Data Sources

#	Database	Script	Source	Type
1	国家法律法规数据库 (NPC)	scripts/download.py	flk.npc.gov.cn	JSON API
2	国家规章库 (Gov Rules)	scripts/gov_rules_crawler.py	gov.cn/zhengce/xxgk/gjgzk	Athena API + HTML
3	外交条约库 (Treaty)	scripts/treaty_crawler.py	treaty.mfa.gov.cn	HTML
4	国务院政策文件库	scripts/gov_policy_library.py	sousuo.www.gov.cn	REST API (GET)
5	司法部行政法规库	scripts/moj_law_crawler.py	xzfg.moj.gov.cn	HTML
6	党内法规库	scripts/party_law_crawler.py	www.12371.cn	HTML
7	国防部法规文库	scripts/mod_law_crawler.py	www.mod.gov.cn	HTTP
8	税务法规库	scripts/tax_law_crawler.py	fgk.chinatax.gov.cn	REST API (POST)
9	生态环境部法规规章	scripts/mee_law_crawler.py	mee.gov.cn	HTML
10	最高人民法院发布栏目	scripts/court_law_crawler.py	court.gov.cn	HTML

All scripts share --search, --info, --size, --output, --rate-limit, --no-cache, --cache-stats, --cache-clear. Run python <script> --help for full parameter lists.

---

Core Rules

Search Strategy (applies to ALL data sources)

Decide before calling the script — never default to fuzzy for everything:

• Known title (e.g. "物业管理条例", "北京市生活垃圾管理条例"): use --exact to reduce noise.
• Broad topic (e.g. "出租车", "环境保护"): use fuzzy search (default).
• Full-text match: add --range content when you need documents where the keyword appears in the body, not just the title.
• Ambiguous: ask the user, or run both exact+fuzzy and compare.

Status Filtering

NPC status codes (sxx / --status):

Code	Meaning
1	已废止
2	已修改
3	现行有效
4	尚未生效

Other sources use their own status values (e.g. --status effective for MoJ). Check --help per script.

Article-Level Queries (NPC)

For individual articles instead of full documents:

Command	When	Output
--preview <id>	Understand structure first	Title, article count, TOC
--article <id> "第X条"	Know the article number	Single article text
--article <id> --grep "kw"	Find articles by keyword in one law	All matching articles

Supports Chinese numerals (第三十八条), Arabic (第38条), or bare numbers (38).

For cross-law article search, use scripts/article_search.py <keyword> --max-laws N.

Rate Limiting

Auto mode picks by estimated requests: OFF (≤10), FIXED 5 req/s (11–100), ADAPTIVE 1–8 req/s with 429 backoff (>100). Override: --rate-limit {off|fixed|adaptive}.

Caching

Local file cache enabled (~/.cache/ per namespace). Search results: 1h TTL. Detail metadata: 24h. DOCX: 7 days. Use --no-cache to bypass, --cache-stats to inspect.

---

Validity Rules — MANDATORY

1. Only state 现行有效, 已废止, 已修改, or 尚未生效 when the official page or API response explicitly provides that status.
2. If the source has no explicit validity field, say: "官方页面未明确标注效力状态".
3. Do not infer validity from file accessibility or recency of publication date.
4. Always return: official URL, file type, issuing authority, publication date, and retrieval date — even when status is absent.
5. For same-name / multi-version documents, flag the ambiguity and verify amendment, repeal, and effective-date information.
6. Distinguish document types: 法律, 行政法规, 地方性法规, 规章, 司法解释, 司法文件, 党内法规, 政策文件, 通知, 案例, 条约. Do not conflate.
7. Never write "现行法律" as a generic label for all official publications.

---

Document Type Attribution

When reporting results, always identify the document type and source database explicitly. Use the format:

来源: [数据库名称] ([URL]) 文件类型: [法律/行政法规/规章/司法解释/党内法规/政策文件/通知/...] 发布机关: [authority] 发布日期: [date] 效力状态: [现行有效/已废止/...] 或 "官方页面未明确标注效力状态"

---

Script Routing Summary

Script	Purpose	Key flags	Output dir
scripts/download.py	NPC search, download, article, preview	--search, --exact, --range, --status, --download, --article, --preview, --urls-only	stdout, files
scripts/article_search.py	Cross-law article-level search	keyword, --law, --range, --max-laws, --context, --json, --offset, --resume	stdout
scripts/gov_rules_crawler.py	Gov.cn rules	--search, --categories, --download, --info	gov_rules_output/
scripts/treaty_crawler.py	MFA treaties	--search, --collections, --download, --info	treaty_output/
scripts/gov_policy_library.py	State Council policies	--search, --range, --category, --year	gov_policy_output/
scripts/moj_law_crawler.py	MoJ admin regulations	--search, --range, --status	moj_law_output/
scripts/party_law_crawler.py	Party regulations (12371.cn)	--search, --category	party_law_output/
scripts/mod_law_crawler.py	MOD law library	--search, --category	mod_law_output/
scripts/tax_law_crawler.py	Tax regulations (chinatax)	--search, --category	tax_law_output/
scripts/mee_law_crawler.py	MEE env regulations	--search, --category	mee_law_output/
scripts/court_law_crawler.py	SPC announcements	--search, --category	court_law_output/
scripts/region_classifier.py	Province/city classification	--classify, --matrix	JSON/CSV

Each script has --help. Run it for detailed parameter lists.

---

Reference Files

Read these on demand, not proactively:

File	Read when...
references/setup.md	First run, dependency issues, or switching agent environments
references/api_reference.md	Need full NPC API payloads, field mappings, auth, or pagination details
references/gov_rules_api_reference.md	Gov Rules auth issues (401/403) or response field parsing
references/gov_policy_api_reference.md	Need State Council API parameter details beyond --help
references/treaty_api_reference.md	Treaty HTML structure or field extraction patterns
references/page_structure.md	Browser automation — page layout, selectors, UI-only search flow
references/kimi_bridge_adapter.md	Running on Kimi Agent or Claude Code + kimi-webbridge
references/codex_adapter.md	Running on Codex
references/batch_collection.md	Large-scale collection (200+ items), batch download workflows

Browser fallback: only when the API/script fails or UI-only search is required. Read references/page_structure.md first.

---

Key Notes by Data Source

• NPC: ossFile paths from detail API are not directly downloadable — always use --download. bbbs = unique law ID.
• Gov Rules: Requires dynamic RSA auth from the frontend JS bundle; may expire during long runs. Restart on 401/403.
• Treaty: Pure HTML, no API. Collections: 全部, 双边, 多边.
• Party Law (12371.cn): No search API; keyword filtering is client-side on titles only.
• MOD: Static HTML, no search API. Client-side title filtering.
• Tax: POST API, requires session cookie (auto-managed). Channel ID auto-discovered.
• MEE: Static HTML, categories pre-loaded in page. No search API. Standards (环保标准) excluded.
• Court: List pages paginated (/fabu/gengduo/{id}.html → {id}_{n}.html). No search API.