@progen-labs/spectra 0.4.0 → 0.4.1

package/README.md

@@ -235,12 +235,13 @@ The harness, not the agent, makes every decision:
 - **Proof.** The change is kept only if `spectra compare` says *improved* with medium or high confidence and no new
   critical finding; with low confidence it measures again.
 - **A record.** Each kept commit carries its evidence; `.spectra/fix/<time>/report.md` summarises the session, and a
-  history skips findings that failed before.
+  history skips findings that failed before. A finding that an earlier kept fix already cleared is skipped too.
 
 By default the agent is Claude Code in headless mode, allowed only to read and edit files (no shell, no web, no MCP),
 with a spending cap per finding. A test bench (`npm run bench`) checks the harness's decisions in CI: a real fix is
 kept, a "faster" change that breaks the page is caught, a cosmetic change doesn't count, and an edit to
-`package.json` is refused. See [the AI fix harness](https://progen-labs.github.io/website/#spectra/docs/harness).
+`package.json` is refused. With Claude Code on the demo app, it kept two of three fixes and cut JavaScript CPU time
+from 1.51 s to 82.5 ms (−95%) for $0.44. See [the AI fix harness](https://progen-labs.github.io/website/#spectra/docs/harness).
 
 ## In VS Code
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@progen-labs/spectra",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "Spectra \u2014 a Chrome DevTools Protocol profiler that identifies the files, functions and components causing frontend slowness, across React, Vue, Svelte, Angular and Next.js builds.",
   "type": "module",
   "bin": {

package/server/fix/harness.js

@@ -4,6 +4,7 @@
  * Everything it touches is passed in (profile, compare, agent, workspace,
  * checks), so the loop itself is tested without a browser or a model.
  */
+import { findingKey } from '../../shared/compare.js';
 import { fixPrompt, judge, rankFindings, scopeProblem } from './policy.js';
 
 // agent errors no other finding would get past: stop instead of trying them all
@@ -34,6 +35,7 @@ export async function runFix(o) {
   const baseline = await profile('baseline');
   session.baseline = { file: baseline.file, summary: baseline.result.summary, url: baseline.result.url };
   let current = [baseline.file];
+  let present = null; // finding keys the latest kept change still measures
 
   const { candidates, skipped } = rankFindings(baseline.result.findings, { exists: ws.exists, history, retry, max: maxFixes });
   session.skipped = skipped.map(s => ({ title: s.finding.title, reason: s.reason }));
@@ -46,6 +48,12 @@ export async function runFix(o) {
   for (const [i, f] of candidates.entries()) {
     if (signal?.aborted) break;
     const n = `${i + 1}/${candidates.length}`;
+    if (present && !present.has(f.key)) {
+      // one fix often clears several findings (a memoised row also stops its hot file)
+      session.skipped.push({ title: f.title, reason: 'already resolved by an earlier kept change' });
+      log(`[${n}] – ${f.title}: already resolved by an earlier kept change.`);
+      continue;
+    }
     const attempt = { finding: { title: f.title, file: f.file, line: f.line, severity: f.severity, impactMs: f.impactMs, key: f.key }, outcome: null, reason: '', changes: [] };
     session.attempts.push(attempt);
     const discard = (outcome, reason) => {
@@ -127,6 +135,7 @@ export async function runFix(o) {
     Object.assign(attempt, { outcome: 'kept', reason: verdict.reason });
     session.kept++;
     current = after.map(a => a.file);
+    present = new Set(after.flatMap(a => a.result.findings || []).map(findingKey));
     log(`[${n}] ✓ kept (${attempt.commit}): ${verdict.reason}`);
   }
 

package/server/fix/test-fix.js

@@ -70,7 +70,8 @@ test('tells the agent the finding and the rules', () => {
 });
 
 // --- the loop, with fakes
-function fakes({ agentChanges, checks = [true, true, true, true], verdicts }) {
+const bothFindings = () => [finding('row-render'), finding('heavy', { file: 'src/heavy.js', line: 2, culprits: [] })];
+function fakes({ agentChanges, checks = [true, true, true, true], verdicts, measured = bothFindings }) {
   const calls = { commits: [], undos: 0, profiles: 0, compares: [] };
   let pending = [];
   let checkRun = 0;
@@ -91,7 +92,7 @@ function fakes({ agentChanges, checks = [true, true, true, true], verdicts }) {
       workspace,
       checks: [{ name: 'npm test' }],
       runChecks: async () => [{ name: 'npm test', ok: checks[checkRun++] ?? true, tail: 'expected 120 rows' }],
-      profile: async label => ((calls.profiles++), { file: `${label}.json`, result: { url: 'http://localhost:5199', summary: {}, findings: [finding('row-render'), finding('heavy', { file: 'src/heavy.js', line: 2, culprits: [] })] } }),
+      profile: async label => ((calls.profiles++), { file: `${label}.json`, result: { url: 'http://localhost:5199', summary: {}, findings: measured(label) } }),
       compare: async (before, after) => (calls.compares.push([before, after]), verdicts[verdictRun++] || comparison('neutral')),
     },
   };
@@ -112,6 +113,21 @@ test('keeps a change that passes the checks and measures faster, and builds on i
   assert.match(reportMarkdown({ ...s, runs: 3, branch: 'spectra/fix-x' }), /1 of 2 findings fixed/);
 });
 
+test('skips a finding that an earlier kept change already resolved', async () => {
+  // the memoised row also stops the hot file: after the first fix, neither is measured
+  const measured = label => (label === 'baseline' ? bothFindings() : []);
+  const { calls, options } = fakes({ agentChanges: [edit, edit], verdicts: [comparison('improved'), comparison('improved')], measured });
+  let asked = 0;
+  const run = options.agent.run;
+  options.agent.run = (...a) => (asked++, run(...a));
+  const s = await runFix(options);
+  assert.equal(asked, 1, 'the agent is not asked about the resolved finding');
+  assert.deepEqual(s.attempts.map(a => a.outcome), ['kept']);
+  assert.deepEqual(s.skipped, [{ title: 'heavy is slow', reason: 'already resolved by an earlier kept change' }]);
+  assert.equal(calls.profiles, 2, 'baseline and one measurement');
+  assert.match(reportMarkdown({ ...s, runs: 3, branch: 'spectra/fix-x' }), /heavy is slow: already resolved by an earlier kept change/);
+});
+
 test('undoes a change whose checks fail, without measuring it', async () => {
   const { calls, options } = fakes({ agentChanges: [edit], checks: [true, false], verdicts: [] });
   options.workspace.exists = f => f === 'src/Row.jsx';