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 | /**
* Block M3 — AI pre-merge risk score.
*
* For every open PR we compute a transparent, auditable 0-10 risk score
* the reviewer can see at a glance before clicking Merge. The score is a
* pure function over a small set of signals (file count, line count,
* teams affected, schema migrations touched, dependency churn, test
* ratio). The LLM (Sonnet 4.6) only writes the one-paragraph
* prose summary; it never influences the numeric score.
*
* Architecture:
*
* 1. `computePrRiskScore(signals)` — pure helper. Same input always
* yields the same score. Documented inline; auditable.
* 2. `generatePrRiskSummary(args)` — calls Sonnet for prose. Never
* throws — falls back to a deterministic sentence when no API key
* is set or the call fails.
* 3. `computePrRiskForPullRequest(prId)` — DB-backed orchestrator.
* Resolves the PR + head SHA, gathers signals, computes the score,
* persists it to `pr_risk_scores`, returns the full row.
* 4. `getCachedPrRisk(prId)` — cheap lookup for the current head SHA.
*
* Cache key: `(pull_request_id, commit_sha)`. When the head branch is
* force-pushed (new SHA), the cache miss naturally re-triggers a score
* computation.
*/
import { and, desc, eq } from "drizzle-orm";
import { db } from "../db";
import {
codeOwners,
prRiskScores,
pullRequests,
repoDependencies,
repositories,
users,
type PrRiskScoreRow,
} from "../db/schema";
import { getDiff, getRepoPath, resolveRef } from "../git/repository";
import {
getAnthropic,
isAiAvailable,
MODEL_HAIKU,
MODEL_SONNET,
extractText,
} from "./ai-client";
import { ownersForPath, parseCodeowners, type OwnerRule } from "./codeowners";
// ---------------------------------------------------------------------------
// Public types
// ---------------------------------------------------------------------------
export type PrRiskBand = "low" | "medium" | "high" | "critical";
export interface PrRiskSignals {
filesChanged: number;
linesAdded: number;
linesRemoved: number;
/** Distinct CODEOWNERS owners touched (count). */
teamsAffected: number;
schemaMigrationTouched: boolean;
/** Paths under sensitive/ etc. */
lockedPathTouched: boolean;
addsNewDependency: boolean;
bumpsMajorDependency: boolean;
testsAddedForNewCode: boolean;
/** 0..1, 0 = lots of tests, 1 = no tests. */
diffMinusTestRatio: number;
}
export interface PrRiskScore {
score: number;
band: PrRiskBand;
signals: PrRiskSignals;
aiSummary: string;
generatedAt: Date;
commitSha: string;
}
// ---------------------------------------------------------------------------
// Pure score formula
// ---------------------------------------------------------------------------
/**
* Transparent risk formula (intentionally simple, intentionally not LLM):
*
* + 0.3 per file changed (cap at 3.0)
* + 0.005 per (linesAdded + linesRemoved) (cap at 2.0)
* + 0.8 per team beyond the first (cap at 2.0)
* + 1.5 if schemaMigrationTouched
* + 1.0 if lockedPathTouched
* + 0.5 if addsNewDependency
* + 1.2 if bumpsMajorDependency
* + 1.0 * (testsAddedForNewCode ? 0 : diffMinusTestRatio)
*
* Clamp to [0, 10] and round to nearest integer.
*
* Band thresholds:
* 0–2 → low
* 3–4 → medium
* 5–7 → high
* 8–10 → critical
*
* Monotonicity: every weight is non-negative, so flipping any boolean
* signal from false to true (or increasing any counter) can only raise
* the score (until the cap). Verified by tests.
*/
export function computePrRiskScore(
signals: PrRiskSignals
): { score: number; band: PrRiskBand } {
let raw = 0;
// File count contribution (cap at 3.0).
raw += Math.min(3.0, Math.max(0, signals.filesChanged) * 0.3);
// Line count contribution (cap at 2.0).
const totalLines =
Math.max(0, signals.linesAdded) + Math.max(0, signals.linesRemoved);
raw += Math.min(2.0, totalLines * 0.005);
// Teams-beyond-the-first contribution (cap at 2.0). teamsAffected=0 or 1
// contributes nothing; each additional team adds 0.8.
const extraTeams = Math.max(0, signals.teamsAffected - 1);
raw += Math.min(2.0, extraTeams * 0.8);
// Boolean signals.
if (signals.schemaMigrationTouched) raw += 1.5;
if (signals.lockedPathTouched) raw += 1.0;
if (signals.addsNewDependency) raw += 0.5;
if (signals.bumpsMajorDependency) raw += 1.2;
// Test-coverage penalty.
if (!signals.testsAddedForNewCode) {
const ratio = clamp01(signals.diffMinusTestRatio);
raw += 1.0 * ratio;
}
const score = Math.max(0, Math.min(10, Math.round(raw)));
return { score, band: bandFor(score) };
}
function clamp01(n: number): number {
if (!Number.isFinite(n)) return 0;
return Math.max(0, Math.min(1, n));
}
function bandFor(score: number): PrRiskBand {
if (score <= 2) return "low";
if (score <= 4) return "medium";
if (score <= 7) return "high";
return "critical";
}
// ---------------------------------------------------------------------------
// AI prose summary — Haiku call with deterministic fallback
// ---------------------------------------------------------------------------
/**
* Generate a 1-3 sentence prose summary of the risk profile. Calls Haiku
* for fluency; on any failure (no key, API error, malformed response)
* falls back to a deterministic sentence built from the signal map.
*
* Never throws. Always returns a non-empty string.
*/
export async function generatePrRiskSummary(args: {
signals: PrRiskSignals;
title: string;
baseBranch: string;
headBranch: string;
}): Promise<string> {
const fallback = deterministicSummary(args.signals);
if (!isAiAvailable()) return fallback;
try {
const client = getAnthropic();
const message = await client.messages.create({
model: MODEL_SONNET,
max_tokens: 200,
messages: [
{
role: "user",
content: `You are summarising the risk profile of a pull request for a reviewer who is about to click Merge.
PR title: ${args.title}
Base branch: ${args.baseBranch}
Head branch: ${args.headBranch}
Signals (these are the FACTS — do not invent any others):
- files changed: ${args.signals.filesChanged}
- lines added: ${args.signals.linesAdded}
- lines removed: ${args.signals.linesRemoved}
- distinct CODEOWNERS owners touched: ${args.signals.teamsAffected}
- schema migration touched: ${args.signals.schemaMigrationTouched}
- locked / sensitive path touched: ${args.signals.lockedPathTouched}
- adds a new dependency: ${args.signals.addsNewDependency}
- bumps a major dependency: ${args.signals.bumpsMajorDependency}
- tests added for new code: ${args.signals.testsAddedForNewCode}
- diff-minus-test ratio (0=lots of tests, 1=no tests): ${args.signals.diffMinusTestRatio.toFixed(2)}
Write ONE to THREE plain sentences describing the riskiest aspects of this PR.
Tone: blunt, factual, no marketing. Do not include a numeric score (the reviewer
already sees that). Do not use bullet points. Output the prose only.`,
},
],
});
const text = extractText(message).trim();
if (!text) return fallback;
return text.slice(0, 600);
} catch {
return fallback;
}
}
/**
* Deterministic prose composed from the signal map. Used as the always-
* available fallback. Public-internal — exported for tests.
*/
export function deterministicSummary(signals: PrRiskSignals): string {
const parts: string[] = [];
parts.push(
`Touches ${signals.filesChanged} file(s) with ${signals.linesAdded} added and ${signals.linesRemoved} removed across ${signals.teamsAffected} owner(s).`
);
const hotspots: string[] = [];
if (signals.schemaMigrationTouched) hotspots.push("a schema migration");
if (signals.lockedPathTouched) hotspots.push("a locked / sensitive path");
if (signals.bumpsMajorDependency) hotspots.push("a major dependency bump");
else if (signals.addsNewDependency) hotspots.push("a new dependency");
if (hotspots.length > 0) {
parts.push(`Includes ${joinList(hotspots)}.`);
}
if (!signals.testsAddedForNewCode && signals.diffMinusTestRatio > 0.5) {
parts.push("No tests were added for the new code.");
}
return parts.join(" ");
}
function joinList(items: string[]): string {
if (items.length === 0) return "";
if (items.length === 1) return items[0];
if (items.length === 2) return `${items[0]} and ${items[1]}`;
return `${items.slice(0, -1).join(", ")}, and ${items[items.length - 1]}`;
}
// ---------------------------------------------------------------------------
// Signal-gathering helpers (DB + git)
// ---------------------------------------------------------------------------
const SCHEMA_MIGRATION_PATTERNS = [
/^drizzle\//i,
/^db\/migrations\//i,
/^migrations\//i,
/\/migrations\//i,
/^src\/db\/schema\.ts$/i,
];
const LOCKED_PATH_PATTERNS = [
/^sensitive\//i,
/^secrets\//i,
/^infra\/secrets\//i,
/\.pem$/i,
/\.key$/i,
];
const TEST_PATH_PATTERNS = [
/(^|\/)__tests__\//i,
/\.test\.[tj]sx?$/i,
/\.spec\.[tj]sx?$/i,
/(^|\/)tests?\//i,
];
function matchesAny(path: string, patterns: RegExp[]): boolean {
for (const re of patterns) if (re.test(path)) return true;
return false;
}
function isSchemaMigrationPath(path: string): boolean {
return matchesAny(path, SCHEMA_MIGRATION_PATTERNS);
}
function isLockedPath(path: string): boolean {
return matchesAny(path, LOCKED_PATH_PATTERNS);
}
function isTestPath(path: string): boolean {
return matchesAny(path, TEST_PATH_PATTERNS);
}
/** Diff-file shape we accept — keeps signal gather DI-friendly for tests. */
export interface DiffFileInput {
path: string;
additions: number;
deletions: number;
}
/**
* Compute the raw signal map from a parsed diff plus owner rules and the
* pre/post dependency sets. Pure helper — no DB, no git, no LLM. Exported
* so it can be exercised directly by tests with synthetic inputs.
*/
export function buildSignalsFromDiff(args: {
files: DiffFileInput[];
raw: string;
ownerRules: OwnerRule[];
/** Existing dep set on the base side, keyed `ecosystem:name`. */
baseDeps: Map<string, string | null>;
/** Dep set after the PR's changes (best-effort), same key shape. */
headDeps: Map<string, string | null>;
}): PrRiskSignals {
let linesAdded = 0;
let linesRemoved = 0;
let schemaMigrationTouched = false;
let lockedPathTouched = false;
let testFiles = 0;
let testLines = 0;
const ownerSet = new Set<string>();
for (const f of args.files) {
linesAdded += Math.max(0, f.additions || 0);
linesRemoved += Math.max(0, f.deletions || 0);
if (isSchemaMigrationPath(f.path)) schemaMigrationTouched = true;
if (isLockedPath(f.path)) lockedPathTouched = true;
if (isTestPath(f.path)) {
testFiles += 1;
testLines += Math.max(0, (f.additions || 0) + (f.deletions || 0));
}
if (args.ownerRules.length > 0) {
for (const o of ownersForPath(f.path, args.ownerRules)) {
ownerSet.add(o);
}
}
}
const filesChanged = args.files.length;
const teamsAffected = ownerSet.size;
const testsAddedForNewCode = testFiles > 0 && linesAdded > 0;
const totalLines = linesAdded + linesRemoved;
const diffMinusTestRatio =
totalLines === 0 ? 0 : 1 - Math.min(1, testLines / totalLines);
// Dependency comparison — adds vs bumps.
let addsNewDependency = false;
let bumpsMajorDependency = false;
for (const [key, headVersion] of args.headDeps) {
if (!args.baseDeps.has(key)) {
addsNewDependency = true;
continue;
}
const baseVersion = args.baseDeps.get(key) ?? null;
if (isMajorBump(baseVersion, headVersion)) {
bumpsMajorDependency = true;
}
}
return {
filesChanged,
linesAdded,
linesRemoved,
teamsAffected,
schemaMigrationTouched,
lockedPathTouched,
addsNewDependency,
bumpsMajorDependency,
testsAddedForNewCode,
diffMinusTestRatio,
};
}
/**
* Heuristic semver-major detector. Strips leading ^ ~ = v and compares
* the leading integer chunks. Anything we cannot parse returns false so
* we err toward "no false positive bump".
*/
export function isMajorBump(
before: string | null | undefined,
after: string | null | undefined
): boolean {
const a = leadingMajor(before);
const b = leadingMajor(after);
if (a === null || b === null) return false;
return b > a;
}
function leadingMajor(spec: string | null | undefined): number | null {
if (!spec) return null;
const stripped = spec.trim().replace(/^[~^=v]+/, "");
const match = stripped.match(/^(\d+)/);
if (!match) return null;
const n = parseInt(match[1], 10);
return Number.isFinite(n) ? n : null;
}
// ---------------------------------------------------------------------------
// DB-touching orchestrator
// ---------------------------------------------------------------------------
interface PrRow {
id: string;
title: string;
baseBranch: string;
headBranch: string;
repositoryId: string;
}
interface RepoRow {
id: string;
name: string;
ownerUsername: string;
}
async function loadPrRow(prId: string): Promise<{
pr: PrRow;
repo: RepoRow;
} | null> {
try {
const [row] = await db
.select({
prId: pullRequests.id,
title: pullRequests.title,
baseBranch: pullRequests.baseBranch,
headBranch: pullRequests.headBranch,
repositoryId: pullRequests.repositoryId,
repoName: repositories.name,
ownerUsername: users.username,
})
.from(pullRequests)
.innerJoin(repositories, eq(repositories.id, pullRequests.repositoryId))
.innerJoin(users, eq(users.id, repositories.ownerId))
.where(eq(pullRequests.id, prId))
.limit(1);
if (!row) return null;
return {
pr: {
id: row.prId,
title: row.title,
baseBranch: row.baseBranch,
headBranch: row.headBranch,
repositoryId: row.repositoryId,
},
repo: {
id: row.repositoryId,
name: row.repoName,
ownerUsername: row.ownerUsername,
},
};
} catch {
return null;
}
}
async function loadOwnerRules(repoId: string): Promise<OwnerRule[]> {
try {
const rows = await db
.select({
pattern: codeOwners.pathPattern,
owners: codeOwners.ownerUsernames,
})
.from(codeOwners)
.where(eq(codeOwners.repositoryId, repoId));
return rows.map((r) => ({
pattern: r.pattern,
owners: (r.owners || "").split(",").filter(Boolean),
}));
} catch {
return [];
}
}
async function loadCurrentDeps(
repoId: string
): Promise<Map<string, string | null>> {
try {
const rows = await db
.select({
ecosystem: repoDependencies.ecosystem,
name: repoDependencies.name,
versionSpec: repoDependencies.versionSpec,
})
.from(repoDependencies)
.where(eq(repoDependencies.repositoryId, repoId));
const out = new Map<string, string | null>();
for (const r of rows) {
out.set(`${r.ecosystem}:${r.name}`, r.versionSpec);
}
return out;
} catch {
return new Map();
}
}
/**
* Parse the dependency state shown by the PR's head commit. We re-use
* the indexed base-side `repo_dependencies` rows as a baseline and look
* for additions/changes by parsing a head-side `package.json` blob via
* `git show`. If git access fails we return null which the caller treats
* as "no signal" (addsNewDependency = bumpsMajorDependency = false).
*/
async function loadHeadDeps(
ownerName: string,
repoName: string,
baseBranch: string,
headBranch: string
): Promise<Map<string, string | null>> {
const out = new Map<string, string | null>();
// We only sniff package.json on the head side (the most common manifest);
// the base-side counts come from the indexed snapshot already. This is a
// best-effort signal — false negatives are fine, false positives would be
// worse.
try {
const cwd = getRepoPath(ownerName, repoName);
const proc = Bun.spawn(
["git", "show", `${headBranch}:package.json`],
{ cwd, stdout: "pipe", stderr: "pipe" }
);
const text = await new Response(proc.stdout).text();
const exit = await proc.exited;
if (exit !== 0 || !text.trim()) return out;
const parsed = JSON.parse(text) as Record<string, unknown>;
for (const key of [
"dependencies",
"devDependencies",
"peerDependencies",
"optionalDependencies",
]) {
const block = parsed[key];
if (!block || typeof block !== "object") continue;
for (const [name, spec] of Object.entries(block as Record<string, unknown>)) {
out.set(
`npm:${name}`,
typeof spec === "string" ? spec : null
);
}
}
// Mark baseBranch silently used so the args object stays meaningful
// even if a future refactor adds a base-side diff.
void baseBranch;
} catch {
/* swallow — return whatever we have */
}
return out;
}
/**
* Compute + persist the risk score for one PR. Returns null when the PR
* cannot be found or the git head SHA cannot be resolved. Idempotent on
* the unique (pull_request_id, commit_sha) constraint.
*/
export async function computePrRiskForPullRequest(
pullRequestId: string,
opts: { now?: Date } = {}
): Promise<PrRiskScore | null> {
const loaded = await loadPrRow(pullRequestId);
if (!loaded) return null;
const { pr, repo } = loaded;
const headSha = await resolveRef(
repo.ownerUsername,
repo.name,
pr.headBranch
).catch(() => null);
if (!headSha) return null;
let diffFiles: DiffFileInput[] = [];
let diffRaw = "";
try {
const d = await getDiff(repo.ownerUsername, repo.name, headSha);
diffFiles = d.files.map((f) => ({
path: f.path,
additions: f.additions,
deletions: f.deletions,
}));
diffRaw = d.raw;
} catch {
/* swallow — buildSignalsFromDiff handles empty arrays */
}
const ownerRules = await loadOwnerRules(repo.id);
const baseDeps = await loadCurrentDeps(repo.id);
const headDeps = await loadHeadDeps(
repo.ownerUsername,
repo.name,
pr.baseBranch,
pr.headBranch
);
const signals = buildSignalsFromDiff({
files: diffFiles,
raw: diffRaw,
ownerRules,
baseDeps,
headDeps,
});
const { score, band } = computePrRiskScore(signals);
const aiSummary = await generatePrRiskSummary({
signals,
title: pr.title,
baseBranch: pr.baseBranch,
headBranch: pr.headBranch,
});
const generatedAt = opts.now ?? new Date();
// Best-effort upsert: try insert, fall back to existing row on the
// unique-constraint hit.
try {
await db.insert(prRiskScores).values({
pullRequestId,
commitSha: headSha,
score,
band,
signals,
aiSummary,
generatedAt,
});
} catch {
// Already cached for this SHA — that's fine, the caller's view
// converges on the same shape via getCachedPrRisk below. No throw.
}
return {
score,
band,
signals,
aiSummary,
generatedAt,
commitSha: headSha,
};
}
/**
* Return the cached risk score for the current head SHA of a PR, if any.
* Returns null when no row matches (cache miss) — the caller is expected
* to kick off `computePrRiskForPullRequest` async and show a placeholder.
*/
export async function getCachedPrRisk(
pullRequestId: string
): Promise<PrRiskScore | null> {
try {
const loaded = await loadPrRow(pullRequestId);
if (!loaded) return null;
const headSha = await resolveRef(
loaded.repo.ownerUsername,
loaded.repo.name,
loaded.pr.headBranch
).catch(() => null);
if (!headSha) return null;
const [row] = await db
.select()
.from(prRiskScores)
.where(
and(
eq(prRiskScores.pullRequestId, pullRequestId),
eq(prRiskScores.commitSha, headSha)
)
)
.orderBy(desc(prRiskScores.generatedAt))
.limit(1);
if (!row) return null;
return rowToPrRiskScore(row);
} catch {
return null;
}
}
/**
* Lookup variant that doesn't need git access — returns the most-recent
* cached score for a PR regardless of which SHA it was pinned to. Used
* by the MCP merge tool where we want the score even if the SHA has
* since moved on.
*/
export async function getLatestCachedPrRisk(
pullRequestId: string
): Promise<PrRiskScore | null> {
try {
const [row] = await db
.select()
.from(prRiskScores)
.where(eq(prRiskScores.pullRequestId, pullRequestId))
.orderBy(desc(prRiskScores.generatedAt))
.limit(1);
if (!row) return null;
return rowToPrRiskScore(row);
} catch {
return null;
}
}
function rowToPrRiskScore(row: PrRiskScoreRow): PrRiskScore {
return {
score: row.score,
band: row.band as PrRiskBand,
signals: (row.signals as unknown) as PrRiskSignals,
aiSummary: row.aiSummary ?? "",
generatedAt: row.generatedAt,
commitSha: row.commitSha,
};
}
// ---------------------------------------------------------------------------
// Test-only surface
// ---------------------------------------------------------------------------
export const __test = {
bandFor,
clamp01,
deterministicSummary,
isMajorBump,
leadingMajor,
isSchemaMigrationPath,
isLockedPath,
isTestPath,
rowToPrRiskScore,
};
|