CodeIssuesDiscussionsWikiPull RequestsProjectsCommitsActionsReleasesContributorsPulse● GatesSecuritySettingsDeploymentsPipelineInsightsAgents✨ Explain✨ Ask AI✨ Workspace✨ Spec✨ Tests▓ Debt Map✨ NL Search🏛 Archaeology
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 | /**
* CI Auto-Fix — when a gate run or workflow run fails on a PR, the repair
* flywheel cache is consulted first (Tier 0); on a miss Claude reads the
* error logs, the failing test file, and the PR diff. Either way a
* ready-to-apply patch is posted as a comment on the PR.
*
* Entry points:
* triggerCiAutofix(gateRunId) — fire-and-forget; call after a gate_run
* row is written with status="failed".
* applyAutofix(prCommentId, userId) — apply the patch from a comment onto
* a new branch and return the branch name. Settles the flywheel entry
* (success/failed) so the cache's confidence accumulates.
*
* Route wiring (src/routes/pulls.tsx or src/routes/api.ts):
* POST /api/pr-comments/:commentId/apply-autofix → applyAutofix
*
* Flywheel wiring (BUILD_BIBLE §7 finding 1): every failure is fingerprinted
* via repair-flywheel.ts; a previously-successful patch with the same
* signature and a good success rate is served WITHOUT an AI call. All served
* fixes are recorded as 'pending' flywheel rows and settled on apply.
* Flywheel/DB errors never break this path — a broken cache degrades to the
* old always-call-AI behaviour.
*/
import { and, eq } from "drizzle-orm";
import { mkdtemp, rm, writeFile } from "fs/promises";
import { join } from "path";
import { tmpdir } from "os";
import { db } from "../db";
import {
findCachedRepair,
recordRepair,
updateOutcome,
type CachedRepair,
type RecordRepairInput,
} from "./repair-flywheel";
import {
gateRuns,
pullRequests,
prComments,
repositories,
users,
repoCollaborators,
} from "../db/schema";
import { getRepoPath } from "../git/repository";
import { getBotUserIdOrFallback } from "./bot-user";
import {
getAnthropic,
isAiAvailable,
MODEL_SONNET,
extractText,
parseJsonResponse,
} from "./ai-client";
import {
getAutomationSettings,
type AutomationSettingsLoader,
} from "./automation-settings";
// ---------------------------------------------------------------------------
// Types
// ---------------------------------------------------------------------------
export interface AutofixResult {
prNumber: number;
repoId: string;
gateRunId: string;
patch: string; // unified diff format
explanation: string; // 2-3 sentence explanation
confidence: "high" | "medium" | "low";
affectedFiles: string[];
}
export interface ClaudeAutofixResponse {
patch: string;
explanation: string;
confidence: "high" | "medium" | "low";
affectedFiles: string[];
}
/**
* DI seam for the repair-flywheel wiring. Production callers omit this and
* get the real implementations; tests inject fakes so no DB is touched.
*/
export interface CiAutofixDeps {
findCachedRepair?: typeof findCachedRepair;
recordRepair?: typeof recordRepair;
updateOutcome?: typeof updateOutcome;
aiAvailable?: () => boolean;
/** Inject the per-repo automation settings loader (tests). */
loadAutomationSettings?: AutomationSettingsLoader;
}
/** The fix triggerCiAutofix decided to post, plus its flywheel bookkeeping. */
export interface AutofixPlan {
/** 'cache' = Tier-0 flywheel replay (no AI call); 'ai' = fresh Sonnet patch. */
source: "cache" | "ai";
fix: ClaudeAutofixResponse;
/** Flywheel row recorded as 'pending' for this attempt; null if recording failed. */
flywheelEntryId: string | null;
/** On a cache hit: the parent pattern that was replayed. */
cachedPatternId: string | null;
}
// ---------------------------------------------------------------------------
// Constants
// ---------------------------------------------------------------------------
/** Idempotency marker embedded in every autofix comment. */
export const CI_AUTOFIX_MARKER = "<!-- gluecron:ci-autofix:v1 -->";
/**
* Marker carrying the pending flywheel entry id, embedded in the autofix
* comment so applyAutofix can settle the outcome (success/failed) later.
*/
export const FLYWHEEL_MARKER_PREFIX = "<!-- gluecron:ci-autofix:flywheel:";
/**
* Minimum settled success rate before a cached pattern is replayed instead
* of calling the AI. Below this the cache is considered unreliable for the
* signature and we fall through to a fresh Sonnet patch.
*/
export const CACHE_MIN_SUCCESS_RATE = 0.5;
/** Max bytes of PR diff sent to Claude. */
const MAX_DIFF_BYTES = 80 * 1024;
/** Max bytes of error log sent to Claude. */
const MAX_LOG_BYTES = 3 * 1024;
/** Max bytes per test file read. */
const MAX_FILE_BYTES = 10 * 1024;
/** Max number of failing test files to read. */
const MAX_TEST_FILES = 3;
// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------
async function spawnGit(
args: string[],
cwd: string
): Promise<{ stdout: string; stderr: string; exitCode: number }> {
const proc = Bun.spawn(["git", ...args], {
cwd,
stdout: "pipe",
stderr: "pipe",
});
const [stdout, stderr] = await Promise.all([
new Response(proc.stdout).text(),
new Response(proc.stderr).text(),
]);
const exitCode = await proc.exited;
return { stdout, stderr, exitCode };
}
function truncate(s: string, maxBytes: number): string {
const buf = Buffer.from(s, "utf8");
if (buf.length <= maxBytes) return s;
return buf.slice(0, maxBytes).toString("utf8") + "\n[truncated]";
}
/**
* Extract failing test file paths, error message, and stack trace from a
* raw CI error log string. Heuristic — good enough for most JS/TS test
* runners (Jest, Vitest, Bun test) and Python pytest output.
*/
function parseErrorLog(errorLog: string): {
testFiles: string[];
errorSummary: string;
} {
const lines = errorLog.split("\n");
// Collect candidate test file paths:
// - lines mentioning .test.ts/.spec.ts/.test.js/.spec.js/.test.py paths
// - lines with "FAIL <path>" (Jest pattern)
const filePatterns = [
/(?:^|\s)([\w./\-]+\.(?:test|spec)\.[jt]sx?)/gm,
/(?:^|\s)([\w./\-]+_test\.py)/gm,
/(?:^FAIL\s+)([\w./\-]+)/gm,
];
const filesSet = new Set<string>();
for (const pattern of filePatterns) {
let m: RegExpExecArray | null;
pattern.lastIndex = 0;
while ((m = pattern.exec(errorLog)) !== null) {
const path = m[1].trim();
if (path && !path.startsWith("-") && !path.startsWith("+")) {
filesSet.add(path);
}
}
}
// Error summary: first 3KB of the log (most runners put the error first).
const errorSummary = truncate(errorLog, MAX_LOG_BYTES);
return {
testFiles: Array.from(filesSet).slice(0, MAX_TEST_FILES),
errorSummary,
};
}
// ---------------------------------------------------------------------------
// Repair flywheel wiring (Tier 0)
// ---------------------------------------------------------------------------
/** Matches the id inside a FLYWHEEL_MARKER_PREFIX comment marker. */
const FLYWHEEL_MARKER_RE = /<!-- gluecron:ci-autofix:flywheel:([0-9a-fA-F-]{8,64}) -->/;
/** Parse the flywheel entry id out of an autofix comment body, if present. */
export function extractFlywheelEntryId(commentBody: string): string | null {
const m = commentBody.match(FLYWHEEL_MARKER_RE);
return m ? m[1] : null;
}
/**
* Record a 'pending' flywheel row for a fix we're about to post. Never
* throws — a flywheel write failure must not stop the fix from shipping,
* it only means this attempt won't contribute to the cache's learning.
*/
async function safeRecordRepair(
input: RecordRepairInput,
deps: CiAutofixDeps
): Promise<string | null> {
const record = deps.recordRepair ?? recordRepair;
try {
return await record(input);
} catch (err) {
console.warn(
"[ci-autofix] flywheel record failed (fix still served):",
err instanceof Error ? err.message : err
);
return null;
}
}
/**
* Settle the flywheel entry referenced by an autofix comment (if any) so
* the pattern's success rate accumulates. Never throws — flywheel
* bookkeeping must not break the apply path.
*/
export async function recordAutofixOutcome(
commentBody: string,
outcome: "success" | "failed",
deps: CiAutofixDeps = {}
): Promise<void> {
const entryId = extractFlywheelEntryId(commentBody);
if (!entryId) return;
const settle = deps.updateOutcome ?? updateOutcome;
try {
await settle(entryId, outcome);
} catch (err) {
console.warn(
"[ci-autofix] flywheel outcome update failed:",
err instanceof Error ? err.message : err
);
}
}
/**
* Tier-0-then-AI resolution for a CI failure (BUILD_BIBLE §7 finding 1).
*
* 1. Tier 0 — consult the repair flywheel for a previously-successful fix
* with the same failure signature. On a usable hit (success rate ≥
* CACHE_MIN_SUCCESS_RATE and a stored patch) the cached patch is
* served directly: no Anthropic call at all.
* 2. Tier 2 — fall through to a fresh Claude Sonnet patch via the
* `generateAiFix` thunk (guarded by isAiAvailable(); the thunk also
* keeps the expensive git context-gathering off the cache-hit path).
*
* Every served fix is recorded as a 'pending' flywheel row; applyAutofix
* settles it via recordAutofixOutcome. Flywheel/DB failures NEVER break
* this path — a broken cache degrades to the old always-call-AI behaviour.
*/
export async function resolveAutofix(args: {
repositoryId: string;
failureText: string;
generateAiFix: () => Promise<ClaudeAutofixResponse | null>;
deps?: CiAutofixDeps;
}): Promise<AutofixPlan | null> {
const deps = args.deps ?? {};
const lookup = deps.findCachedRepair ?? findCachedRepair;
const aiOk = deps.aiAvailable ?? isAiAvailable;
// ── Tier 0: flywheel cache. Fail open — any error counts as a miss.
let cached: CachedRepair | null = null;
if (args.failureText.trim()) {
try {
cached = await lookup(args.repositoryId, args.failureText);
} catch (err) {
console.warn(
"[ci-autofix] flywheel lookup failed (falling through to AI):",
err instanceof Error ? err.message : err
);
}
}
// A usable hit needs a stored full patch (pre-0105 rows have none) AND a
// good settled success rate; anything else falls through to the AI tier.
if (
cached &&
cached.patch?.trim() &&
cached.successRate >= CACHE_MIN_SUCCESS_RATE
) {
const fix: ClaudeAutofixResponse = {
patch: cached.patch,
explanation:
cached.patchSummary ||
"Replayed a previously-successful repair for this failure signature.",
// A pattern that keeps working is high confidence; anything that has
// failed at least occasionally is medium.
confidence: cached.successRate >= 0.9 ? "high" : "medium",
affectedFiles: cached.filesChanged,
};
const entryId = await safeRecordRepair(
{
repositoryId: args.repositoryId,
failureText: args.failureText,
classification: cached.classification,
tier: "cached",
patchSummary: fix.explanation,
// Carry the patch onto the new row so it is itself replayable once
// it settles (findCachedRepair prefers the most recent success).
patch: cached.patch,
filesChanged: fix.affectedFiles,
commitSha: null,
parentPatternId: cached.id,
},
deps
);
// Audit the saved AI call — this line is the flywheel's whole point.
console.log(
`[ci-autofix] flywheel cache HIT — pattern ${cached.id} (success rate ${(cached.successRate * 100).toFixed(0)}%, ${cached.hitCount} prior hits) served without an AI call`
);
return { source: "cache", fix, flywheelEntryId: entryId, cachedPatternId: cached.id };
}
// ── Tier 2: fresh AI patch. All AI features degrade gracefully when no
// API key is configured — cached fixes above still work without one.
if (!aiOk()) return null;
const fix = await args.generateAiFix();
if (!fix || !fix.patch || !fix.explanation) return null;
if (fix.confidence === "low") return null;
const entryId = await safeRecordRepair(
{
repositoryId: args.repositoryId,
failureText: args.failureText,
classification: null,
tier: "ai-sonnet",
patchSummary: fix.explanation,
// Stored so the entry is replayable by the Tier-0 cache once it
// settles to 'success'.
patch: fix.patch,
filesChanged: fix.affectedFiles,
commitSha: null,
},
deps
);
return { source: "ai", fix, flywheelEntryId: entryId, cachedPatternId: null };
}
// ---------------------------------------------------------------------------
// Main entry point
// ---------------------------------------------------------------------------
/**
* Fire-and-forget entry point called after a gate_run row is set to 'failed'.
* Never throws — all errors are swallowed after logging.
*
* Note: no isAiAvailable() gate here — the Tier-0 flywheel cache needs no
* API key, so cached fixes still flow when AI is unconfigured. The actual
* Anthropic call is guarded inside resolveAutofix (graceful degradation).
*/
export async function triggerCiAutofix(
gateRunId: string,
deps: CiAutofixDeps = {}
): Promise<void> {
try {
await _runAutofix(gateRunId, deps);
} catch (err) {
console.error(
"[ci-autofix] crashed:",
err instanceof Error ? err.message : err
);
}
}
async function _runAutofix(
gateRunId: string,
deps: CiAutofixDeps = {}
): Promise<void> {
// 1. Load the gate run
const [gateRun] = await db
.select()
.from(gateRuns)
.where(eq(gateRuns.id, gateRunId))
.limit(1);
if (!gateRun) return;
if (gateRun.status !== "failed") return;
if (!gateRun.pullRequestId) return;
// 2. Load the PR row
const [pr] = await db
.select()
.from(pullRequests)
.where(eq(pullRequests.id, gateRun.pullRequestId))
.limit(1);
if (!pr) return;
// 3. Load repo (owner/name)
const [repoRow] = await db
.select({
id: repositories.id,
name: repositories.name,
diskPath: repositories.diskPath,
ownerId: repositories.ownerId,
ownerUsername: users.username,
})
.from(repositories)
.innerJoin(users, eq(repositories.ownerId, users.id))
.where(eq(repositories.id, gateRun.repositoryId))
.limit(1);
if (!repoRow) return;
// Per-repo automation gate — 'off' skips CI autofix entirely; 'suggest'
// (the fail-open default) posts the patch comment as before; 'auto'
// additionally applies the patch onto a fix/ branch below.
const automation = await (deps.loadAutomationSettings ?? getAutomationSettings)(
repoRow.id
);
if (automation.ciAutofixMode === "off") return;
// 4. Check idempotency — skip if already posted for this gateRunId
const idempotencyMarker = `<!-- gluecron:ci-autofix:run:${gateRunId} -->`;
const existing = await db
.select({ id: prComments.id })
.from(prComments)
.where(
and(
eq(prComments.pullRequestId, gateRun.pullRequestId),
// We check by looking for comments with the autofix marker.
// drizzle doesn't have LIKE with dynamic params easily, but
// we query all AI comments and filter client-side (there won't be many).
eq(prComments.isAiReview, true)
)
)
.limit(50);
for (const row of existing) {
// Load the body to check idempotency marker
const [full] = await db
.select({ body: prComments.body })
.from(prComments)
.where(eq(prComments.id, row.id))
.limit(1);
if (full?.body?.includes(idempotencyMarker)) return;
}
const repoDir = getRepoPath(repoRow.ownerUsername, repoRow.name);
// 5. Failure text — drives both the flywheel signature and the AI prompt.
const errorLog = gateRun.summary || gateRun.details || "";
const failureText =
typeof errorLog === "string" ? errorLog : JSON.stringify(errorLog);
// 6. Tier 0 (flywheel cache) first, then the AI fallback. The expensive
// context gathering (PR diff + failing test files + Sonnet call) lives
// inside the thunk so a cache hit never touches git or the API.
const plan = await resolveAutofix({
repositoryId: repoRow.id,
failureText,
deps,
generateAiFix: async () => {
// 6a. Get the PR diff (max 80KB)
const diffResult = await spawnGit(
["diff", `${pr.baseBranch}...${pr.headBranch}`],
repoDir
);
const prDiff = truncate(diffResult.stdout, MAX_DIFF_BYTES);
if (!prDiff.trim()) return null; // nothing to work with
// 6b. Parse errorLog to extract test files + error summary
const { testFiles, errorSummary } = parseErrorLog(failureText);
// 6c. Read failing test files via git show HEAD:path
let testFileContent = "";
for (const filePath of testFiles) {
const showResult = await spawnGit(
["show", `${pr.headBranch}:${filePath}`],
repoDir
);
if (showResult.exitCode === 0 && showResult.stdout) {
const content = truncate(showResult.stdout, MAX_FILE_BYTES);
testFileContent += `\n\n--- ${filePath} ---\n${content}`;
}
}
// 6d. Call Claude Sonnet 4.6
const client = getAnthropic();
const prompt = `You are a senior engineer fixing a CI failure.
PR diff (what changed):
${prDiff}
Failing test output:
${errorSummary}
Test file content:${testFileContent || "\n(no test files detected)"}
Produce a minimal unified diff patch that fixes the CI failure. The patch must:
1. Be valid unified diff format (--- a/file, +++ b/file, @@ lines)
2. Fix only what's needed — no refactoring
3. Not modify the test itself unless the test expectation is genuinely wrong
Return JSON: {"patch": "...", "explanation": "...", "confidence": "high|medium|low", "affectedFiles": ["..."]}`;
const message = await client.messages.create({
model: MODEL_SONNET,
max_tokens: 4096,
messages: [{ role: "user", content: prompt }],
});
const rawText = extractText(message);
return parseJsonResponse<ClaudeAutofixResponse>(rawText);
},
});
// No usable fix (cache miss + AI declined/low-confidence/unavailable).
if (!plan) return;
// 7. Build and post the comment
const commentBody = buildAutofixComment(plan, idempotencyMarker, gateRunId);
const botAuthorId = await getBotUserIdOrFallback(repoRow.id);
if (!botAuthorId) return;
const [posted] = await db
.insert(prComments)
.values({
pullRequestId: gateRun.pullRequestId,
authorId: botAuthorId,
body: commentBody,
isAiReview: true,
})
.returning({ id: prComments.id });
// 'auto' mode — also apply the patch onto a fix/ branch via the same
// path the Apply Fix button drives. Best-effort: an apply failure (it
// also settles the flywheel entry as 'failed') leaves the suggest-mode
// comment in place for a human to act on.
if (automation.ciAutofixMode === "auto" && posted) {
try {
const { branchName } = await applyAutofix(posted.id, repoRow.ownerId, deps);
console.log(
`[ci-autofix] auto-applied patch from comment ${posted.id} onto ${branchName}`
);
} catch (err) {
console.warn(
"[ci-autofix] auto-apply failed (patch comment still posted):",
err instanceof Error ? err.message : err
);
}
}
}
function buildAutofixComment(
plan: AutofixPlan,
idempotencyMarker: string,
gateRunId: string
): string {
const result = plan.fix;
const confidenceBadge =
result.confidence === "high"
? "🟢 High confidence"
: result.confidence === "medium"
? "🟡 Medium confidence"
: "🔴 Low confidence";
// Flywheel bookkeeping marker — applyAutofix parses this to settle the
// pending entry's outcome. Absent when the flywheel write failed.
const flywheelMarker = plan.flywheelEntryId
? `\n${FLYWHEEL_MARKER_PREFIX}${plan.flywheelEntryId} -->`
: "";
const sourceNote =
plan.source === "cache"
? `\n\n♻️ **Served from the repair cache** — this failure signature was fixed successfully before; no AI call was made.`
: "";
return `${CI_AUTOFIX_MARKER}
${idempotencyMarker}${flywheelMarker}
## 🔧 AI Auto-Fix
${result.explanation}${sourceNote}
**Confidence:** ${confidenceBadge}
\`\`\`diff
${result.patch}
\`\`\`
<details><summary>Apply this fix</summary>
Copy the patch above or click **Apply Fix** to commit it automatically.
<form method="post" action="/api/pr-comments/COMMENT_ID/apply-autofix" style="display:inline">
<button type="submit" style="margin-top:8px;padding:6px 14px;background:#6c63ff;color:#fff;border:none;border-radius:6px;cursor:pointer">
⚡ Apply Fix
</button>
</form>
</details>
<sub>Gate run: <code>${gateRunId}</code> · Affected files: ${result.affectedFiles.join(", ") || "see patch above"}</sub>`;
}
// ---------------------------------------------------------------------------
// Apply autofix
// ---------------------------------------------------------------------------
/**
* Applies the patch from a PR comment onto a new branch.
* Returns the new branch name so the caller can redirect to compare view.
*
* Also settles the comment's pending flywheel entry: a clean apply+commit
* records 'success' (the pattern becomes replayable by the Tier-0 cache),
* an apply failure records 'failed' so unreliable patterns lose confidence.
*/
export async function applyAutofix(
prCommentId: string,
userId: string,
deps: CiAutofixDeps = {}
): Promise<{ branchName: string }> {
// 1. Load the comment
const [comment] = await db
.select({
id: prComments.id,
pullRequestId: prComments.pullRequestId,
body: prComments.body,
isAiReview: prComments.isAiReview,
})
.from(prComments)
.where(eq(prComments.id, prCommentId))
.limit(1);
if (!comment) throw new Error("Comment not found");
if (!comment.body.includes(CI_AUTOFIX_MARKER)) {
throw new Error("Not an autofix comment");
}
// 2. Load PR + repo for access check
const [pr] = await db
.select()
.from(pullRequests)
.where(eq(pullRequests.id, comment.pullRequestId))
.limit(1);
if (!pr) throw new Error("PR not found");
const [repoRow] = await db
.select({
id: repositories.id,
name: repositories.name,
ownerId: repositories.ownerId,
ownerUsername: users.username,
})
.from(repositories)
.innerJoin(users, eq(repositories.ownerId, users.id))
.where(eq(repositories.id, pr.repositoryId))
.limit(1);
if (!repoRow) throw new Error("Repository not found");
// Verify write access: must be repo owner or collaborator
const isOwner = repoRow.ownerId === userId;
if (!isOwner) {
const [collab] = await db
.select({ id: repoCollaborators.id })
.from(repoCollaborators)
.where(
and(
eq(repoCollaborators.repositoryId, repoRow.id),
eq(repoCollaborators.userId, userId)
)
)
.limit(1);
if (!collab) throw new Error("Forbidden: no write access");
}
// 3. Extract patch from comment body
const patchMatch = comment.body.match(/```diff\n([\s\S]*?)```/);
if (!patchMatch) throw new Error("No patch found in comment");
const patch = patchMatch[1];
// 4. Create a new branch from the PR's head
const branchName = `fix/autofix-${Date.now()}`;
const repoDir = getRepoPath(repoRow.ownerUsername, repoRow.name);
// Create the branch at the PR head SHA
const headSha = await spawnGit(
["rev-parse", pr.headBranch],
repoDir
);
if (headSha.exitCode !== 0) throw new Error("Cannot resolve head branch");
await spawnGit(
["branch", branchName, headSha.stdout.trim()],
repoDir
);
// 5. Apply the patch via git apply in a temp worktree
const tmpDir = await mkdtemp(join(tmpdir(), "autofix-"));
try {
// Add worktree for the new branch
const wtResult = await spawnGit(
["worktree", "add", tmpDir, branchName],
repoDir
);
if (wtResult.exitCode !== 0) {
throw new Error(`git worktree add failed: ${wtResult.stderr}`);
}
// Write the patch to a temp file
const patchFile = join(tmpDir, "autofix.patch");
await writeFile(patchFile, patch, "utf8");
// Apply the patch
const applyResult = await spawnGit(
["apply", "--index", patchFile],
tmpDir
);
if (applyResult.exitCode !== 0) {
throw new Error(`git apply failed: ${applyResult.stderr}`);
}
// 6. Commit
const commitResult = await spawnGit(
[
"commit",
"-m",
"fix: apply AI autofix for CI failure",
"--author",
"gluecron[bot] <bot@gluecron.com>",
],
tmpDir
);
if (commitResult.exitCode !== 0) {
throw new Error(`git commit failed: ${commitResult.stderr}`);
}
// 7. Push the branch back to the bare repo
// In a bare-repo + worktree setup the push target is the bare repo itself.
await spawnGit(
["push", repoDir, `HEAD:refs/heads/${branchName}`],
tmpDir
);
// 8. The repair landed — settle the flywheel entry so this pattern's
// success rate climbs and future identical failures hit the Tier-0 cache.
await recordAutofixOutcome(comment.body, "success", deps);
} catch (err) {
// The patch failed to apply/commit/push — settle as 'failed' so the
// flywheel learns this pattern is unreliable. Best-effort: the original
// error is always rethrown for the route to surface.
await recordAutofixOutcome(comment.body, "failed", deps);
throw err;
} finally {
// Cleanup worktree
await spawnGit(["worktree", "remove", "--force", tmpDir], repoDir).catch(
() => {}
);
await rm(tmpDir, { recursive: true, force: true }).catch(() => {});
}
return { branchName };
}
|