tewip-ai 0.70.8 → 0.70.9

package/dist/engine/pursue/pursue-command.js

@@ -11,6 +11,7 @@ import { datasetFiles, readJsonl, readTriageRecords } from '../../reporters/stor
 import { alreadyTried, cannotAct, planFor } from './act.js';
 import { decide } from './policy.js';
 import { CANNOT_FAIL } from '../vouch/vouch.js';
+import { applyCorrection, correctionFor, loadCorrections } from '../corrections.js';
 import { vouchAll } from '../vouch/vouch-command.js';
 const words = (text) => (text.toLowerCase().match(/[a-z][a-z0-9]{2,}/g) ?? []).filter((w) => !['the', 'and', 'for', 'this', 'that', 'with', 'keep', 'green', 'sprint', 'flow', 'test', 'tests'].includes(w));
 /** Does this test or scenario belong to the goal? Deterministic, so a person can predict it. */
@@ -25,6 +26,7 @@ export function observe(root, goal, only) {
     const dir = join(root, '.tewip', 'history');
     const runs = [];
     const records = [];
+    const corrections = [];
     if (existsSync(dir)) {
         for (const project of only ? [only] : readdirSync(dir)) {
             const where = join(dir, project);
@@ -32,16 +34,34 @@ export function observe(root, goal, only) {
                 continue;
             runs.push(...readJsonl(datasetFiles(where).runs));
             records.push(...readTriageRecords(datasetFiles(where).triage));
+            corrections.push(...loadCorrections(where));
         }
     }
     runs.sort((a, b) => (a.finishedAt > b.finishedAt ? 1 : -1));
     const newest = runs.at(-1);
     const wanted = words(goal);
+    // A person's correction decides the category, as it does everywhere else.
+    //
+    // This read `r.diagnosis` straight off the record, so a correction never reached the decision.
+    // Field-tested: a person marked a failure REAL_BUG, and `pursue` went on calling it LOCATOR_DRIFT
+    // and concluding "the goal is failing and Tewip cannot say why". That matters more here than in a
+    // report, because this is the command that **acts** - and its own documented order is "a real bug
+    // first... it will not touch a real bug", a safety rule that depends on recognising one. The
+    // strongest possible signal that something is a real bug is a person saying so.
+    //
+    // Phase 74 fixed this inside `buildItems` for every command that goes through it. This one does
+    // not, so it applies them itself.
     const failing = records
         .filter((r) => r.runId === newest?.runId && about(wanted, r.testId))
         .map((r) => {
-        const d = r.diagnosis;
-        return { testId: r.testId, category: String(d?.category ?? 'UNKNOWN'), confidence: Number(d?.confidence ?? 0) };
+        const correction = corrections.length > 0 ? correctionFor(r, corrections) : null;
+        const own = r.diagnosis;
+        const d = correction && own ? applyCorrection(own, correction) : own;
+        return {
+            testId: r.testId,
+            category: String((correction?.category ?? d?.category) ?? 'UNKNOWN'),
+            confidence: Number(d?.confidence ?? (correction ? 1 : 0)),
+        };
     });
     // A test that failed in more than one earlier run and is not failing now has come and gone.
     const failedBefore = new Map();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tewip-ai",
-  "version": "0.70.8",
+  "version": "0.70.9",
   "mcpName": "io.github.hmamut39/tewip",
   "description": "Tewip finds why each failing test failed - real bug, flaky test, environment, test data or changed selector - for any stack, from the command line, CI, or your editor.",
   "keywords": [