Pre-launch — Gluecron is in final validation. Public signups and git hosting for non-owner users open after launch review.
CodeIssuesDiscussionsWikiPull RequestsProjectsCommitsActionsReleasesContributorsPulse● GatesSecuritySettingsDeploymentsPipelineInsightsAgents✨ Explain✨ Ask AI✨ Workspace✨ Spec✨ Tests▓ Debt Map✨ NL Search🏛 Archaeology
claude/adoring-hopper-5x74bqclaude/affectionate-feynman-ykrf1hclaude/architecture-audit-design-wxprenclaude/build-status-update-3MXsfclaude/charming-meitner-mllb5rclaude/compare-gate-gluecron-s4mFQclaude/confident-faraday-tikcwbclaude/continue-work-XMTlIclaude/crontech-gluecron-deploy-7MIECclaude/crontech-platform-setup-SeKfwclaude/design-2026claude/ecstatic-ptolemy-jMdigclaude/enhance-github-integration-QNHdGclaude/fix-aa-loop-issue-PonMQclaude/fix-actions-and-processclaude/fix-desktop-errors-XqoW8claude/fix-red-workflowsclaude/fix-website-access-6FKJNclaude/gatetest-integration-hardeningclaude/github-audit-improvements-bDFr9claude/gluecron-launch-status-FoMRlclaude/hopeful-lamport-olfCTclaude/issue-to-pr-and-protectionsclaude/jolly-heisenberg-2sg1Qclaude/launch-preparation-QmTb6claude/new-session-xk1l7claude/plan-platform-architecture-kkN4yclaude/platform-analysis-roadmap-1nUGLclaude/platform-launch-assessment-8dWV8claude/polish-platform-release-AeDrUclaude/resume-previous-work-KzyLwclaude/review-crontech-handoff-qYEVqclaude/review-project-completeness-lHhS2claude/review-readme-docs-ulqPKclaude/serene-edison-rj87weclaude/setup-multi-repo-dev-BCwNQclaude/ship-fixes-and-tests-Jvz1cclaude/site-audit-competitive-pctlwgclaude/site-migration-vercel-XstpKclaude/standalone-product-repos-XHFTDcopilot/feat-smart-empty-states-keyboard-first-enhancementcopilot/feat-smart-morning-digest-review-context-restorecopilot/fix-and-process-workflowscopilot/update-ai-powered-code-reviewfeat/debt-mapfeat/push-policy-codeowners-hardeningfeat/smart-digest-contextfeat/stage-impactfeat/t1-secret-migrationfeat/u-polishfeat/w-self-hostfeat/w2-claude-configfix/agent-journey-orphan-sweepgatetest/auto-fix-1776586424172gatetest/auto-fix-1776586534814gatetest/auto-fix-1776590685143gatetest/auto-fix-1776590808199mainops/redeploy-retriggerstyle/dxt-cta-themeworktree-agent-a3377aad30d55da26worktree-agent-a7ef607b7ee1d6c74
pr-splitter.ts6.1 KB · 214 lines
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
/**
 * PR Split Suggestions — AI-powered guidance for decomposing large PRs.
 *
 * When a PR has >400 lines changed, Claude Sonnet is asked to suggest how
 * to split it into 2-4 smaller, independently-mergeable PRs grouped by
 * logical concern (schema / API / UI / tests etc.).
 *
 * Results are cached in memory for 1 hour per PR — the AI call is expensive
 * and the diff doesn't change between page loads.
 *
 * Returns `null` when:
 *   - PR has <=400 changed lines
 *   - AI is unavailable (no ANTHROPIC_API_KEY)
 *   - Claude returns fewer than 2 suggestions
 *   - Any error occurs (always degrades gracefully)
 */

import { getAnthropic, MODEL_SONNET, extractText, parseJsonResponse, isAiAvailable } from "./ai-client";
import { join } from "path";
import { config } from "./config";

// ---------------------------------------------------------------------------
// Public types
// ---------------------------------------------------------------------------

export interface SplitPr {
  title: string;
  rationale: string;
  files: string[];
  estimatedLines: number;
  suggestedBranch: string;
}

export interface SplitSuggestion {
  originalPrTitle: string;
  totalFiles: number;
  totalLines: number;
  suggestedPrs: SplitPr[];
  mergeOrder: string[];
}

// ---------------------------------------------------------------------------
// In-memory cache (1h TTL per prId)
// ---------------------------------------------------------------------------

interface CacheEntry {
  suggestion: SplitSuggestion | null;
  cachedAt: number;
}

const _cache = new Map<string, CacheEntry>();
const CACHE_TTL_MS = 60 * 60 * 1000; // 1 hour

function getCached(prId: string): SplitSuggestion | null | undefined {
  const entry = _cache.get(prId);
  if (!entry) return undefined; // cache miss
  if (Date.now() - entry.cachedAt > CACHE_TTL_MS) {
    _cache.delete(prId);
    return undefined;
  }
  return entry.suggestion;
}

function setCached(prId: string, suggestion: SplitSuggestion | null): void {
  _cache.set(prId, { suggestion, cachedAt: Date.now() });
}

// ---------------------------------------------------------------------------
// Diff stat parsing
// ---------------------------------------------------------------------------

interface FileStatLine {
  path: string;
  added: number;
  deleted: number;
  total: number;
}

function parseNumstat(raw: string): FileStatLine[] {
  return raw
    .trim()
    .split("\n")
    .filter(Boolean)
    .map((line) => {
      const parts = line.split("\t");
      if (parts.length < 3) return null;
      const added = parts[0] === "-" ? 0 : parseInt(parts[0], 10) || 0;
      const deleted = parts[1] === "-" ? 0 : parseInt(parts[1], 10) || 0;
      return { path: parts[2], added, deleted, total: added + deleted };
    })
    .filter((x): x is FileStatLine => x !== null);
}

function getRepoDir(owner: string, repo: string): string {
  return join(config.gitReposPath, `${owner}/${repo}.git`);
}

async function getDiffStat(
  owner: string,
  repo: string,
  baseBranch: string,
  headBranch: string
): Promise<FileStatLine[]> {
  const repoDir = getRepoDir(owner, repo);
  const proc = Bun.spawn(
    ["git", "--git-dir", repoDir, "diff", "--numstat", `${baseBranch}...${headBranch}`],
    { stdout: "pipe", stderr: "pipe" }
  );
  const raw = await new Response(proc.stdout).text();
  await proc.exited;
  return parseNumstat(raw);
}

// ---------------------------------------------------------------------------
// Public API
// ---------------------------------------------------------------------------

/**
 * Suggest how to split a large PR.
 * Returns null when the PR is small, AI is unavailable, or any error occurs.
 */
export async function suggestPrSplit(
  prId: string,
  prTitle: string,
  ownerName: string,
  repoName: string,
  baseBranch: string,
  headBranch: string
): Promise<SplitSuggestion | null> {
  // Check cache first
  const cached = getCached(prId);
  if (cached !== undefined) return cached;

  try {
    const fileStats = await getDiffStat(ownerName, repoName, baseBranch, headBranch);
    const totalLines = fileStats.reduce((s, f) => s + f.total, 0);
    const totalFiles = fileStats.length;

    if (totalLines < 400) {
      setCached(prId, null);
      return null;
    }

    if (!isAiAvailable()) {
      setCached(prId, null);
      return null;
    }

    const fileList = fileStats
      .map((f) => `${f.path}  +${f.added} -${f.deleted}`)
      .join("\n");

    const prompt = `This PR is too large to review effectively (${totalLines} lines across ${totalFiles} files).
Suggest how to split it into 2-4 smaller PRs that can be reviewed and merged independently.

PR title: ${prTitle}
Files changed:
${fileList}

Return JSON with this exact shape (no extra keys, no prose outside the JSON block):
{
  "suggestedPrs": [
    {
      "title": "...",
      "rationale": "...",
      "files": ["..."],
      "estimatedLines": N,
      "suggestedBranch": "..."
    }
  ],
  "mergeOrder": ["PR title 1", "PR title 2"]
}

Rules:
- Group by logical concern (schema changes together, API layer together, UI together)
- Each suggested PR should be independently mergeable
- Suggest merge order to minimise conflicts
- suggestedBranch should be kebab-case derived from the PR title (e.g. feat/auth-schema-only)
- Return between 2 and 4 suggested PRs`;

    const anthropic = getAnthropic();
    const message = await anthropic.messages.create({
      model: MODEL_SONNET,
      max_tokens: 1024,
      messages: [{ role: "user", content: prompt }],
    });

    const text = extractText(message);
    const parsed = parseJsonResponse<{
      suggestedPrs: SplitPr[];
      mergeOrder: string[];
    }>(text);

    if (!parsed || !Array.isArray(parsed.suggestedPrs) || parsed.suggestedPrs.length < 2) {
      setCached(prId, null);
      return null;
    }

    const suggestion: SplitSuggestion = {
      originalPrTitle: prTitle,
      totalFiles,
      totalLines,
      suggestedPrs: parsed.suggestedPrs,
      mergeOrder: Array.isArray(parsed.mergeOrder) ? parsed.mergeOrder : [],
    };

    setCached(prId, suggestion);
    return suggestion;
  } catch {
    // Always degrade gracefully
    setCached(prId, null);
    return null;
  }
}