@iceinvein/agent-skills 0.1.38 → 0.1.39

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

@@ -2,10 +2,35 @@ import { expect, test } from 'bun:test'
 import { readFile } from 'node:fs/promises'
 
 const SKILL = new URL('../../SKILL.md', import.meta.url).pathname
+const SKILL_JSON = new URL('../../skill.json', import.meta.url).pathname
+const ref = (name: string) => new URL(`../../references/${name}`, import.meta.url).pathname
 const FOCUSES = ['security', 'bugs', 'performance', 'code-smells', 'architecture'] as const
 
-test('SKILL.md has a specialist block for every focus', async () => {
+test('every references/ path SKILL.md cites exists on disk', async () => {
   const text = await readFile(SKILL, 'utf8')
+  const cited = [...text.matchAll(/references\/[a-z0-9-]+\.md/g)].map((m) => m[0])
+  expect(cited.length).toBeGreaterThan(0)
+  for (const rel of new Set(cited)) {
+    const body = await readFile(new URL(`../../${rel}`, import.meta.url).pathname, 'utf8')
+    expect(body.length).toBeGreaterThan(0)
+  }
+})
+
+test('references/ ships in the install bundle', async () => {
+  // resolveBundlePaths silently drops include entries that match nothing, so a
+  // missing entry here would install a SKILL.md whose prompts are all 404s.
+  const manifest = JSON.parse(await readFile(SKILL_JSON, 'utf8')) as {
+    bundle: { include: string[] }
+  }
+  const covers = (p: string) =>
+    manifest.bundle.include.some((inc) => inc === p || p.startsWith(`${inc}/`))
+  for (const name of ['specialists.md', 'critic.md', 'peer-review.md']) {
+    expect(covers(`references/${name}`)).toBe(true)
+  }
+})
+
+test('references/specialists.md has a block for every focus', async () => {
+  const text = await readFile(ref('specialists.md'), 'utf8')
   for (const focus of FOCUSES) {
     const tag = `magpie-specialist-${focus}`
     const fence = `\`\`\`${tag}`
@@ -13,23 +38,69 @@ test('SKILL.md has a specialist block for every focus', async () => {
     const start = text.indexOf(fence)
     const end = text.indexOf('```', start + tag.length + 3)
     const block = text.slice(start, end)
-    expect(block).toContain('orchestrator')
+    expect(block).toContain('Output Contract')
     expect(block).toContain(focus)
   }
 })
 
-test('SKILL.md §4 specifies the findings file path', async () => {
-  const text = await readFile(SKILL, 'utf8')
-  // The file-write instruction lives once, in the orchestrator template before §5.
-  const orchestrator = text.slice(0, text.indexOf('```magpie-specialist-'))
-  expect(orchestrator).toMatch(/findings\/<focus>\.json/)
-  expect(orchestrator).toMatch(/Write findings to/i)
+test('references/specialists.md carries the output contract next to the blocks', async () => {
+  const text = await readFile(ref('specialists.md'), 'utf8')
+  // The contract has to travel with the prompts: the orchestrator assembles
+  // both into one subagent prompt from this single file.
+  const contract = text.slice(0, text.indexOf('```magpie-specialist-'))
+  expect(contract).toContain('## Output Contract')
+  expect(contract).toMatch(/findings\/<focus>\.json/)
+  expect(contract).toMatch(/Write findings to/i)
+  // Severity, impact, likelihood, confidence, action enums must all be listed.
+  expect(contract).toMatch(/"blocker".*"high".*"medium".*"low"/)
+  expect(contract).toMatch(/"critical".*"high".*"medium".*"low"/)
+  expect(contract).toMatch(/"likely".*"possible".*"edge-case".*"unknown"/)
+  expect(contract).toMatch(/"must-fix".*"should-fix".*"consider".*"optional"/)
+  expect(contract).toMatch(/"impact"[\s\S]*"likelihood"[\s\S]*"confidence"[\s\S]*"action"/)
+  expect(contract).toMatch(/"body"[\s\S]*"startLine"[\s\S]*"endLine"/)
+  // Anti-patterns flagged explicitly so subagents don't repeat the JSON-shape mistakes.
+  expect(contract).toMatch(/NOT "lines"/)
+  expect(contract).toMatch(/NOT "recommendation"/)
 })
 
-test('SKILL.md has the critic and peer-review blocks', async () => {
-  const text = await readFile(SKILL, 'utf8')
+test('references/critic.md holds the rubric and both placeholders', async () => {
+  const text = await readFile(ref('critic.md'), 'utf8')
   expect(text).toContain('```magpie-critic')
+  expect(text).toContain('<<DEDUPED_FINDINGS_COMPACT>>')
+  expect(text).toContain('<<DIFF_EXCERPT>>')
+  expect(text).toContain('review-critic')
+})
+
+test('references/peer-review.md holds the prompt and the Claude preamble', async () => {
+  const text = await readFile(ref('peer-review.md'), 'utf8')
   expect(text).toContain('```magpie-peer-review')
+  expect(text).toContain('```magpie-peer-review-claude-preamble')
+  expect(text).toContain('review-peer-review')
+  for (const ph of ['<<PRIMARY_PROVIDER>>', '<<PEER_PROVIDER>>', '<<KEPT_FINDINGS_COMPACT>>']) {
+    expect(text).toContain(ph)
+  }
+})
+
+test('SKILL.md sends each stage to the reference file it needs', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  const section = (heading: string) => {
+    const start = text.indexOf(heading)
+    expect(start).toBeGreaterThan(-1)
+    const next = text.indexOf('\n### ', start + heading.length)
+    return text.slice(start, next === -1 ? undefined : next)
+  }
+  expect(section('### 3. Specialists')).toContain('references/specialists.md')
+  expect(section('### 5. Critic')).toContain('references/critic.md')
+  expect(section('### 6. Peer review')).toContain('references/peer-review.md')
+})
+
+test('SKILL.md no longer inlines the prompt bodies it moved out', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  for (const tag of ['magpie-specialist-', 'magpie-critic', 'magpie-peer-review']) {
+    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)
 })
 
 test('styles.css declares a prefers-color-scheme:dark block that overrides core tokens', async () => {
@@ -104,25 +175,42 @@ test('severity chips and submit button no longer hardcode `color: white` (must f
   }
 })
 
-test('SKILL.md §4 inlines the full specialist schema with enum vocabularies', async () => {
+test('SKILL.md checks for a resumable run before minting a new run id', async () => {
   const text = await readFile(SKILL, 'utf8')
-  // The schema block sits inside the specialist orchestrator template (§4),
-  // before the specialist blocks begin (the first `magpie-specialist-` tag).
-  const schemaCutoff = text.indexOf('```magpie-specialist-')
-  expect(schemaCutoff).toBeGreaterThan(-1)
-  const orchestrator = text.slice(0, schemaCutoff)
-  // Severity, impact, likelihood, confidence, action enums must all be listed.
-  expect(orchestrator).toMatch(/"blocker".*"high".*"medium".*"low"/)
-  expect(orchestrator).toMatch(/"critical".*"high".*"medium".*"low"/)
-  expect(orchestrator).toMatch(/"likely".*"possible".*"edge-case".*"unknown"/)
-  expect(orchestrator).toMatch(/"must-fix".*"should-fix".*"consider".*"optional"/)
-  // The risk field must be documented as an object with the four required keys.
-  expect(orchestrator).toMatch(/"impact"[\s\S]*"likelihood"[\s\S]*"confidence"[\s\S]*"action"/)
-  // suggestion shape: body/startLine/endLine.
-  expect(orchestrator).toMatch(/"body"[\s\S]*"startLine"[\s\S]*"endLine"/)
-  // Anti-patterns flagged explicitly so subagents don't repeat the JSON-shape mistakes.
-  expect(orchestrator).toMatch(/NOT "lines"/)
-  expect(orchestrator).toMatch(/NOT "recommendation"/)
+  // §0 always computed a fresh `pr-<n>-<epoch>` id, so the resume section
+  // (which keyed off that brand-new directory) could never fire.
+  const setupSection = text.slice(0, text.indexOf('### 1. Setup'))
+  expect(setupSection).toContain('magpie --list-runs')
+  expect(setupSection).toMatch(/active/i)
+})
+
+test('SKILL.md does not gate resume on state/server-info', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  const start = text.indexOf('## Resuming a crashed run')
+  expect(start).toBeGreaterThan(-1)
+  const section = text.slice(start, text.indexOf('## Aborting', start))
+  // server-info is deleted when the server idles out, so a resumable run
+  // fails that check. log.jsonl is the durable signal.
+  expect(section).toContain('log.jsonl')
+  expect(section).not.toMatch(/if .*server-info.* exists/i)
+  // Resuming must restart the server; the old one is gone.
+  expect(section).toContain('magpie serve')
+  // `context` is a no-op stage: say so, or the agent stalls on it.
+  expect(section).toContain('context')
+})
+
+test('SKILL.md names the report buttons that actually exist', async () => {
+  const text = await readFile(SKILL, 'utf8')
+  const actionBar = await readFile(
+    new URL('../render-action-bar.ts', import.meta.url).pathname,
+    'utf8',
+  )
+  for (const label of ['Post Selected', 'Post Recommended']) {
+    expect(actionBar).toContain(label)
+    expect(text).toContain(label)
+  }
+  // The old label was never rendered anywhere.
+  expect(text).not.toContain('Post to PR')
 })
 
 test('SKILL.md has the stage walkthrough', async () => {

package/skills/magpie/scripts/__tests__/status-cmd.test.ts

@@ -31,6 +31,19 @@ test('empty log reports nothing completed', async () => {
   expect(result.next).toBe('setup')
 })
 
+test('a skipped stage advances the resume pointer past it', async () => {
+  // `context` has no work in the pipeline; SKILL.md tells the orchestrator to
+  // log it as skipped. If `skipped` did not advance the pointer, a resume
+  // would be sent back to a stage that has no step to run.
+  await writeFile(
+    join(runDir, 'log.jsonl'),
+    `${JSON.stringify({ stage: 'setup', status: 'done' })}\n${JSON.stringify({ stage: 'context', status: 'skipped' })}\n`,
+  )
+  const result = await runStatus(runDir)
+  expect(result.lastCompleted).toBe('context')
+  expect(result.next).toBe('specialists')
+})
+
 test('error stage halts progression', async () => {
   await writeFile(
     join(runDir, 'log.jsonl'),

package/skills/magpie/scripts/helper.js

@@ -352,6 +352,23 @@
     }
   }
 
+  function isSelected(id) {
+    return findCheckboxes(id).some((cb) => cb.checked && !cb.disabled)
+  }
+
+  // Assigning `cb.checked` in script fires no change event, so any programmatic
+  // selection has to emit its own /events record. `state/events` is the only
+  // channel the orchestrator can read when the user types `post` in the
+  // terminal instead of using the in-page buttons; a silent set drops the
+  // finding from that post.
+  function setCheckedAndNotify(id, checked) {
+    const before = isSelected(id)
+    setChecked(id, checked)
+    const after = isSelected(id)
+    if (after === before) return
+    post({ type: after ? 'select' : 'deselect', findingId: id, timestamp: Date.now() })
+  }
+
   // ---------------------------------------------------------------------------
   // Bulk selection
   // ---------------------------------------------------------------------------
@@ -367,8 +384,8 @@
       if (el.getAttribute('data-posted') === 'true') continue
       ids.add(el.getAttribute('data-finding-id'))
     }
-    for (const id of ids) setChecked(id, true)
-    updateSelectedCount()
+    for (const id of ids) setCheckedAndNotify(id, true)
+    recountSelected()
   }
 
   function handleSelectRecommended() {
@@ -378,8 +395,8 @@
       if (el.getAttribute('data-posted') === 'true') continue
       ids.add(el.getAttribute('data-finding-id'))
     }
-    for (const id of ids) setChecked(id, true)
-    updateSelectedCount()
+    for (const id of ids) setCheckedAndNotify(id, true)
+    recountSelected()
   }
 
   // ---------------------------------------------------------------------------
@@ -399,14 +416,8 @@
     if (cbs.length === 0) return
     const cb = cbs[0]
     if (cb.disabled) return
-    const next = !cb.checked
-    setChecked(id, next)
-    post({
-      type: next ? 'select' : 'deselect',
-      findingId: id,
-      timestamp: Date.now(),
-    })
-    updateSelectedCount()
+    setCheckedAndNotify(id, !cb.checked)
+    recountSelected()
   }
 
   // ---------------------------------------------------------------------------
@@ -535,7 +546,7 @@
         if (cb.disabled) continue
         cb.checked = target.checked
       }
-      updateSelectedCount()
+      recountSelected()
       post({
         type: target.checked ? 'select' : 'deselect',
         findingId: id,

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

@@ -13,6 +13,12 @@ const ORDER = [
 ] as const
 
 export type StatusResult = {
+  /**
+   * Highest stage the log says is behind us. A stage logged `skipped` counts:
+   * `context` has no work in the pipeline and is always logged that way, so
+   * treating it as unfinished would send a resume back to a stage that has no
+   * step to run.
+   */
   lastCompleted: (typeof ORDER)[number] | null
   next: (typeof ORDER)[number] | 'cleanup'
   error: string | null
@@ -35,7 +41,10 @@ export async function runStatus(runDir: string): Promise<StatusResult> {
       error = stage
       break
     }
-    if (status === 'done' && (ORDER as readonly string[]).includes(stage)) {
+    if (
+      (status === 'done' || status === 'skipped') &&
+      (ORDER as readonly string[]).includes(stage)
+    ) {
       lastCompleted = stage as StatusResult['lastCompleted']
     }
   }

package/skills/magpie/skill.json

@@ -1,6 +1,6 @@
 {
   "name": "magpie",
-  "version": "0.7.1",
+  "version": "0.8.0",
   "description": "Interactive PR review pipeline. Runs five parallel specialist subagents (security, bugs, performance, code-smells, architecture), dedupes findings, applies a critic rubric, peer-reviews via codex exec (falling back to a Claude second opinion when codex is unavailable), and serves an interactive HTML report for selecting findings to post via gh. Bundles a Bun CLI installed onto PATH via the skill's postinstall step. Use when the user asks to review a GitHub pull request.",
   "author": "iceinvein",
   "type": "prompt",
@@ -13,6 +13,7 @@
   "bundle": {
     "include": [
       "bin",
+      "references",
       "scripts",
       "templates",
       "fixtures/example-pr",