feat: implement memory and evidence administration with guided repairs

Add PostgreSQL-backed memory, editable evidence with source review and activation, and human-approved archive repairs across the harness, API, and UI. Include migrations, deployment support, regression coverage, and validation documentation. Refresh permissions from validated session roles so existing administrator logins can access newly deployed archive management features.
2026-09-10 10:31:34 +02:00
parent 8fe526dd6e
commit 82e2c91f42
168 changed files with 11914 additions and 1772 deletions
@@ -0,0 +1,61 @@
+const test = require("node:test");
+const assert = require("node:assert/strict");
+const { installRepairGate } = require("../memory/repair.js");
+
+function gate({ answers, fail = false, resume = false }) {
+    let tool;
+    const calls = [], widgets = [];
+    const repair = { repair_id: "receipt", choice: null, status: "proposed", can_apply: true,
+        saved: false, indexed: false, options: [{ id: "fix", content: { detail: "Correction" } }] };
+    installRepairGate({ registerTool: def => { tool = def; } }, {
+        workflow: { activate() {}, phase: () => ({ id: "F4" }) },
+        memory: { execute: (_ctx, args) => {
+            calls.push(args);
+            if (args[0] === "repair-apply") {
+                const choice = args[args.indexOf("--choice") + 1];
+                if (choice === "reject") Object.assign(repair, { choice, status: "rejected" });
+                else Object.assign(repair, { choice, saved: true, indexed: !fail,
+                    status: fail ? "pending_activation" : "active" });
+                fail = false;
+            }
+            return JSON.stringify(repair);
+        } },
+        waitForReviewer: async (_ctx, descriptor) => {
+            widgets.push(descriptor);
+            assert.ok(answers.length, "gate must not ask an unbounded extra question");
+            return answers.shift();
+        },
+        toTextResult: text => text,
+    });
+    return { calls, widgets, run: () => tool.execute("id", {
+        session: "s", ...(resume ? { repair_id: "receipt" } : { proposal: { reason: "conflict", options: [] } }),
+    }, null, null, {}) };
+}
+
+test("human choice applies once and the next widget reports activation", async () => {
+    const g = gate({ answers: [{ choices: ["fix"] }, { choices: ["continue"] }] });
+    assert.equal(JSON.parse(await g.run()).indexed, true);
+    assert.deepEqual(g.calls.map(args => args[0]), ["repair-prepare", "repair-apply"]);
+    assert.equal(g.widgets[0].repair.saved, false);
+    assert.equal(g.widgets[1].repair.status, "active");
+});
+test("rejection asks for reformulation without a normal workflow decision", async () => {
+    const g = gate({ answers: [{ choices: ["reject"] }] });
+    assert.match(await g.run(), /rejected as inadequate/);
+    assert.equal(g.calls.length, 2);
+});
+test("retry keeps the receipt and selected choice, then reports true activation", async () => {
+    const g = gate({ resume: true, fail: true,
+        answers: [{ choices: ["fix"] }, { choices: ["fix"] }, { choices: ["continue"] }] });
+    assert.equal(JSON.parse(await g.run()).indexed, true);
+    assert.equal(g.widgets[1].repair.status, "pending_activation");
+    assert.equal(g.calls[0][0], "repair-show");
+    assert.deepEqual(g.calls[1], g.calls[2]);
+});
+test("free text and forged choices never apply a correction", async () => {
+    for (const response of [{ control: "freetext", text: "Explain the conflict" }, { choices: ["forged"] }]) {
+        const g = gate({ answers: [response] });
+        await g.run();
+        assert.equal(g.calls.length, 1);
+    }
+});