@1aboveio/skills 0.0.0-stage → 1.0.0

package/skills/core/polish-cn/scripts/polish-cn.test.mjs

@@ -0,0 +1,244 @@
+import assert from 'node:assert/strict';
+import { spawnSync } from 'node:child_process';
+import { mkdtempSync, readFileSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { dirname, join } from 'node:path';
+import { fileURLToPath } from 'node:url';
+import test from 'node:test';
+
+const script = join(dirname(fileURLToPath(import.meta.url)), 'polish_blocks.py');
+
+const REPORT = `---
+title: 月报
+---
+# 2026 年 10 月报告
+
+本期授权尝试 4,946 笔，授权成功率 45.79%。0 笔销售未能匹配授权，覆盖六个月。
+详见 [方法说明](method.zh.md)。
+
+| 指标 | 值 |
+|---|---|
+| 授权尝试 | 4,946 |
+
+- 授权拒绝 1,111 笔，\`C1\` 为高风险。
+- 疑似欺诈响应不是已确认损失。
+
+\`\`\`python
+print("本期授权尝试 4,946 笔")
+\`\`\`
+`;
+
+function polish(...args) {
+  return spawnSync('python3', [script, ...args], { encoding: 'utf8' });
+}
+
+// Writes a document, extracts it, and returns helpers. `propose` keeps every block it does not
+// rewrite, so the proposal covers all blocks unless `partial` is set.
+function setup(content = REPORT) {
+  const dir = mkdtempSync(join(tmpdir(), 'polish-cn-test-'));
+  const doc = join(dir, 'report.zh.md');
+  writeFileSync(doc, content);
+  const blocksPath = join(dir, 'blocks.json');
+  const result = polish('extract', doc, '--out', blocksPath);
+  assert.equal(result.status, 0, result.stderr);
+  const blocks = JSON.parse(readFileSync(blocksPath, 'utf8')).blocks;
+  const byText = (needle) => blocks.find((block) => block.text.includes(needle));
+  const propose = (entries, { partial = false } = {}) => {
+    const path = join(dir, `proposals-${Math.random().toString(36).slice(2)}.json`);
+    const rewritten = {};
+    for (const [needle, text] of entries) {
+      const block = byText(needle);
+      rewritten[block.id] = { sha: block.sha, text };
+    }
+    const kept = partial ? {} : Object.fromEntries(
+      blocks.filter((block) => !rewritten[block.id]).map((block) => [block.id, 'already natural']));
+    writeFileSync(path, JSON.stringify({ blocks: rewritten, kept, notes: ['note kept'] }));
+    return path;
+  };
+  return { dir, doc, blocksPath, blocks, byText, propose };
+}
+
+test('extract returns prose only: front matter, tables and code fences are never blocks', () => {
+  const { blocks } = setup();
+  assert.deepEqual(blocks.map((block) => block.kind), ['heading', 'paragraph', 'list-item', 'list-item']);
+  assert.ok(blocks.every((block) => !block.text.includes('|') && !block.text.includes('print(')));
+  assert.equal(blocks[0].prefix, '# ');
+  assert.deepEqual(blocks[1].lines, [6, 7]);
+});
+
+test('extract skips HTML comments and blocks, indented code, pipe tables without edge pipes, and link definitions', () => {
+  const { blocks } = setup([
+    '正文一。', '', '<!--', '注释 99', '-->', '', '<div>', '内嵌 HTML 7', '</div>', '',
+    '    indented_code(3)', '', '指标 | 值', '--- | ---', '尝试 | 12', '', '[ref]: https://example.com/x', '', '正文二。', '',
+  ].join('\n'));
+  assert.deepEqual(blocks.map((block) => block.text), ['正文一。', '正文二。']);
+});
+
+test('check accepts a rewrite that keeps every protected token, even in a new order', () => {
+  const { blocksPath, propose } = setup();
+  const proposals = propose([[
+    '本期授权尝试',
+    '本期授权成功率为 45.79%，授权尝试共 4,946 笔。0 笔销售未能匹配授权，覆盖六个月。详见 [方法](method.zh.md)。',
+  ]]);
+  const result = polish('check', blocksPath, proposals);
+  assert.equal(result.status, 0, result.stdout + result.stderr);
+  assert.equal(JSON.parse(result.stdout).accepted.length, 1);
+});
+
+test('check fails when a block is neither rewritten nor kept, or is both, unless --partial', () => {
+  const { dir, blocksPath, byText, propose } = setup();
+  const partial = propose([['疑似欺诈', '疑似欺诈响应并非已确认的损失。']], { partial: true });
+  const strict = polish('check', blocksPath, partial);
+  assert.equal(strict.status, 1);
+  assert.equal(JSON.parse(strict.stdout).unreviewed.length, 3);
+  assert.equal(polish('check', blocksPath, partial, '--partial').status, 0);
+
+  const block = byText('疑似欺诈');
+  const both = JSON.parse(readFileSync(propose([['疑似欺诈', '疑似欺诈响应并非已确认的损失。']]), 'utf8'));
+  both.kept[block.id] = 'also kept';
+  const bothPath = join(dir, 'both.json');
+  writeFileSync(bothPath, JSON.stringify(both));
+  const result = polish('check', blocksPath, bothPath);
+  assert.equal(result.status, 1);
+  assert.deepEqual(JSON.parse(result.stdout).partitionProblems, [block.id]);
+});
+
+test('check rejects changed, added or dropped numbers, signs, currency, links, images, code and protected terms', () => {
+  const { blocksPath, propose } = setup();
+  const cases = [
+    ['本期授权尝试', '本期授权尝试 4,964 笔，授权成功率 45.79%。0 笔销售未能匹配授权，覆盖六个月。详见 [方法说明](method.zh.md)。', /number '4,946'/],
+    ['本期授权尝试', '本期授权尝试 4,946 笔，授权成功率 45.79%。0 笔销售未能匹配授权，覆盖 6 个月。详见 [方法说明](method.zh.md)。', /added number '6'/],
+    ['本期授权尝试', '本期授权尝试 4,946 笔，授权成功率 45.79%。0 笔销售未能匹配授权，覆盖六个月。', /link 'method\.zh\.md'/],
+    ['本期授权尝试', '本期授权尝试 4,946 笔，授权成功率 45.79%。0 笔销售未能匹配授权，覆盖六个月。详见 ![方法说明](method.zh.md)。', /image 'method\.zh\.md'/],
+    ['本期授权尝试', '本期授权尝试 4,946 笔，授权成功率 +45.79%。0 笔销售未能匹配授权，覆盖六个月。详见 [方法说明](method.zh.md)。', /symbol '\+'/],
+    ['本期授权尝试', '本期授权尝试 4,946 笔，授权成功率 45.79%，约 ¥0。0 笔销售未能匹配授权，覆盖六个月。详见 [方法说明](method.zh.md)。', /symbol '¥'/],
+    ['授权拒绝', '拒付 1,111 笔，`C1` 为高风险。', /term '授权拒绝'/],
+    ['授权拒绝', '授权拒绝 1,111 笔，C2 为高风险。', /code '`C1`'/],
+  ];
+  for (const [needle, text, reason] of cases) {
+    const result = polish('check', blocksPath, propose([[needle, text]]), '--protect', '授权拒绝', '--protect', '拒付');
+    assert.equal(result.status, 1, `expected rejection for: ${text}`);
+    const rejected = Object.values(JSON.parse(result.stdout).rejected);
+    assert.match(rejected[0], reason);
+  }
+});
+
+test('check reads TERMINOLOGY.md as concepts: mentions are kept, Avoid terms may become preferred but never appear', () => {
+  const { dir, blocks, blocksPath, propose } = setup('本月拒绝交易 1 笔，授权拒绝 2 笔。\n\n疑似欺诈响应不是已确认损失。\n');
+  const terms = join(dir, 'TERMINOLOGY.md');
+  writeFileSync(terms, [
+    '# Terminology: test', '', '## Disputes', '',
+    '| Concept | English | 简体 | 繁體（港） | Avoid | Meaning | Status | Source |',
+    '| --- | --- | --- | --- | --- | --- | --- | --- |',
+    '| Decline | decline | 授权拒绝 | 授權拒絕 | 拒绝交易、拒付（指授权拒绝） | Issuer refusal | user, 2026-10-20 | user |',
+    '| Chargeback | chargeback | 拒付 | — | 退单（指 Visa 拒付） | Dispute | proposed | a.md |',
+    '| Piped | pipe | 甲\\|乙 | — |  | Escaped pipe | proposed | a.md |', '',
+    '## Sources', '', '| id | Source |', '| --- | --- |', '| a.md | 授权拒绝 |', '',
+  ].join('\n'));
+  const run = (needle, text) => polish('check', blocksPath, propose([[needle, text]]), '--terminology', terms);
+
+  const normalized = run('拒绝交易', '本月授权拒绝 1 笔，授权拒绝 2 笔。');
+  assert.equal(normalized.status, 0, normalized.stdout);
+  assert.deepEqual(Object.keys(JSON.parse(normalized.stdout).avoidedTermsInSource), [blocks[0].id]);
+
+  const added = run('拒绝交易', '本月拒绝交易 1 笔，拒绝交易 2 笔。');
+  assert.equal(added.status, 1);
+  assert.match(Object.values(JSON.parse(added.stdout).rejected)[0], /avoids: 拒绝交易/);
+
+  const dropped = run('拒绝交易', '本月有 1 笔，授权拒绝 2 笔。');
+  assert.equal(dropped.status, 1);
+  assert.match(Object.values(JSON.parse(dropped.stdout).rejected)[0], /concept 'c1'.*c1 = 授权拒绝/);
+
+  const scopedNote = run('疑似欺诈', '疑似欺诈响应不是已确认损失，与拒付无关。');
+  assert.equal(scopedNote.status, 1, 'adding 拒付 adds a mention of the chargeback concept');
+  assert.match(Object.values(JSON.parse(scopedNote.stdout).rejected)[0], /c2 = 拒付/);
+
+  const emDash = run('疑似欺诈', '疑似欺诈响应——不是已确认损失。');
+  assert.equal(emDash.status, 0, 'an em dash is not a protected term: — means an empty cell');
+});
+
+test('check flags ids in kept that are not blocks', () => {
+  const { dir, blocksPath, propose } = setup();
+  const proposals = JSON.parse(readFileSync(propose([]), 'utf8'));
+  proposals.kept.b999_note = 'invented';
+  const path = join(dir, 'unknown.json');
+  writeFileSync(path, JSON.stringify(proposals));
+  const result = polish('check', blocksPath, path);
+  assert.equal(result.status, 1);
+  assert.deepEqual(JSON.parse(result.stdout).partitionProblems, ['b999_note (unknown id in kept)']);
+});
+
+test('check refuses line breaks and new Markdown structure inside a proposal', () => {
+  const { blocksPath, propose } = setup();
+  for (const text of ['疑似欺诈响应\n# 新标题', '疑似欺诈响应\r# 注入', '# 疑似欺诈响应不是已确认损失。', '- 疑似欺诈响应不是已确认损失。']) {
+    const result = polish('check', blocksPath, propose([['疑似欺诈', text]]));
+    assert.equal(result.status, 1, `expected rejection for: ${JSON.stringify(text)}`);
+    assert.match(Object.values(JSON.parse(result.stdout).rejected)[0], /one line|block syntax/);
+  }
+});
+
+test('check refuses deletions, stale blocks and malformed proposal files', () => {
+  const { dir, blocksPath, byText, propose } = setup();
+  const empty = polish('check', blocksPath, propose([['疑似欺诈', '  ']]));
+  assert.equal(empty.status, 1);
+  assert.match(empty.stdout, /never deletes a block/);
+
+  const block = byText('疑似欺诈');
+  const stale = JSON.parse(readFileSync(propose([['疑似欺诈', '疑似欺诈响应并非已确认的损失。']]), 'utf8'));
+  stale.blocks[block.id].sha = 'old';
+  const stalePath = join(dir, 'stale.json');
+  writeFileSync(stalePath, JSON.stringify(stale));
+  const result = polish('check', blocksPath, stalePath);
+  assert.equal(result.status, 1);
+  assert.match(result.stdout, /does not match block sha/);
+
+  const nullPath = join(dir, 'null.json');
+  writeFileSync(nullPath, 'null');
+  const malformed = polish('check', blocksPath, nullPath);
+  assert.equal(malformed.status, 2);
+  assert.match(malformed.stderr, /needs an object/);
+});
+
+test('apply rewrites accepted prose, keeps tables and code byte-identical, and writes the record', () => {
+  const { dir, doc, blocksPath, byText, propose } = setup();
+  const proposals = propose([
+    ['疑似欺诈', '疑似欺诈响应并非已确认的损失。'],
+    ['授权拒绝', '授权拒绝共 1,111 笔，`C1` 列为高风险。'],
+  ]);
+  const record = join(dir, 'record.json');
+  const result = polish('apply', doc, blocksPath, proposals, '--reject', byText('授权拒绝').id,
+    '--record', record, '--model', 'deepseek/deepseek-flash');
+  assert.equal(result.status, 0, result.stderr);
+  const after = readFileSync(doc, 'utf8');
+  assert.match(after, /^- 疑似欺诈响应并非已确认的损失。$/m);
+  assert.match(after, /^- 授权拒绝 1,111 笔，`C1` 为高风险。$/m, 'a caller-rejected block stays unchanged');
+  assert.ok(after.includes('| 授权尝试 | 4,946 |\n'));
+  assert.ok(after.includes('print("本期授权尝试 4,946 笔")'));
+  assert.ok(after.endsWith('```\n'), 'the trailing newline survives');
+  const written = JSON.parse(readFileSync(record, 'utf8'));
+  assert.equal(written.model, 'deepseek/deepseek-flash');
+  assert.equal(written.applied.length, 1);
+  assert.deepEqual(written.notes, ['note kept']);
+});
+
+test('apply keeps CRLF line endings and leaves HTML comments intact', () => {
+  const crlf = ['正文 123。', '', '<!--', '注释 99', '-->', '', '结尾。', ''].join('\r\n');
+  const { doc, blocksPath, propose } = setup(crlf);
+  const result = polish('apply', doc, blocksPath, propose([['正文', '正文为 123。']]));
+  assert.equal(result.status, 0, result.stderr);
+  assert.equal(readFileSync(doc, 'utf8'), crlf.replace('正文 123。', '正文为 123。'));
+});
+
+test('apply refuses an incomplete proposal file and a document changed after extract', () => {
+  const { doc, blocksPath, propose } = setup();
+  const incomplete = polish('apply', doc, blocksPath, propose([['疑似欺诈', '疑似欺诈响应并非已确认的损失。']], { partial: true }));
+  assert.equal(incomplete.status, 1);
+  assert.match(incomplete.stderr, /every block needs/);
+
+  const proposals = propose([['疑似欺诈', '疑似欺诈响应并非已确认的损失。']]);
+  writeFileSync(doc, REPORT.replace('疑似欺诈响应不是', '疑似欺诈响应并不是'));
+  const result = polish('apply', doc, blocksPath, proposals);
+  assert.equal(result.status, 1);
+  assert.match(result.stderr, /changed since extract/);
+  assert.ok(readFileSync(doc, 'utf8').includes('疑似欺诈响应并不是'));
+});