EAAE,CAAC;IAC7C,IACE,CAAC,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAE,QAAQ,EAAE,aAAa;QACrE,aAAa,EAAE,WAAW,EAAE,WAAW,EAAE,eAAe,EAAE,YAAY;QACtE,QAAQ,EAAE,QAAQ,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EACjC,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;QACnE,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,GAAW,EAAE,MAAc,IAAI,CAAC,GAAG,EAAE;IAC1E,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;QACnC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;YAAE,OAAO,SAAS,CAAC;QACrC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,EAAE,OAAO,CAAC,CAAuC,CAAC;QACxF,MAAM,IAAI,GAAG,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC3C,6EAA6E;QAC7E,4EAA4E;QAC5E,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;YACrB,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;YAC1D,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,GAAG,GAAG,KAAK,GAAG,YAAY;gBAAE,OAAO,SAAS,CAAC;QAC9E,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,GAAW,EAAE,IAAe;IAC3D,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;QACnC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3C,aAAa,CAAC,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACxF,CAAC;IAAC,MAAM,CAAC;QACP,kEAAkE;IACpE,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,IAIhC;IACC,MAAM,SAAS,GAAG,kBAAkB,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACrD,IAAI,SAAS,EAAE,CAAC;QACd,gBAAgB,CAAC,IAAI,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;QACtC,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,OAAO,CACL,sBAAsB,CAAC,IAAI,CAAC,GAAG,CAAC;QAChC,kBAAkB,CAAC,IAAI,CAAC,UAAU,CAAC;QACnC,KAAK,CACN,CAAC;AACJ,CAAC;AAED,gFAAgF;AAChF,MAAM,CAAC,MAAM,sBAAsB,GACjC,8EAA8E;IAC9E,+EAA+E;IAC/E,iFAAiF;IACjF,yEAAyE;IACzE,iFAAiF;IACjF,+EAA+E;IAC/E,gFAAgF;IAChF,gFAAgF;IAChF,+EAA+E;IAC/E,6EAA6E;IAC7E,6EAA6E;IAC7E,kFAAkF;IAClF,yEAAyE,CAAC;AAE5E;;;;;;GAMG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAA0C;IAC5E,MAAM,IAAI,GAAG;QACX,EAAE;QACF,EAAE;QACF,sFAAsF;QACtF,gTAAgT;KACjT,CAAC;IACF,MAAM,IAAI,GAAkC;QAC1C,MAAM,EAAE;YACN,4FAA4F;gBAC1F,gLAAgL;gBAChL,wFAAwF;SAC3F;QACD,QAAQ,EAAE;YACR,qFAAqF;gBACnF,2FAA2F;gBAC3F,sGAAsG;SACzG;QACD,UAAU,EAAE;YACV,8DAA8D;gBAC5D,sFAAsF;gBACtF,+FAA+F;SAClG;KACF,CAAC;IACF,OAAO,CAAC,GAAG,IAAI,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC7C,CAAC"}

package/dist/evidence/interaction_evidence.d.ts

@@ -8,7 +8,7 @@ export interface InteractionEvidence {
     /** Screenshots captured during the journey. */
     screenshots: number;
     /** Mobile interaction engine used, when target was mobile. */
-    mobile_engine?: "maestro" | "coordinate";
+    mobile_engine?: "flutter_driver" | "maestro" | "coordinate";
     /** Maestro flow pass/fail, when applicable. */
     mobile_flows_passed?: number;
     mobile_flows_total?: number;

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

@@ -1 +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"}
+{"version":3,"file":"interaction_evidence.d.ts","sourceRoot":"","sources":["../../src/evidence/interaction_evidence.ts"],"names":[],"mappings":"AAmBA,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,gBAAgB,GAAG,SAAS,GAAG,YAAY,CAAC;IAC5D,+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

@@ -10,8 +10,9 @@
  * 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.
+ * Applicable-or-n/a: the caller only adds it for UI projects. It applies in
+ * EVERY mode — driving the app is verification, not a code edit, so a
+ * "don't modify my code" request never makes it n/a.
  */
 import { existsSync, readFileSync, writeFileSync, readdirSync, statSync } from "fs";
 import { join } from "path";

package/dist/evidence/interaction_evidence.js.map

@@ -1 +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"}
+{"version":3,"file":"interaction_evidence.js","sourceRoot":"","sources":["../../src/evidence/interaction_evidence.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;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

@@ -1801,19 +1801,21 @@ server.tool("codeloop_run_journey", TOOL_BOOTSTRAP + `DEEP-E2E EXECUTOR — laun
 in ONE hands-free call. This is the autonomous counterpart to the manual
 plan→start_recording→interact→stop→replay sequence: CodeLoop does it for you.
 
+Runs in EVERY mode — launching + driving the app is verification, not a code edit, so a
+"don't modify my code" request never disables it (that only pauses the agent's source edits).
+
 What it does, in order:
-1. REFUSES in audit/read-only mode (driving the app modifies its state).
-2. Detects the target (browser / desktop / android_emulator / ios_simulator / Flutter) — overridable.
-3. READY/LAUNCH per target: web → headed Playwright at e2e.web_url; desktop → launch evidence.target_app;
+1. Detects the target (browser / desktop / android_emulator / ios_simulator / Flutter) — overridable.
+2. READY/LAUNCH per target: web → headed Playwright at e2e.web_url; desktop → launch evidence.target_app;
    Android/iOS → BOOT the emulator/simulator (reuses one already booted; honors e2e.android_avd /
    e2e.ios_device; opt out with e2e.boot_device:false). If a mobile device can't be booted it returns
    a copy-paste directive instead of failing.
-4. Plans the journey (codeloop_plan_user_journey) — entity CRUD arcs + the AI-chatbox arc.
-5. Starts a background video recording (best-effort).
-6. DRIVES every deterministic step via the interaction engine (fill known fields, type the AI prompt +
+3. Plans the journey (codeloop_plan_user_journey) — entity CRUD arcs + the AI-chatbox arc.
+4. Starts a background video recording (best-effort).
+5. DRIVES every deterministic step via the interaction engine (fill known fields, type the AI prompt +
    submit + read back / assert a non-empty reply), capturing a screenshot after each step.
-7. Visits EVERY discovered screen (codeloop_discover_screens) and screenshots each.
-8. Stops the recording and returns a directive to run codeloop_interaction_replay + visual_review +
+6. Visits EVERY discovered screen (codeloop_discover_screens) and screenshots each.
+7. Stops the recording and returns a directive to run codeloop_interaction_replay + visual_review +
    design_compare + gate_check.
 
 Steps it can't resolve deterministically (ambiguous navigation, raw-coordinate targets, missing
@@ -1830,7 +1832,6 @@ screens_captured[], screenshots[], unsupported_count, manual_followups[], direct
     target_type: targetTypeSchema.optional().describe("Override the auto-detected interaction target. Accepts synonyms (web→browser, android→android_emulator, ios→ios_simulator, *_desktop→desktop)."),
     web_url: z.string().optional().describe("URL to open for browser targets (e.g. http://localhost:3000). Defaults to e2e.web_url from config. Start your dev server first."),
     max_duration_seconds: z.number().int().min(10).max(600).optional().describe("Max video recording length. Default 180s."),
-    mode: z.string().optional().describe(AGENT_MODE_PARAM_DESC),
 }, async (params) => {
     const result = await withAuth(async () => {
         const cwd = resolveCwd(params);
@@ -1839,7 +1840,6 @@ screens_captured[], screenshots[], unsupported_count, manual_followups[], direct
         const { runJourney } = await import("./tools/run_journey.js");
         return runJourney({
             cwd,
-            paramMode: params.mode,
             e2e: { ...cfg.e2e, web_url: params.web_url ?? cfg.e2e?.web_url },
             targetApp: cfg.evidence?.target_app,
             targetType: params.target_type,
@@ -2685,14 +2685,15 @@ 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.)
+        // Required for every UI project regardless of mode: driving the app is
+        // verification, not a code edit, so "don't modify my code" never waives it.
         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";
+            // baseDir already ends in `/runs` (getArtifactsBaseDir) — pass it as-is,
+            // NOT join(baseDir,"runs") which double-nests and hid the evidence so
+            // step 3b stayed PENDING forever even after run_journey drove the app.
+            hasJourneyEvidence = loadLatestInteractionEvidence(baseDir) != null;
         }
         catch { /* best-effort */ }
         // Interaction coverage: compare interaction_log selectors/URLs against discover_screens
@@ -2816,20 +2817,18 @@ Returns: checklist of completed and pending verification steps.`, {
             },
             {
                 step: "3b. Deep-E2E journey (run_journey)",
-                status: !isUIProject || workflowAuditMode
+                status: !isUIProject
                     ? "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.",
+                    : 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",