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 | /**
* AI CI Healer — autonomous failure → root-cause → patch PR loop.
*
* When a `workflow_runs` row lands in status="failure", this module:
* 1. Pulls the run + all its `workflow_jobs` (especially the failed ones'
* `logs` column).
* 2. Asks Claude to identify the root cause + whether it's fixable from
* inside this repo.
* 3. If patchable, hands the suggested fixes off to the existing
* `generatePatchForGateTestFinding` (the finding-shape maps cleanly
* onto its `GateTestFinding[]` API — both surfaces want
* `{ path, description, severity }`).
* 4. Records an `ai.ci.healed` or `ai.ci.gave_up` audit row keyed on the
* run id so the autopilot poller doesn't re-process the same failure
* every 5 minutes.
*
* Degrades silently when ANTHROPIC_API_KEY is unset (autopilot also gates
* on the env var, but the lib double-checks so it's safe to call directly).
* Everything is wrapped in try/catch — `analyzeFailedWorkflowRun` returns
* `null` rather than throwing on any failure.
*
* Marker convention:
* - Successful patch open → audit action `ai.ci.healed`, targetId = runId.
* - Claude says unfixable → audit action `ai.ci.gave_up`, targetId = runId.
* The autopilot poller skips any run that already has either marker, so we
* never retry forever.
*/
import { and, desc, eq, gte, lt, sql } from "drizzle-orm";
import type Anthropic from "@anthropic-ai/sdk";
import { db } from "../db";
import {
auditLog,
repositories,
workflowJobs,
workflowRuns,
workflows,
} from "../db/schema";
import {
MODEL_SONNET,
extractText,
getAnthropic,
isAiAvailable,
parseJsonResponse,
} from "./ai-client";
import { audit } from "./notify";
import {
generatePatchForGateTestFinding,
type GateTestFinding,
} from "./ai-patch-generator";
import { assertAiQuota, AiQuotaExceededError } from "./billing";
// ---------------------------------------------------------------------------
// Tunables
// ---------------------------------------------------------------------------
/** Cap per-job log slice we ship to Claude — protect the prompt budget. */
const LOG_SNIPPET_BYTES = 8 * 1024;
/** Hard cap on runs processed per autopilot tick — runaway protection. */
const HEAL_CAP_PER_TICK = 5;
/** Only consider runs that finished at least this long ago. Gives the
* workflow runner time to flush logs + final job rows before we sample. */
const HEAL_MIN_AGE_MS = 60 * 1_000;
/** Don't bother healing ancient failures — older than this and we assume
* the human has already triaged or the SHA has been rewritten. */
const HEAL_MAX_AGE_MS = 24 * 60 * 60 * 1_000;
// ---------------------------------------------------------------------------
// Public surface
// ---------------------------------------------------------------------------
export interface SuggestedFix {
/** Relative path inside the repo that the AI thinks needs touching. */
path: string;
/** What's wrong / what the fix should look like (one-liner). */
description: string;
/** Severity inherited from the AI's confidence call. */
severity?: "low" | "medium" | "high" | "critical";
}
export interface CiHealAnalysis {
/** Plain-English root cause, 1-3 sentences. */
rootCause: string;
/** Concrete fix proposals (may be empty if Claude can't pinpoint files). */
suggestedFixes: SuggestedFix[];
/** Convenience: the paths from `suggestedFixes` deduped. */
patchablePaths: string[];
}
export interface CiHealerTickSummary {
considered: number;
healed: number;
gaveUp: number;
skipped: number;
}
/**
* Claude's response shape. Kept loose so we tolerate minor schema drift.
*/
interface ClaudeCiResponse {
rootCause?: string;
fixable?: boolean;
suggestedFixes?: Array<{
path?: string;
description?: string;
severity?: string;
}>;
/** Optional: Claude's reason it gave up. We just persist it for ops. */
unfixableReason?: string;
}
// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------
function truncate(s: string | null | undefined, max: number): string {
if (!s) return "";
if (s.length <= max) return s;
return s.slice(0, max) + "\n…(truncated)";
}
function normaliseSeverity(s: unknown): SuggestedFix["severity"] | undefined {
if (typeof s !== "string") return undefined;
const v = s.toLowerCase();
if (v === "low" || v === "medium" || v === "high" || v === "critical") {
return v;
}
return undefined;
}
/**
* Map our AI-derived `SuggestedFix` set onto the
* `GateTestFinding[]` shape that `generatePatchForGateTestFinding`
* already accepts. The patch generator only needs `{path, description,
* severity}` — exactly what we already have. Each fix becomes one
* candidate finding; the generator stops at the first one that produces a
* non-empty patch set so a bad suggestion doesn't drown a good one.
*/
export function fixesToFindings(
fixes: SuggestedFix[],
runShortId: string
): GateTestFinding[] {
return fixes
.filter((f) => f.path && f.path.trim().length > 0)
.map((f, i) => ({
id: `ci-heal-${runShortId}-${i}`,
ruleId: "ci-failure",
path: f.path.trim(),
severity: f.severity || "high",
title: "CI failure auto-heal",
description: f.description || "CI failure",
}));
}
/**
* Build the Claude prompt. Pure function so tests can pin the shape.
*/
export function buildCiHealPrompt(args: {
repoFullName: string;
commitSha: string;
workflowYaml: string;
failedJobs: Array<{ name: string; conclusion: string | null; logs: string }>;
}): string {
const jobsBlock = args.failedJobs
.map(
(j) =>
`### Job: ${j.name} (${j.conclusion || "failure"})\n\`\`\`\n${truncate(
j.logs,
LOG_SNIPPET_BYTES
)}\n\`\`\``
)
.join("\n\n");
return [
"You are GlueCron's CI healer. A workflow run just failed. Decide whether the failure is a code bug fixable in THIS repository, and if so propose concrete file edits.",
"",
`**Repository:** ${args.repoFullName}`,
`**Commit:** ${args.commitSha.slice(0, 12)}`,
"",
"## Workflow YAML",
"```yaml",
truncate(args.workflowYaml, 4_000),
"```",
"",
"## Failed job logs",
jobsBlock || "(no failed-job logs available)",
"",
"Respond ONLY with JSON of this exact shape:",
"{",
' "rootCause": "1-3 sentence diagnosis (what failed, why)",',
' "fixable": true | false,',
' "suggestedFixes": [',
' { "path": "src/foo.ts", "description": "what to change", "severity": "high" }',
" ],",
' "unfixableReason": "(only if fixable=false) why the human must intervene"',
"}",
"",
"Rules:",
"- `fixable` MUST be false if the failure is an env/infra/external-service problem (missing secret, registry down, network, GitHub Actions runner image, etc.) — anything you can't fix by editing files in this repo.",
"- `suggestedFixes` must be empty when fixable=false.",
"- Only suggest paths you can identify with high confidence from the logs or YAML. Do not guess at random files.",
"- Keep `description` short — the patch generator will be invoked with this finding to produce the actual diff.",
].join("\n");
}
interface FailedJobRow {
name: string;
conclusion: string | null;
logs: string;
}
async function loadFailedJobs(runId: string): Promise<FailedJobRow[]> {
try {
const rows = await db
.select({
name: workflowJobs.name,
conclusion: workflowJobs.conclusion,
logs: workflowJobs.logs,
status: workflowJobs.status,
})
.from(workflowJobs)
.where(eq(workflowJobs.runId, runId));
return rows
.filter((r) => r.status === "failure" || r.conclusion === "failure")
.map((r) => ({
name: r.name,
conclusion: r.conclusion,
logs: r.logs || "",
}));
} catch (err) {
console.error("[ai-ci-healer] loadFailedJobs failed:", err);
return [];
}
}
/**
* Has this run already been processed (success or give-up)? We use the
* audit log as the marker store so we don't need a schema change.
*/
async function hasMarker(runId: string): Promise<boolean> {
try {
const [row] = await db
.select({ id: auditLog.id })
.from(auditLog)
.where(
and(
eq(auditLog.targetType, "workflow_run"),
eq(auditLog.targetId, runId),
sql`${auditLog.action} IN ('ai.ci.healed', 'ai.ci.gave_up')`
)
)
.limit(1);
return !!row;
} catch (err) {
console.warn("[ai-ci-healer] hasMarker query failed:", err);
// Fail closed — if we can't check, skip to avoid duplicate work.
return true;
}
}
// ---------------------------------------------------------------------------
// analyzeFailedWorkflowRun — public entry point #1
// ---------------------------------------------------------------------------
export interface AnalyzeOptions {
/** Test-only Anthropic client injection. */
client?: Pick<Anthropic, "messages">;
}
/**
* Diagnose a failed run. Returns `null` when:
* - The run doesn't exist or isn't actually a failure.
* - ANTHROPIC_API_KEY is unset AND no client was injected.
* - Claude says the failure isn't fixable from this repo (env/infra).
* - Any DB / network step blows up (logged, swallowed).
*/
export async function analyzeFailedWorkflowRun(
runId: string,
opts: AnalyzeOptions = {}
): Promise<CiHealAnalysis | null> {
// Resolve client lazily — tests inject, production reads env.
let client: Pick<Anthropic, "messages">;
if (opts.client) {
client = opts.client;
} else {
if (!isAiAvailable()) return null;
try {
client = getAnthropic();
} catch {
return null;
}
}
// Load run row.
let run: typeof workflowRuns.$inferSelect | null = null;
try {
const [row] = await db
.select()
.from(workflowRuns)
.where(eq(workflowRuns.id, runId))
.limit(1);
run = row || null;
} catch (err) {
console.error("[ai-ci-healer] loadRun failed:", err);
return null;
}
if (!run || run.status !== "failure") return null;
// Load workflow + repo (for YAML + naming context).
let workflowYaml = "";
let repoFullName = "unknown/unknown";
try {
const [w] = await db
.select({ yaml: workflows.yaml })
.from(workflows)
.where(eq(workflows.id, run.workflowId))
.limit(1);
if (w?.yaml) workflowYaml = w.yaml;
} catch (err) {
console.warn("[ai-ci-healer] load workflow failed:", err);
}
try {
const [r] = await db
.select({
name: repositories.name,
ownerId: repositories.ownerId,
})
.from(repositories)
.where(eq(repositories.id, run.repositoryId))
.limit(1);
if (r) {
repoFullName = `${r.ownerId.slice(0, 8)}/${r.name}`;
}
} catch (err) {
console.warn("[ai-ci-healer] load repo failed:", err);
}
const failedJobs = await loadFailedJobs(runId);
// Ask Claude.
let parsed: ClaudeCiResponse | null = null;
try {
const message = await client.messages.create({
model: MODEL_SONNET,
max_tokens: 2048,
messages: [
{
role: "user",
content: buildCiHealPrompt({
repoFullName,
commitSha: run.commitSha || "(unknown)",
workflowYaml,
failedJobs,
}),
},
],
});
try {
const { recordAiCost, extractUsage } = await import("./ai-cost-tracker");
const usage = extractUsage(message);
await recordAiCost({
repositoryId: run.repositoryId ?? null,
model: MODEL_SONNET,
inputTokens: usage.input,
outputTokens: usage.output,
category: "ci_healer",
sourceId: runId,
sourceKind: "workflow_run",
});
} catch {
/* swallow — best-effort */
}
parsed = parseJsonResponse<ClaudeCiResponse>(extractText(message));
} catch (err) {
console.warn(
"[ai-ci-healer] Claude call failed:",
err instanceof Error ? err.message : err
);
return null;
}
if (!parsed) return null;
const rootCause =
typeof parsed.rootCause === "string" && parsed.rootCause.trim()
? parsed.rootCause.trim()
: "(no root cause provided)";
// "Not fixable" branch — return null so the caller can mark `ai.ci.gave_up`.
if (parsed.fixable === false) {
return null;
}
const rawFixes = Array.isArray(parsed.suggestedFixes)
? parsed.suggestedFixes
: [];
const suggestedFixes: SuggestedFix[] = rawFixes
.filter(
(f): f is { path: string; description?: string; severity?: string } =>
!!f && typeof f.path === "string" && f.path.trim().length > 0
)
.map((f) => ({
path: f.path.trim(),
description:
typeof f.description === "string" && f.description.trim()
? f.description.trim()
: rootCause,
severity: normaliseSeverity(f.severity),
}));
// Claude said fixable but produced zero usable paths → treat as
// unfixable so we don't loop.
if (suggestedFixes.length === 0) return null;
const patchablePaths = Array.from(new Set(suggestedFixes.map((f) => f.path)));
return { rootCause, suggestedFixes, patchablePaths };
}
// ---------------------------------------------------------------------------
// healOneRun — drives one run end-to-end. Public so callers (autopilot,
// tests, manual retrigger) share the same pipeline.
// ---------------------------------------------------------------------------
export interface HealOneOptions extends AnalyzeOptions {
/** Test override — pin the patch generator output. */
generatePatch?: typeof generatePatchForGateTestFinding;
}
export interface HealOneResult {
outcome: "healed" | "gave_up" | "skipped";
prNumber?: number;
branch?: string;
reason?: string;
}
export async function healOneRun(
runId: string,
opts: HealOneOptions = {}
): Promise<HealOneResult> {
if (process.env.AUTOPILOT_DISABLED === "1") {
return { outcome: "skipped", reason: "autopilot disabled" };
}
if (!opts.client && !isAiAvailable()) {
return { outcome: "skipped", reason: "ANTHROPIC_API_KEY missing" };
}
// Skip if already processed (audit marker present).
if (await hasMarker(runId)) {
return { outcome: "skipped", reason: "already processed" };
}
// Look up the run for audit metadata (repo + sha) and owner for quota check.
let repositoryId: string | null = null;
let commitSha: string | null = null;
let repoOwnerId: string | null = null;
try {
const [row] = await db
.select({
repositoryId: workflowRuns.repositoryId,
commitSha: workflowRuns.commitSha,
})
.from(workflowRuns)
.where(eq(workflowRuns.id, runId))
.limit(1);
if (row) {
repositoryId = row.repositoryId;
commitSha = row.commitSha;
}
} catch (err) {
console.warn("[ai-ci-healer] post-analyze run lookup failed:", err);
}
// Resolve repo owner for the quota check.
if (repositoryId) {
try {
const [repoRow] = await db
.select({ ownerId: repositories.ownerId })
.from(repositories)
.where(eq(repositories.id, repositoryId))
.limit(1);
if (repoRow) repoOwnerId = repoRow.ownerId;
} catch {
/* tolerate */
}
}
// Hard quota gate — skip silently when the repo owner's AI budget is
// exhausted. We don't fail the CI run; the autopilot marker is not written
// so the healer will retry on the next tick once budget resets.
if (repoOwnerId) {
try {
await assertAiQuota(repoOwnerId);
} catch (err) {
if (err instanceof AiQuotaExceededError) {
console.log(
`[ai-ci-healer] skipping run=${runId}: AI quota exceeded for owner=${repoOwnerId}`
);
return { outcome: "skipped", reason: "AI quota exceeded" };
}
// Unexpected error — log and proceed (fail open).
console.warn("[ai-ci-healer] assertAiQuota failed unexpectedly:", err);
}
}
const analysis = await analyzeFailedWorkflowRun(runId, {
client: opts.client,
});
if (!analysis) {
await audit({
userId: null,
repositoryId,
action: "ai.ci.gave_up",
targetType: "workflow_run",
targetId: runId,
metadata: { commitSha, reason: "unfixable or analysis returned null" },
});
return { outcome: "gave_up", reason: "unfixable" };
}
if (!repositoryId || !commitSha) {
// No base sha → can't seed a patch branch. Mark gave_up so we don't loop.
await audit({
userId: null,
repositoryId,
action: "ai.ci.gave_up",
targetType: "workflow_run",
targetId: runId,
metadata: {
commitSha,
reason: "missing repositoryId or commitSha for patch branch",
},
});
return { outcome: "gave_up", reason: "missing base sha" };
}
const generator = opts.generatePatch ?? generatePatchForGateTestFinding;
const findings = fixesToFindings(analysis.suggestedFixes, runId.slice(0, 8));
const patch = await generator({
repositoryId,
baseSha: commitSha,
findings,
client: opts.client,
});
if (!patch) {
await audit({
userId: null,
repositoryId,
action: "ai.ci.gave_up",
targetType: "workflow_run",
targetId: runId,
metadata: {
commitSha,
reason: "patch generator returned null",
rootCause: analysis.rootCause,
patchablePaths: analysis.patchablePaths,
},
});
return { outcome: "gave_up", reason: "patch generator returned null" };
}
await audit({
userId: null,
repositoryId,
action: "ai.ci.healed",
targetType: "workflow_run",
targetId: runId,
metadata: {
commitSha,
rootCause: analysis.rootCause,
patchablePaths: analysis.patchablePaths,
prNumber: patch.prNumber,
branch: patch.branch,
},
});
return {
outcome: "healed",
prNumber: patch.prNumber,
branch: patch.branch,
};
}
// ---------------------------------------------------------------------------
// runCiHealerTick — public autopilot entry point
// ---------------------------------------------------------------------------
export interface CiHealerTickDeps {
/** Inject a candidate finder for tests. */
findCandidates?: (limit: number) => Promise<{ id: string }[]>;
/** Inject the per-run handler for tests. */
healOne?: (runId: string) => Promise<HealOneResult>;
/** Inject the per-tick cap. */
cap?: number;
}
/**
* Find failed workflow runs that:
* - finished at least HEAL_MIN_AGE_MS ago (let the runner flush logs)
* - finished less than HEAL_MAX_AGE_MS ago (don't chase ancient failures)
* - do NOT yet have an ai.ci.healed / ai.ci.gave_up audit marker
*
* The `hasMarker` filter is applied per-row in `healOneRun` rather than
* the SQL select — keeps the query simple and the index on (status,
* createdAt) doing most of the work.
*/
async function defaultFindCandidates(
limit: number
): Promise<{ id: string }[]> {
const now = Date.now();
const cutoffNew = new Date(now - HEAL_MIN_AGE_MS);
const cutoffOld = new Date(now - HEAL_MAX_AGE_MS);
try {
const rows = await db
.select({ id: workflowRuns.id })
.from(workflowRuns)
.where(
and(
eq(workflowRuns.status, "failure"),
lt(workflowRuns.createdAt, cutoffNew),
gte(workflowRuns.createdAt, cutoffOld)
)
)
.orderBy(desc(workflowRuns.createdAt))
.limit(limit);
return rows;
} catch (err) {
console.error("[ai-ci-healer] candidate query failed:", err);
return [];
}
}
/**
* One autopilot tick: scan recent failures, heal what we can. Returns a
* counts summary. Never throws.
*
* No-op when:
* - AUTOPILOT_DISABLED=1
* - ANTHROPIC_API_KEY is unset
*/
export async function runCiHealerTick(
deps: CiHealerTickDeps = {}
): Promise<CiHealerTickSummary> {
if (process.env.AUTOPILOT_DISABLED === "1") {
return { considered: 0, healed: 0, gaveUp: 0, skipped: 0 };
}
if (!isAiAvailable()) {
return { considered: 0, healed: 0, gaveUp: 0, skipped: 0 };
}
const findCandidates = deps.findCandidates ?? defaultFindCandidates;
const healOne = deps.healOne ?? ((id: string) => healOneRun(id));
const cap = deps.cap ?? HEAL_CAP_PER_TICK;
let candidates: { id: string }[] = [];
try {
candidates = await findCandidates(cap);
} catch (err) {
console.error("[ai-ci-healer] findCandidates threw:", err);
return { considered: 0, healed: 0, gaveUp: 0, skipped: 0 };
}
let healed = 0;
let gaveUp = 0;
let skipped = 0;
for (const cand of candidates) {
try {
const res = await healOne(cand.id);
if (res.outcome === "healed") healed += 1;
else if (res.outcome === "gave_up") gaveUp += 1;
else skipped += 1;
} catch (err) {
skipped += 1;
console.error(
`[ai-ci-healer] per-run failure for run=${cand.id}:`,
err instanceof Error ? err.message : err
);
}
}
return { considered: candidates.length, healed, gaveUp, skipped };
}
// ---------------------------------------------------------------------------
// Test-only re-exports
// ---------------------------------------------------------------------------
export const __test = {
loadFailedJobs,
hasMarker,
defaultFindCandidates,
normaliseSeverity,
truncate,
HEAL_MIN_AGE_MS,
HEAL_MAX_AGE_MS,
HEAL_CAP_PER_TICK,
};
|