Industry Chokepoint Research

Industry Chokepoint Research

Default Behavior

When the user provides only an industry or theme, start a quick scan immediately. Do not ask for a configuration unless a missing choice would materially change the work.

Use this default adapter:

research_depth: "quick scan"
time_horizon: "12-36 months"
geography: "Global by default; keep the industry chain unified rather than splitting China and US into separate chains"
asset_scope:
  - public_equities
  - private_companies
  - suppliers
hidden_node_audit:
  - low-visibility enabling components
  - second-tier and third-tier suppliers
  - passive components / connectors / substrates / materials / test / metrology when relevant
public_market_coverage:
  - China A/H/B shares and important China-listed suppliers
  - US-listed companies and ADRs
  - Europe/Japan/Korea/Taiwan names when they control a relevant node
include_private_companies: true
evidence_policy:
  allow_tier4_as_lead_only: true
  require_tier1_for_company_claims: true
  require_financial_filings_for_public_companies: true
  min_sources_per_major_claim: 2

If the user says full DD, 深度研究, 只看中国, 只看美股, 3-6个月, 不看股票, or similar, override only that field.

Workflow

Follow the detailed framework in references/chokepoint-framework.md.

Minimum quick-scan sequence:

1. Define the demand shock, payer, budget source, why now, and misleading narratives.
2. Choose the 2-4 relevant chains: supply chain, value chain, delivery chain, regulatory chain, funding chain, data chain, or adoption chain. Build one unified global chain; use country/region as attributes on nodes and companies, not as separate chain maps.
3. Break each chain into concrete nodes, not broad concepts.
4. Score candidate bottlenecks with opportunity dimensions and risk deductions.
5. Expand supplier graphs for the top 3 nodes.
6. Run a hidden enabling-node audit before company mapping. For AI, semiconductors, data centers, electronics, power, industrial equipment, telecom, auto, aerospace, or other hardware-heavy chains, explicitly check low-visibility but necessary components such as passive components, MLCC, inductors, connectors, cables, PCB, IC substrates, packaging materials, thermal materials, power modules, test, metrology, and specialty consumables. Include them when demand linkage is real, even if they are second-tier nodes; label revenue purity and evidence quality clearly.
7. Map public and private companies by node, not by ticker popularity. Public-company mapping must include both China and US listed companies when relevant; add Europe/Japan/Korea/Taiwan names if they control an important node.
8. Run a peer-completeness pass within each included subnode. If one upstream material or component leader is included, add the major regional peers and China-listed/HK-listed equivalents when they have credible exposure. If a peer is excluded, state whether the reason is low purity, weak evidence, private status, or weak demand linkage.
9. Harvest evidence using the tier system; use Tier 4 only as leads.
10. Run financial reality checks for top public companies. Use current filings, market data, and company disclosures where available. If a live source is unavailable, mark the field as not verified or not disclosed; do not silently reuse stale caches.
11. Red-team every top thesis before writing conclusions.
12. Write a memo that gives research priority only, not buy/sell advice.

Data Adapter Rules

This public skill does not assume any private database, proprietary API, or local cache. If a user or project has local market-data adapters, use them only when they are available and permitted.

For public-company checks:

• Prefer official filings and company IR for revenue/profit facts.
• Prefer exchange, regulator, or well-documented market-data sources for price and valuation-sensitive fields.
• Label the data date and source family for every current-year price, revenue, profit, or valuation comparison.
• If a source requires an API key, read it from environment variables or local ignored config, never from tracked files.
• For current-year stock return, revenue growth, profit growth, and valuation-sensitive comparisons, do not use stale caches by default.

Output Contract

Create outputs under the current workspace unless the user specifies another path:

outputs/{industry_slug}/
  index.html
  01_research_questions.html
  02_chain_map.html
  03_bottleneck_scores.html
  04_supplier_graph.html
  05_company_universe.html
  06_evidence_log.html
  07_financial_checks.html
  08_red_team.html
  09_catalysts.html
  10_research_memo.html

  01_research_questions.md
  02_chain_map.md
  03_bottleneck_scores.csv
  04_supplier_graph.json
  05_company_universe.csv
  06_evidence_log.jsonl
  07_financial_checks.md
  08_red_team.md
  09_catalysts.csv
  10_research_memo.md

If time is limited, prioritize index.html, 10_research_memo.html, 03_bottleneck_scores.html, 05_company_universe.html, and 08_red_team.html.

Use scripts/render_outputs_html.py after writing sources to generate readable HTML pages and an index.html.

Quality Rules

• Do not output buy/sell recommendations.
• Do not equate industry growth with company benefit.
• Do not list tickers without explaining the chain relationship.
• Do not split the industry chain into China chain and US chain by default; create one global chain and mark geography per node/company.
• Do not omit China or US listed companies from company mapping when relevant public exposures exist.
• Do not omit low-visibility enabling suppliers simply because they are not the headline bottleneck.
• Do not include one regional leader in an upstream material node while omitting obvious peer leaders in other markets.
• Do not deliver only Markdown for reading artifacts; HTML is the default readable output format.
• Distinguish node is important, company has exposure, revenue can grow, profit can grow, and valuation still has room.
• Mark every key conclusion with an evidence tier.
• Treat unsupported conclusions as hypotheses to verify.
• For non-manufacturing sectors, rewrite supply chain as delivery chain, adoption chain, regulatory chain, funding chain, or data chain.
• For current market, filing, pricing, policy, or company evidence, use live sources when tools allow it; otherwise disclose that current verification was not available.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: guangfeizhao
• Source: guangfeizhao/industry-chokepoint-research-kit
• License: MIT

Install and usage instructions live in the source repository linked above.