@iceinvein/agent-skills 0.1.40 → 0.2.0

package/skills/magpie/scripts/__tests__/shard.test.ts

@@ -0,0 +1,165 @@
+import { afterEach, beforeEach, expect, test } from 'bun:test'
+import { existsSync } from 'node:fs'
+import { mkdir, mkdtemp, readdir, readFile, rm, writeFile } from 'node:fs/promises'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+import { DEFAULT_SHARD_BUDGET, groupKey, planShards, shardDiff } from '../shard.ts'
+
+let runDir: string
+
+beforeEach(async () => {
+  runDir = await mkdtemp(join(tmpdir(), 'magpie-shard-'))
+})
+
+afterEach(async () => {
+  await rm(runDir, { recursive: true, force: true })
+})
+
+/** A file chunk with `bodyLines` added lines, shaped like a real patch. */
+function chunk(path: string, bodyLines: number): string {
+  const body = Array.from({ length: bodyLines }, (_, i) => `+line ${i}`).join('\n')
+  return [
+    `diff --git a/${path} b/${path}`,
+    'index 0000000..1111111 100644',
+    `--- a/${path}`,
+    `+++ b/${path}`,
+    `@@ -0,0 +1,${bodyLines} @@`,
+    body,
+    '',
+  ].join('\n')
+}
+
+test('groupKey caps the grouping directory at two segments', () => {
+  expect(groupKey('src/cli/commands/install.ts')).toBe('src/cli')
+  expect(groupKey('src/a.ts')).toBe('src')
+  expect(groupKey('README.md')).toBe('.')
+})
+
+test('planShards keeps a directory group together when it fits', () => {
+  const chunks = [
+    { path: 'src/cli/a.ts', text: '', lines: 100 },
+    { path: 'src/web/b.ts', text: '', lines: 100 },
+    { path: 'src/cli/c.ts', text: '', lines: 100 },
+  ]
+  const shards = planShards(chunks, 1000, 80)
+  expect(shards).toHaveLength(1)
+  // Grouping reorders: both src/cli files are adjacent.
+  expect(shards[0]?.map((c) => c.path)).toEqual(['src/cli/a.ts', 'src/cli/c.ts', 'src/web/b.ts'])
+})
+
+test('planShards splits when the line budget is exceeded', () => {
+  const chunks = [
+    { path: 'src/a/1.ts', text: '', lines: 600 },
+    { path: 'src/b/2.ts', text: '', lines: 600 },
+    { path: 'src/c/3.ts', text: '', lines: 600 },
+  ]
+  const shards = planShards(chunks, 1000, 80)
+  expect(shards).toHaveLength(3)
+})
+
+test('planShards splits when max-files is reached before the budget', () => {
+  const chunks = Array.from({ length: 5 }, (_, i) => ({
+    path: `src/g${i}/f.ts`,
+    text: '',
+    lines: 1,
+  }))
+  const shards = planShards(chunks, 10_000, 2)
+  expect(shards).toHaveLength(3)
+  expect(shards[0]).toHaveLength(2)
+  expect(shards[2]).toHaveLength(1)
+})
+
+test('planShards gives an over-budget file a shard to itself', () => {
+  const chunks = [
+    { path: 'src/a/small.ts', text: '', lines: 10 },
+    { path: 'src/b/huge.ts', text: '', lines: 9000 },
+    { path: 'src/c/other.ts', text: '', lines: 10 },
+  ]
+  const shards = planShards(chunks, 1000, 80)
+  expect(shards).toHaveLength(3)
+  expect(shards[1]?.map((c) => c.path)).toEqual(['src/b/huge.ts'])
+})
+
+test('shardDiff writes a single-shard manifest pointing at diff.patch', async () => {
+  await writeFile(join(runDir, 'diff.patch'), chunk('src/a.ts', 3))
+  const manifest = await shardDiff(runDir)
+  expect(manifest.shards).toHaveLength(1)
+  expect(manifest.shards[0]?.path).toBe('diff.patch')
+  expect(manifest.totalFiles).toBe(1)
+  expect(manifest.budget).toBe(DEFAULT_SHARD_BUDGET)
+  expect(existsSync(join(runDir, 'shards', 'shard-1.patch'))).toBe(false)
+  const onDisk = JSON.parse(await readFile(join(runDir, 'shards', 'manifest.json'), 'utf8'))
+  expect(onDisk.shards[0].files).toEqual(['src/a.ts'])
+})
+
+test('shardDiff writes one patch per shard when the budget forces a split', async () => {
+  const diff = chunk('src/a/one.ts', 200) + chunk('src/b/two.ts', 200)
+  await writeFile(join(runDir, 'diff.patch'), diff)
+  const manifest = await shardDiff(runDir, { budget: 150 })
+  expect(manifest.shards).toHaveLength(2)
+  expect(manifest.shards[0]?.path).toBe('shards/shard-1.patch')
+  const first = await readFile(join(runDir, 'shards', 'shard-1.patch'), 'utf8')
+  expect(first).toContain('diff --git a/src/a/one.ts')
+  expect(first).not.toContain('src/b/two.ts')
+  // diff.patch is a view source, never replaced.
+  const whole = await readFile(join(runDir, 'diff.patch'), 'utf8')
+  expect(whole).toContain('src/a/one.ts')
+  expect(whole).toContain('src/b/two.ts')
+})
+
+test('a re-split into fewer shards leaves no orphaned shard patches behind', async () => {
+  const diff =
+    chunk('src/a/one.ts', 200) + chunk('src/b/two.ts', 200) + chunk('src/c/three.ts', 200)
+  await writeFile(join(runDir, 'diff.patch'), diff)
+  const first = await shardDiff(runDir, { budget: 150 })
+  expect(first.shards).toHaveLength(3)
+  expect(existsSync(join(runDir, 'shards', 'shard-3.patch'))).toBe(true)
+
+  // Re-split with room for two shards. shard-3.patch describes a file set no
+  // shard id owns any more, so it must not survive the new split.
+  const second = await shardDiff(runDir, { budget: 450 })
+  expect(second.shards).toHaveLength(2)
+  const patches = (await readdir(join(runDir, 'shards'))).filter((n) => n.endsWith('.patch')).sort()
+  expect(patches).toEqual(['shard-1.patch', 'shard-2.patch'])
+  expect(existsSync(join(runDir, 'shards', 'manifest.json'))).toBe(true)
+})
+
+test('a re-split down to one shard clears every shard patch and keeps the manifest', async () => {
+  const diff = chunk('src/a/one.ts', 200) + chunk('src/b/two.ts', 200)
+  await writeFile(join(runDir, 'diff.patch'), diff)
+  expect((await shardDiff(runDir, { budget: 150 })).shards).toHaveLength(2)
+  const manifest = await shardDiff(runDir, { budget: 10_000 })
+  expect(manifest.shards).toHaveLength(1)
+  expect(manifest.shards[0]?.path).toBe('diff.patch')
+  // Only the manifest is left: the single-shard entry points at diff.patch, and
+  // diff.patch itself lives outside shards/ and is never touched.
+  expect(await readdir(join(runDir, 'shards'))).toEqual(['manifest.json'])
+  expect(existsSync(join(runDir, 'diff.patch'))).toBe(true)
+})
+
+test('shardDiff handles an empty diff without writing shard files', async () => {
+  await writeFile(join(runDir, 'diff.patch'), '')
+  const manifest = await shardDiff(runDir)
+  expect(manifest.shards).toEqual([])
+  expect(manifest.totalFiles).toBe(0)
+})
+
+test('shardDiff propagates a non-ENOENT read failure instead of treating it as an empty diff', async () => {
+  // Make diff.patch a directory: reading it as text is a real, portable I/O
+  // failure (EISDIR) that is not "file absent" and must not be swallowed.
+  await mkdir(join(runDir, 'diff.patch'))
+  await expect(shardDiff(runDir)).rejects.toThrow()
+})
+
+test('shardDiff treats a diff.patch that does not exist yet as an empty diff', async () => {
+  // No writeFile at all here: diff.patch is genuinely absent, the one case
+  // the ENOENT branch exists to tolerate. Task 4 wires shardDiff into
+  // runSetup and Task 6 reads the manifest unconditionally, so "no diff yet"
+  // must still leave a readable manifest rather than a missing file.
+  expect(existsSync(join(runDir, 'diff.patch'))).toBe(false)
+  const manifest = await shardDiff(runDir)
+  expect(manifest.shards).toEqual([])
+  expect(manifest.totalFiles).toBe(0)
+  expect(manifest.totalLines).toBe(0)
+  expect(existsSync(join(runDir, 'shards', 'manifest.json'))).toBe(true)
+})

package/skills/magpie/scripts/__tests__/skill-lint.test.ts

@@ -117,7 +117,9 @@ test('SKILL.md no longer inlines the prompt bodies it moved out', async () => {
     expect(text).not.toContain(`\`\`\`${tag}`)
   }
   // The walkthrough is the always-read part; keep it small enough to be cheap.
-  expect(text.split(/\s+/).length).toBeLessThan(2600)
+  // Raised from 2600 when the sharded-dispatch and fallback-diff prose was added:
+  // that's real, load-bearing procedure, not bloat.
+  expect(text.split(/\s+/).length).toBeLessThan(3000)
 })
 
 test('SKILL.md never instructs the agent to approve indexing', async () => {
@@ -327,3 +329,96 @@ test('the output contract tells specialists to look before they hedge', async ()
   expect(contract).toContain('Needs verification:')
   expect(contract).toMatch(/look before .*hedg/i)
 })
+
+test('the output contract documents both findings filenames', async () => {
+  const text = await readFile(ref('specialists.md'), 'utf8')
+  const contract = text.slice(0, text.indexOf('```magpie-specialist-'))
+  // The pre-existing assertion must keep holding.
+  expect(contract).toMatch(/findings\/<focus>\.json/)
+  expect(contract).toMatch(/findings\/<focus>\.shard-<n>\.json/)
+})
+
+test('the run header documents the shard lines', async () => {
+  const text = await readFile(ref('specialists.md'), 'utf8')
+  expect(text).toContain('Shard: <n> of <N>')
+  expect(text).toContain('shards/shard-<n>.patch')
+})
+
+test('specialists are told excluded files are still reachable', async () => {
+  const text = await readFile(ref('specialists.md'), 'utf8')
+  expect(text).toContain('diff.full.patch')
+  expect(text).toContain('excluded-files.json')
+})
+
+test('stage 4 pins the shard field on the specialist log entry', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  const start = text.indexOf('### 4. Specialists')
+  expect(start).toBeGreaterThan(-1)
+  const section = text.slice(start, text.indexOf('\n### 5.', start))
+  // render-cmd sums per-focus findings counts keyed on this field. Drop it from a
+  // sharded run and every shard of a focus lands in one 'all' bucket,
+  // last-write-wins, so the progress page under-reports by up to the shard count
+  // with no error anywhere. This prose is the only thing that produces the field.
+  expect(section).toMatch(/\{stage: specialist,[^}]*\bshard: <n>/)
+  // ... and the unsharded path must still omit it, or 'all' would never be used.
+  expect(section).toMatch(/omit `shard`/i)
+})
+
+test('stage 4 reconciles the expected findings files before dedupe', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  const start = text.indexOf('### 4. Specialists')
+  const section = text.slice(start, text.indexOf('\n### 5.', start))
+  // Stage 4 fails only when every specialist fails, so losing 1 of 30 agents
+  // otherwise renders a report indistinguishable from a complete one.
+  expect(section).toMatch(/findings\/<focus>\.shard-<n>\.json/)
+  expect(section).toMatch(/5 × <shard count>/)
+})
+
+test('the stage-4 gate offers only options the pipeline can carry out', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  const start = text.indexOf('**More than four shards')
+  expect(start).toBeGreaterThan(-1)
+  const gate = text.slice(start, text.indexOf('\n\n', start))
+  // Option 2: --max-files is what binds on a PR of many small files, so naming
+  // only --budget sends the user back to an unchanged shard count.
+  expect(gate).toContain('--budget')
+  expect(gate).toContain('--max-files')
+  // Option 3 must name a mechanism that exists. There is no unreviewed marker in
+  // the report, so the record is a diagnostic log entry, on a stage outside
+  // status-cmd's ORDER ladder and never `status: error`.
+  const { ORDER } = await import('../status-cmd.ts')
+  const stage = gate.match(/\{stage: ([a-z-]+), status: ([a-z-]+)/)
+  expect(stage?.[1]).toBeDefined()
+  expect(ORDER as readonly string[]).not.toContain(stage?.[1])
+  expect(stage?.[2]).not.toBe('error')
+  expect(gate).toMatch(/skipped/)
+})
+
+test('the excluded-files note is part of the specialist prompt, not commentary', async () => {
+  const spec = await readFile(ref('specialists.md'), 'utf8')
+  // Every other specialist-directed line lives in a fenced block; this one used to
+  // sit unfenced between two numbered assembly parts, where an assembling agent
+  // reads it as a note to itself and the specialist never sees it.
+  const fences = [...spec.matchAll(/^```[a-z-]*\n([\s\S]*?)^```/gm)].map((m) => m[1] ?? '')
+  const inFence = fences.some((body) => body.includes('diff.full.patch'))
+  expect(inFence).toBe(true)
+  // And it must be a numbered part of the assembly list, not floating text.
+  expect(spec).toMatch(/^\d+\. The excluded-files block below/m)
+})
+
+test('the specialist assembly list is numbered contiguously from 1', async () => {
+  const spec = await readFile(ref('specialists.md'), 'utf8')
+  // Anchor on the heading, not on item 3's inline `## Output Contract` reference.
+  const preamble = spec.slice(0, spec.indexOf('\n## Output Contract'))
+  const numbers = [...preamble.matchAll(/^(\d+)\. /gm)].map((m) => Number(m[1]))
+  expect(numbers.length).toBeGreaterThan(4)
+  expect(numbers).toEqual(numbers.map((_, i) => i + 1))
+})
+
+test('SKILL.md documents the shard manifest and the fan-out gate', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  expect(text).toContain('shards/manifest.json')
+  expect(text).toContain('diff.patch')
+  // The confirmation gate above four shards is the design's only interactive stop.
+  expect(text).toMatch(/more than four shards/i)
+})

package/skills/magpie/scripts/dedupe-cmd.ts

@@ -3,6 +3,7 @@ import { join } from 'node:path'
 import { annotateChangedLines } from './changed-lines.ts'
 import { deduplicateFindings } from './dedupe.ts'
 import { verifyEvidence } from './evidence-filter.ts'
+import { namespaceId, parseFindingsFilename } from './findings-files.ts'
 import { DEFAULT_THRESHOLD, scoreRisk } from './score.ts'
 import { FOCUS_IDS, parseFinding, type ReviewFinding } from './types.ts'
 
@@ -16,6 +17,49 @@ export type RunDedupeOptions = {
   threshold?: number
 }
 
+/**
+ * The five focuses a specialist subagent writes. `tests` is a `FOCUS_IDS` entry
+ * too, but setup writes `findings/tests.json` once for the whole run with no
+ * subagent, so it is never expected per shard.
+ */
+const SPECIALIST_FOCUS_IDS = FOCUS_IDS.filter((id) => id !== 'tests')
+
+type Coverage = { expected: number; missing: string[] }
+
+/**
+ * Reconcile the findings files on disk against the `(focus, shard)` pairs the
+ * shard manifest implies. Stage 4 fails the run only when *every* specialist
+ * fails, so a sharded run that lost one agent out of thirty still logs
+ * `specialists: done` and renders a report indistinguishable from a complete
+ * one. Returns null when there is nothing to reconcile against (no manifest, an
+ * unreadable one, or a manifest with no shards, which is what an empty or
+ * fully-filtered diff produces), leaving pre-sharder runs behaving exactly as
+ * they did before.
+ */
+async function reconcileCoverage(runDir: string, present: Set<string>): Promise<Coverage | null> {
+  let shards: Array<Record<string, unknown>>
+  try {
+    const manifest = (await Bun.file(join(runDir, 'shards', 'manifest.json')).json()) as {
+      shards?: unknown
+    }
+    if (!Array.isArray(manifest.shards) || manifest.shards.length === 0) return null
+    shards = manifest.shards as Array<Record<string, unknown>>
+  } catch {
+    return null
+  }
+  // A single-shard manifest means stage 4 took the unsharded path, so its
+  // specialists write `<focus>.json` with no shard suffix.
+  const sharded = shards.length > 1
+  const expected: string[] = []
+  for (const [i, shard] of shards.entries()) {
+    const id = typeof shard?.id === 'number' ? shard.id : i + 1
+    for (const focus of SPECIALIST_FOCUS_IDS) {
+      expected.push(sharded ? `${focus}.shard-${id}.json` : `${focus}.json`)
+    }
+  }
+  return { expected: expected.length, missing: expected.filter((name) => !present.has(name)) }
+}
+
 export async function runDedupe(runDir: string, options: RunDedupeOptions = {}): Promise<number> {
   const threshold = options.threshold ?? DEFAULT_THRESHOLD
   const findingsDir = join(runDir, 'findings')
@@ -29,8 +73,8 @@ export async function runDedupe(runDir: string, options: RunDedupeOptions = {}):
 
   for (const name of files) {
     if (!name.endsWith('.json')) continue
-    const focus = name.slice(0, -'.json'.length)
-    if (!FOCUS_IDS.includes(focus as (typeof FOCUS_IDS)[number])) {
+    const parsed = parseFindingsFilename(name)
+    if (!parsed || !FOCUS_IDS.includes(parsed.focus as (typeof FOCUS_IDS)[number])) {
       await logLine(runDir, {
         stage: 'dedupe',
         status: 'skip',
@@ -39,6 +83,7 @@ export async function runDedupe(runDir: string, options: RunDedupeOptions = {}):
       })
       continue
     }
+    const { focus, shard } = parsed
     const path = join(findingsDir, name)
     let raw: unknown
     try {
@@ -63,7 +108,8 @@ export async function runDedupe(runDir: string, options: RunDedupeOptions = {}):
     }
     for (const item of raw) {
       try {
-        collected.push(parseFinding(item))
+        const finding = parseFinding(item)
+        collected.push({ ...finding, id: namespaceId(finding.id, focus, shard) })
       } catch (err) {
         await logLine(runDir, {
           stage: 'dedupe',
@@ -107,6 +153,14 @@ export async function runDedupe(runDir: string, options: RunDedupeOptions = {}):
       )}\n`,
     )
   }
+  const coverage = await reconcileCoverage(runDir, new Set(files))
+  if (coverage) {
+    process.stdout.write(
+      coverage.missing.length === 0
+        ? `dedupe: all ${coverage.expected} expected findings files present\n`
+        : `dedupe: ${coverage.missing.length} of ${coverage.expected} expected findings files missing (re-dispatch those specialists): ${coverage.missing.join(', ')}\n`,
+    )
+  }
   await logLine(runDir, {
     stage: 'dedupe',
     status: 'done',
@@ -118,6 +172,7 @@ export async function runDedupe(runDir: string, options: RunDedupeOptions = {}):
       skipped: evidence.skipped,
       dropped: evidence.dropped.length,
     },
+    ...(coverage ? { coverage } : {}),
   })
   return 0
 }

package/skills/magpie/scripts/diff-chunks.ts

@@ -0,0 +1,28 @@
+export type FileChunk = { path: string; text: string; lines: number }
+
+export const FILE_HEADER = /^diff --git a\/(.+?) b\/(.+?)$/m
+
+/** Split a unified diff on file boundaries. The first element may be a preamble
+ *  that is not a file chunk; callers decide what to do with it. */
+export function splitRawChunks(diff: string): string[] {
+  return diff.split(/^(?=diff --git )/m)
+}
+
+/** The b-side path of a file chunk, falling back to the a-side for deletions. */
+export function chunkPath(chunk: string): string | null {
+  const m = chunk.match(FILE_HEADER)
+  return m?.[2] ?? m?.[1] ?? null
+}
+
+/** File chunks only, with their paths and patch-line counts resolved. */
+export function splitFileChunks(diff: string): FileChunk[] {
+  if (!diff.trim()) return []
+  const out: FileChunk[] = []
+  for (const chunk of splitRawChunks(diff)) {
+    if (!chunk.startsWith('diff --git ')) continue
+    const path = chunkPath(chunk)
+    if (!path) continue
+    out.push({ path, text: chunk, lines: chunk.split('\n').length })
+  }
+  return out
+}

package/skills/magpie/scripts/findings-files.ts

@@ -0,0 +1,32 @@
+/**
+ * A specialist writes `findings/<focus>.json`, or `findings/<focus>.shard-<n>.json`
+ * when stage 4 fanned the focus out across shards. Returns null for anything
+ * else, which the caller logs and skips.
+ */
+export function parseFindingsFilename(
+  name: string,
+): { focus: string; shard: number | null } | null {
+  if (!name.endsWith('.json')) return null
+  const stem = name.slice(0, -'.json'.length)
+  const parts = stem.split('.')
+  const focus = parts[0]
+  if (!focus) return null
+  if (parts.length === 1) return { focus, shard: null }
+  if (parts.length > 2) return null
+  const m = /^shard-(\d+)$/.exec(parts[1] ?? '')
+  if (!m) return null
+  return { focus, shard: Number(m[1]) }
+}
+
+/**
+ * Every shard's security specialist mints `security-1` for its first finding,
+ * so a five-shard run yields five findings sharing one id, and the report and
+ * post stages key on id. Rewriting deterministically here beats asking five
+ * prompts to encode the shard themselves.
+ */
+export function namespaceId(id: string, focus: string, shard: number | null): string {
+  if (shard === null) return id
+  const tag = `s${shard}`
+  const prefix = `${focus}-`
+  return id.startsWith(prefix) ? `${prefix}${tag}-${id.slice(prefix.length)}` : `${tag}-${id}`
+}

package/skills/magpie/scripts/gh.ts

@@ -1,16 +1,21 @@
 import { writeFile } from 'node:fs/promises'
 import { join } from 'node:path'
+import { localPrDiff } from './git-diff.ts'
 
 export type FetchPrInput = {
   ghBin: string
   prNumber: number
   runDir: string
-  /** Working directory for the gh invocation. Defaults to process.cwd(). */
+  /** Working directory for the gh and git invocations. Defaults to process.cwd(). */
   cwd?: string
+  /** git binary for the local-diff fallback. Defaults to `git` on PATH. */
+  gitBin?: string
+  /** Extra environment for the gh invocation. Test seam for the fake gh fixtures. */
+  env?: Record<string, string>
 }
 
 export type FetchPrResult =
-  | { ok: true; prJsonPath: string; diffPath: string }
+  | { ok: true; prJsonPath: string; diffPath: string; source: 'gh' | 'git'; mergeBase?: string }
   | { ok: false; error: string }
 
 export const PR_VIEW_FIELDS = [
@@ -31,9 +36,23 @@ export const PR_VIEW_FIELDS = [
   'closingIssuesReferences',
 ].join(',')
 
+function prFileCount(viewStdout: string): number {
+  try {
+    const parsed = JSON.parse(viewStdout) as { files?: unknown }
+    return Array.isArray(parsed.files) ? parsed.files.length : 0
+  } catch {
+    return 0
+  }
+}
+
 export async function fetchPr(input: FetchPrInput): Promise<FetchPrResult> {
   const { ghBin, prNumber, runDir, cwd } = input
-  const spawnOpts = { stdout: 'pipe' as const, stderr: 'pipe' as const, ...(cwd ? { cwd } : {}) }
+  const spawnOpts = {
+    stdout: 'pipe' as const,
+    stderr: 'pipe' as const,
+    ...(cwd ? { cwd } : {}),
+    ...(input.env ? { env: { ...process.env, ...input.env } } : {}),
+  }
   const view = Bun.spawn(
     [ghBin, 'pr', 'view', String(prNumber), '--json', PR_VIEW_FIELDS],
     spawnOpts,
@@ -48,20 +67,52 @@ export async function fetchPr(input: FetchPrInput): Promise<FetchPrResult> {
     }
   }
 
+  const prJsonPath = join(runDir, 'pr.json')
+  const diffPath = join(runDir, 'diff.patch')
+  await writeFile(prJsonPath, viewStdout)
+
   const diff = Bun.spawn([ghBin, 'pr', 'diff', String(prNumber)], spawnOpts)
   const diffStdout = await new Response(diff.stdout).text()
   const diffStderr = await new Response(diff.stderr).text()
   const diffExit = await diff.exited
-  if (diffExit !== 0) {
-    return {
-      ok: false,
-      error: `gh pr diff exit ${diffExit}: ${diffStderr.trim()}`,
-    }
+
+  // GitHub refuses the `.diff` media type above roughly 300 files (HTTP 406),
+  // and a truncating proxy can answer 200 with nothing at all. Reviewing an
+  // empty diff as though it were the PR is the worse failure of the two, so
+  // both conditions fall through to the local clone.
+  const emptyForNonEmptyPr = diffStdout.trim() === '' && prFileCount(viewStdout) > 0
+  if (diffExit === 0 && !emptyForNonEmptyPr) {
+    await writeFile(diffPath, diffStdout)
+    return { ok: true, prJsonPath, diffPath, source: 'gh' }
   }
 
-  const prJsonPath = join(runDir, 'pr.json')
-  const diffPath = join(runDir, 'diff.patch')
-  await writeFile(prJsonPath, viewStdout)
-  await writeFile(diffPath, diffStdout)
-  return { ok: true, prJsonPath, diffPath }
+  const ghError =
+    diffExit !== 0
+      ? `gh pr diff exit ${diffExit}: ${diffStderr.trim()}`
+      : 'gh pr diff returned an empty diff for a PR with changed files'
+
+  let pr: { baseRefName?: unknown; headRefOid?: unknown }
+  try {
+    pr = JSON.parse(viewStdout) as { baseRefName?: unknown; headRefOid?: unknown }
+  } catch (err) {
+    return { ok: false, error: `${ghError}; pr.json is unparseable: ${String(err)}` }
+  }
+  const baseRefName = typeof pr.baseRefName === 'string' ? pr.baseRefName : ''
+  const headRefOid = typeof pr.headRefOid === 'string' ? pr.headRefOid : ''
+  if (!baseRefName || !headRefOid) {
+    return { ok: false, error: `${ghError}; pr.json has no baseRefName/headRefOid to diff from` }
+  }
+
+  const local = await localPrDiff({
+    gitBin: input.gitBin ?? 'git',
+    repoPath: cwd ?? process.cwd(),
+    prNumber,
+    baseRefName,
+    headRefOid,
+  })
+  if (!local.ok) {
+    return { ok: false, error: `${ghError}; local fallback failed: ${local.error}` }
+  }
+  await writeFile(diffPath, local.diff)
+  return { ok: true, prJsonPath, diffPath, source: 'git', mergeBase: local.mergeBase }
 }

package/skills/magpie/scripts/git-diff.ts

@@ -0,0 +1,111 @@
+export type LocalDiffInput = {
+  gitBin: string
+  repoPath: string
+  prNumber: number
+  baseRefName: string
+  headRefOid: string
+}
+
+export type LocalDiffResult =
+  | { ok: true; diff: string; mergeBase: string }
+  | { ok: false; error: string }
+
+async function git(
+  gitBin: string,
+  cwd: string,
+  args: string[],
+): Promise<{ exit: number; stdout: string; stderr: string }> {
+  const proc = Bun.spawn([gitBin, ...args], { cwd, stdout: 'pipe', stderr: 'pipe' })
+  const stdout = await new Response(proc.stdout).text()
+  const stderr = await new Response(proc.stderr).text()
+  const exit = await proc.exited
+  return { exit, stdout, stderr }
+}
+
+async function resolveCommit(gitBin: string, cwd: string, ref: string): Promise<string | null> {
+  const r = await git(gitBin, cwd, ['rev-parse', '--verify', '--quiet', `${ref}^{commit}`])
+  const sha = r.stdout.trim()
+  return r.exit === 0 && sha.length > 0 ? sha : null
+}
+
+/**
+ * Render a fetch's outcome for an error message, so a genuine failure (wrong
+ * remote, auth failure, rate limiting) can be told apart from a benign one
+ * (no origin, offline, non-GitHub remote) when resolution ends up failing.
+ * Empty when the fetch succeeded, since a successful fetch explains nothing.
+ */
+function fetchDiagnostic(result: { exit: number; stderr: string }): string {
+  if (result.exit === 0) return ''
+  const stderr = result.stderr.trim()
+  return stderr ? `; git fetch: ${stderr}` : `; git fetch exited ${result.exit}`
+}
+
+/**
+ * Reproduce `gh pr diff` from the local clone. GitHub refuses the `.diff` media
+ * type above roughly 300 files, and a review that cannot get a diff is worth
+ * less than one built from local objects.
+ *
+ * Diffs from the merge base rather than the base tip so the output matches
+ * `gh pr diff`'s three-dot semantics: the PR's own changes, not every commit
+ * the base branch gained since the branch point.
+ */
+export async function localPrDiff(input: LocalDiffInput): Promise<LocalDiffResult> {
+  const { gitBin, repoPath, prNumber, baseRefName, headRefOid } = input
+  const prRef = `refs/magpie/pr-${prNumber}`
+
+  // Best effort. `pull/<n>/head` is a GitHub convention and resolves fork PRs
+  // that `headRefName` cannot, but a repo with no `origin`, an offline run, or
+  // a non-GitHub remote must still work when the commits are already local.
+  // Each fetch's own result is kept (not discarded) so, if resolution below
+  // still fails, the error can say why the fetch didn't help.
+  const headFetch = await git(gitBin, repoPath, [
+    'fetch',
+    'origin',
+    `pull/${prNumber}/head:${prRef}`,
+  ])
+  const baseFetch = await git(gitBin, repoPath, ['fetch', 'origin', baseRefName])
+
+  let head: string | null = null
+  for (const ref of [prRef, headRefOid]) {
+    head = await resolveCommit(gitBin, repoPath, ref)
+    if (head) break
+  }
+  if (!head) {
+    return {
+      ok: false,
+      error: `cannot resolve PR head ${headRefOid} locally (tried ${prRef} and the SHA); fetch the PR branch and retry${fetchDiagnostic(headFetch)}`,
+    }
+  }
+  // Reviewing a stale local ref and reporting it as a review of the PR is worse
+  // than failing, so a mismatch is fatal rather than a warning.
+  if (head !== headRefOid) {
+    return {
+      ok: false,
+      error: `local PR head ${head} does not match the PR head ${headRefOid}; the local ref is stale`,
+    }
+  }
+
+  let base: string | null = null
+  for (const ref of [`origin/${baseRefName}`, baseRefName]) {
+    base = await resolveCommit(gitBin, repoPath, ref)
+    if (base) break
+  }
+  if (!base) {
+    return {
+      ok: false,
+      error: `cannot resolve base branch ${baseRefName} locally (tried origin/${baseRefName} and ${baseRefName})${fetchDiagnostic(baseFetch)}`,
+    }
+  }
+
+  const mb = await git(gitBin, repoPath, ['merge-base', base, head])
+  if (mb.exit !== 0) {
+    return { ok: false, error: `git merge-base exit ${mb.exit}: ${mb.stderr.trim()}` }
+  }
+  const mergeBase = mb.stdout.trim()
+
+  const d = await git(gitBin, repoPath, ['diff', '--no-color', '--find-renames', mergeBase, head])
+  if (d.exit !== 0) {
+    return { ok: false, error: `git diff exit ${d.exit}: ${d.stderr.trim()}` }
+  }
+  return { ok: true, diff: d.stdout, mergeBase }
+}