Agent Wiki Hub is a reusable Agent Knowledge Pack repository. wikis/ remains the agent execution source of truth; v2 adds a generated Obsidian vault, local dashboard, RAG scaffold, controlled crawler scaffold, multimodal ingestion scaffold, source-review queueing, and safety audits.
v2.1 expands the hub from a complete framework into a high-density stable knowledge base. The added pages are model-synthesized stable domain knowledge: they are not authoritative sources, do not contain current facts, and do not change source-review evidence state.
New local capabilities:
- Knowledge density reports and audits for every wiki.
- Stable concepts, rules, workflows, cases, prompts, and evals across all 12 wikis.
- Wiki root
MOC.mdpages plus enhanced Obsidian MOCs and density dashboards. - Dashboard JSON for knowledge density, high-risk boundaries, current-fact gates, and MOC status.
- RAG fallback metadata for
current_fact,source_status,generated_by,risk_level, andhuman_gate_required.
High-risk or changing claims still require Source Review and human gates. This includes finance, legal, health, security, node operations, customs, airdrop/Web3, platform policies, current laws, current software versions, prices, fees, project status, and live vulnerability status.
python scripts/run_acceptance.py
python scripts/generate_knowledge_density_report.py
python scripts/audit_knowledge_density.py
python scripts/audit_current_fact_leakage.py
python scripts/audit_high_risk_boundaries.py
python scripts/generate_obsidian_vault.py
python scripts/generate_obsidian_canvas.py
python dashboard/scripts/collect_dashboard_data.py
python rag/search_knowledge.py --query "source review human gate" --top-k 5Open Obsidian, choose Open folder as vault, select obsidian-vault/, then open 05_Dashboard/Wiki Status.md or 02_Knowledge/MOCs/.
Run python dashboard/scripts/collect_dashboard_data.py, then open dashboard/index.html. The dashboard is static and does not require node_modules/.
The RAG layer is local-first. Chroma is optional; keyword fallback works without external dependencies.
python rag/build_chroma_index.py
python rag/search_knowledge.py --query "risk control" --top-k 5crawler/ contains conservative public-source crawler scaffolding. It defaults to dry-run/no-op, writes Raw notes only, respects configured source limits, and never writes directly to wikis/.
Raw material belongs in obsidian-vault/01_Raw/. Candidate knowledge belongs in obsidian-vault/02_Knowledge/Candidates/. OCR, Whisper, and Chroma are optional capabilities and are not required for acceptance.
Only low-risk stable_knowledge candidates with source metadata, sufficient confidence, and no secret/current-fact markers may be promoted. Promotion writes only to automation-generated wiki files and never overwrites human-authored pages.
Current facts and high-risk candidates go to source-review queues. They are planning-only until authoritative, dated, scoped evidence and required human gates are completed.
- No
.env, token, cookie, private key, bearer token, password, or credential is committed. - No current facts are written directly into stable wiki pages.
- High-risk domains retain human confirmation gates.
- Crawler output is Raw and unverified by default.
- Optional dependency absence is a warning, not an acceptance blocker.
- 35 open source update topics still require human or connected source verification.
- Verified tickets may remain 0 and
current_fact_readymay remain false. - OCR, Whisper, Chroma, network crawling, and GitHub Actions PR creation require optional dependencies, network access, or repository permission settings.
python scripts/generate_obsidian_vault.py
python scripts/generate_obsidian_canvas.py
python scripts/audit_obsidian_vault.py
python dashboard/scripts/collect_dashboard_data.py
python scripts/audit_dashboard.py
python ingestion/generate_ingestion_report.py
python scripts/audit_ingestion_pipeline.py
python crawler/generate_crawl_report.py
python scripts/classify_candidate_knowledge.py --dry-run
python scripts/promote_stable_knowledge.py --dry-run
python scripts/generate_source_review_from_candidates.py --dry-run
python scripts/audit_crawler_outputs.py
python scripts/audit_knowledge_promotion.py
python scripts/audit_rag_config.py
python scripts/audit_secret_leaks.py
python scripts/generate_knowledge_density_report.py
python scripts/audit_knowledge_density.py
python scripts/audit_current_fact_leakage.py
python scripts/audit_high_risk_boundaries.py
python scripts/generate_knowledge_expansion_summary.py
python scripts/generate_wiki_moc_pages.py
python scripts/generate_obsidian_backlinks.py
python scripts/run_acceptance.py这是一个面向 Codex / AI Agent 的行业知识库仓库模板。它把每个行业知识库做成可下载、可索引、可调用、可持续更新的 Agent Knowledge Pack。
生成日期：2026-05-26
- 每个行业一个独立 Wiki：
wikis/<domain>-agent-wiki/ - 每个 Wiki 都包含：概念、规则、流程、案例、工具、提示词、评测、来源记录
- Codex 通过根目录
AGENTS.md和各 Wiki 内部AGENTS.md判断什么时候读取哪个知识库 - 所有会变化的信息必须写入
sources/source-notes.md，并标记更新时间 - 所有高风险领域默认只做研究、模拟、审核与风控，不自动执行真实交易、真实法律/医疗/安全操作
cd agent-wiki-hub-starter
python3 scripts/validate_wiki.py
python3 scripts/check_registry_consistency.py
python3 scripts/audit_ci_workflow.py
python3 scripts/audit_page_metadata.py
python3 scripts/audit_content_coverage.py
python3 scripts/audit_links.py
python3 scripts/audit_pack_integrity.py
python3 scripts/check_eval_files.py
python3 scripts/update_index.py
python3 scripts/report_wiki_status.py
python3 scripts/list_source_updates.py
python3 scripts/generate_source_refresh_playbook.py
python3 scripts/generate_source_refresh_tickets.py
python3 scripts/generate_source_refresh_logs.py
python3 scripts/record_source_evidence.py --ticket-id TICKET-SRC-006 --status pending --dry-run
python3 scripts/import_source_evidence_packet.py
python3 scripts/generate_source_evidence_packet_fixtures.py
python3 scripts/audit_source_refresh_completion.py
python3 scripts/audit_source_evidence_quality.py
python3 scripts/generate_source_refresh_dashboard.py
python3 scripts/generate_source_refresh_wave_runner.py
python3 scripts/generate_source_reviewer_queue.py
python3 scripts/generate_source_review_session_plan.py
python3 scripts/generate_source_review_readiness_matrix.py
python3 scripts/generate_source_review_work_orders.py
python3 scripts/generate_source_review_packet_bundle.py
python3 scripts/generate_source_review_packet_classification.py
python3 scripts/generate_source_review_wave_plan.py --wave wave-2
python3 scripts/generate_source_review_wave_2_batch_plan.py
python3 scripts/generate_source_review_wave_3_plan.py
python3 scripts/generate_source_review_wave_packet_bundle.py --plan registry/source-review-wave-3-plan.json --stem source-review-session-wave-3-pending --work-order-dir registry/source-review-work-orders-wave-3
python3 scripts/generate_source_review_final_status.py
python3 scripts/audit_source_review_packets.py
python3 scripts/rehearse_source_review_packet_imports.py
python3 scripts/audit_safety_boundaries.py
python3 scripts/generate_hub_navigation.py
python3 scripts/generate_agent_routing_cards.py
python3 scripts/generate_agent_handoff.py
python3 scripts/generate_release_notes.py
python3 scripts/generate_change_summary.py
python3 scripts/route_wiki.py --query "risk control backtest"
python3 scripts/run_acceptance.py
python3 scripts/search_wiki.py --query "risk control" --wiki finance-agent-wikivalidate_wiki.py now checks both wiki structure and eval file integrity. check_eval_files.py remains available as a focused eval-only check.
check_registry_consistency.py verifies that registry records, wiki directories, and manifest metadata stay aligned.
audit_ci_workflow.py verifies GitHub Actions acceptance workflow expectations and writes docs/CI_AUDIT.md.
audit_page_metadata.py verifies front matter for reusable knowledge pages and writes docs/PAGE_METADATA_AUDIT.md.
audit_content_coverage.py verifies required topic coverage for each wiki and writes docs/COVERAGE_AUDIT.md.
audit_links.py verifies local Markdown links and internal path references, then writes docs/LINK_AUDIT.md.
audit_pack_integrity.py verifies generated zip packages and writes docs/PACK_AUDIT.md.
run_acceptance.py runs the full local acceptance suite and writes reports to docs/ and registry/.
list_source_updates.py builds a prioritized queue of topics that require authoritative source verification.
generate_source_refresh_playbook.py turns source-update topics into refresh tasks and writes docs/SOURCE_REFRESH_PLAYBOOK.md.
generate_source_refresh_tickets.py turns refresh tasks into execution tickets and writes docs/SOURCE_REFRESH_TICKETS.md.
generate_source_refresh_logs.py creates per-wiki source refresh log templates and writes docs/SOURCE_REFRESH_LOG_STATUS.md.
record_source_evidence.py records user-provided source evidence for a ticket and writes to the matching wiki source refresh log.
import_source_evidence_packet.py imports JSON/JSONL evidence packets after preflight validation and writes docs/SOURCE_EVIDENCE_PACKET_IMPORTER.md.
generate_source_evidence_packet_fixtures.py writes safe dry-run packet fixtures and expected-failure examples to registry/source-evidence-fixtures/.
audit_source_refresh_completion.py checks source refresh ticket completion evidence and writes docs/SOURCE_REFRESH_COMPLETION_AUDIT.md.
audit_source_evidence_quality.py checks completed evidence entry quality and writes docs/SOURCE_EVIDENCE_QUALITY_AUDIT.md.
generate_source_refresh_dashboard.py summarizes source refresh readiness and writes docs/SOURCE_REFRESH_DASHBOARD.md.
generate_source_refresh_wave_runner.py organizes open source-refresh tickets into wave, priority, and human-review batches without verifying external facts.
generate_source_reviewer_queue.py assigns open source-refresh tickets to generic reviewer roles and preserves human confirmation gates.
generate_source_review_session_plan.py turns reviewer queue cards into a concrete source-review session plan with safe packet placeholders.
generate_source_review_readiness_matrix.py summarizes per-ticket readiness across reviewer assignment, session selection, packet coverage, packet audit, rehearsal, and completion state.
generate_source_review_work_orders.py turns ready source-review tickets into offline work orders for human or connected source reviewers without verifying facts.
generate_source_review_packet_bundle.py exports the current source-review session as pending JSON/JSONL packet templates and a human checklist.
audit_source_review_packets.py checks source-review packet files for ticket validity, final-status readiness, human-review gates, duplicates, and obvious secrets.
rehearse_source_review_packet_imports.py dry-runs all source-review packet imports without writing evidence logs.
generate_agent_handoff.py writes docs/AGENT_HANDOFF.md and registry/agent-handoff.json for the next agent entry point.
audit_safety_boundaries.py checks high/medium risk wiki safety boundaries and writes docs/SAFETY_AUDIT.md.
generate_release_notes.py writes docs/RELEASE_NOTES.md and registry/release-manifest.json.
generate_change_summary.py writes docs/CHANGE_SUMMARY.md and registry/change-summary.json.
generate_hub_navigation.py writes docs/HUB_NAVIGATION.md and registry/hub-navigation.json.
generate_agent_routing_cards.py writes docs/AGENT_ROUTING_CARDS.md and registry/agent-routing-cards.json.
route_wiki.py routes a query to the best wiki and prints reading order, source gates, safety rules, and prohibited actions.
GitHub Actions workflow: .github/workflows/wiki-acceptance.yml
It runs python3 scripts/run_acceptance.py for pull requests, pushes to main or master, and manual dispatches. See docs/CI_USAGE.md.
~/agent-wiki-hub/具体项目通过软链接引用：
mkdir -p ~/projects/my-project/knowledge
ln -s ~/agent-wiki-hub/wikis/finance-agent-wiki ~/projects/my-project/knowledge/finance-agent-wiki把 CODEX_BUILD_PROMPT.md 的内容完整复制给 Codex。Codex 会根据这些规则继续扩展、校验、索引和打包知识库。