factiq

name	factiq
description	Answer economic and financial data questions with real FactIQ (worlddb) data: US indicators (BLS employment/CPI, BEA GDP, Census trade, EIA energy, USDA ERS, BTS transport); international data (China NBS/customs, India MOSPI/RBI/trade, EU Comext, Singapore, IMF, World Bank); stocks, commodities/forex; earnings-call transcripts; a curated business-news feed; and satellite-derived data: nighttime lights by country and state, lake and reservoir water levels, daily shipping and port activity (chokepoint transits, port calls, seaborne trade estimates), plus on-demand signals (fire detections/crop burning with mappable footprints, NO2/SO2/CO industrial and combustion activity, smoke and dust aerosol, crop-condition NDVI, monsoon rainfall, heatwaves, soil moisture — by country, state, or bounding box). Use for unemployment, inflation, GDP, trade flows, energy, wages, markets, recent news context, stubble burning, wildfires, air quality, monsoon or drought conditions, nighttime lights, reservoir levels, shipping or chokepoint traffic, economic charts or maps, terminal previews, multi-section research reports, and custom HTML dashboards. Discover series, query SQL, compute, then return a sourced answer or publish/render the requested output. For bilateral trade, use the bundled SQL generators instead of hand-writing schema-specific queries.
allowed-tools	mcp__plugin_factiq_factiq__*, mcp__factiq__*, mcp__claude_ai_FactIQ__*, Bash(python3:*), Bash(python:*), Read, Write, Agent

FactIQ Data Tools

You are the analyst. FactIQ provides authenticated MCP tools for the whole loop — discover the data (catalog, dataset/series search, read-only SQL, series lookup, market data, earnings-transcript search), then publish the result (share_chart, share_report). There is no server-side agent: you decompose the question, find the data with the MCP tools, do the math with your own tokens, then either answer directly in a sentence or author the output and publish it with a tool call.

Five output modes:

• Direct answer — a plain-text sentence, no chart and no share link. Use when the question asks for a single current value or a simple scalar lookup where a chart would add nothing: "what's the US unemployment rate right now?", "latest CPI print", "Apple's trailing P/E". Still fetch the value with the MCP tools — only the presentation is a sentence. State the number with its period and source (e.g. "US unemployment was 4.1% in May 2026, per BLS."). The moment the question wants a trend, a history, a comparison across categories or entities, a breakdown, or explicitly asks for a chart or report, switch to one of the modes below.
• Quick chart (share_chart tool + term_chart.py) — one focused chart published to FactIQ as a share link, plus an inline terminal preview rendered from the same ChartSpec. Default for a single metric over time or a comparison across categories or entities. When geography is the finding this includes maps: choropleths by country or state/province and coordinate bubbles — see references/output/chart-spec.md (Maps) for the format and region-name rules (the terminal preview degrades to a ranked table).
• Terminal chart (term_chart.py) — an ANSI/ASCII preview without a share link. Use only when the user explicitly asks for terminal-only, ASCII-only, or local text output. See Terminal charts below.
• Detailed report (share_report tool) — summary + sections of narrative and charts + methodology, rendered on FactIQ's share-report page exactly like the in-house agent's reports, plus inline terminal previews of its charts. For broad or analytical questions. See Detailed reports below. Broad questions in a covered domain (bilateral trade, economic policy, monetary policy, fiscal revenue, business formation — see references/report-patterns/README.md) default here unless the user explicitly asks for only a quick chart. If the request is vague or comprehensive, run the explorer-agent interview in references/report-patterns/interview-step.md before data work; it clarifies scope without replacing the dialectical report method.
• Bespoke local viz (build_viz.py) — a self-contained HTML file you author freely and save locally, not published to FactIQ. Use when the answer needs something the ChartSpec can't express: a custom layout, a multi-panel dashboard, a force/flow/chord diagram, a novel encoding, or fine-grained visual control. See Bespoke local visualizations below.

Data in, output out:

• All discovery, fetching, and publishing go through the FactIQ MCP tools (factiq MCP). No codebase, database, or API key is needed — the coding agent calls them directly over one authenticated connection.

• The local scripts never touch the API:

  python3 "{plugin_root}/scripts/term_chart.py" render ... # terminal ChartSpec preview
  python3 "{plugin_root}/scripts/build_viz.py"  ...   # local visualization helper
  python3 "{plugin_root}/scripts/comext_sql.py" ...   # SQL generator: Eurostat Comext (EU) trade
  python3 "{plugin_root}/scripts/trade_sql.py"  ...   # SQL generator: US/China/India/Korea/Japan/Taiwan customs
  python3 "{plugin_root}/scripts/hs_codes.py"   ...   # HS commodity code <-> name, offline
  python3 "{plugin_root}/scripts/series_math.py" ...  # YoY/YTD/share/index/merge on saved results
  

  Resolve {plugin_root} once, then reuse it. In Claude Code it is ${CLAUDE_PLUGIN_ROOT}. In Codex, start from the absolute path supplied for this SKILL.md: the plugin root is two directories above the directory containing this file (skills/factiq/../..). Never resolve these scripts from the shell's current working directory or from a similarly named scripts/ directory in the user's project. Keep the quotes around the absolute path so installations under a directory containing spaces still work.

  For any bilateral-trade question, generate the SQL instead of writing it. comext_sql.py and trade_sql.py encode each schema's series-ID grammar, partner-code system, units, and HS-level rules, so the query is correct by construction — run --help on either for the subcommands (total / products / trend). Label the HS codes a ranking returns with hs_codes.py (zero server calls), and do growth/share arithmetic with series_math.py on results saved via build_viz.py save rather than in your own output.

Setup

One connection covers everything: the FactIQ MCP server bundled with this plugin (.mcp.json), authorized over OAuth. On first use the coding agent runs FactIQ's browser-based Connect flow. If the FactIQ tools are missing or return an auth error, the connection isn't set up yet — tell the user to authorize the MCP server:

• Claude Code: run /mcp, pick factiq, and complete the sign-in.
• Codex: run codex mcp login factiq and complete the sign-in.

The same FactIQ login works everywhere (email, Google, or passkey). Nothing to copy or paste, and no separate key for publishing — the same connection authorizes share_chart / share_report.

Local development. The bundled MCP URL is https://api.factiq.com/mcp. For a local backend, edit .mcp.json in your development checkout or configure a standalone factiq MCP server in Codex or Claude Code that points at the local URL.

Tools

All FactIQ tools are MCP tools provided by the factiq MCP server.

Data

Tool	Purpose
get_data_catalog (schemas?, full?)	Per-schema index + the shared table DDL. Call once per session before anything else. full=true returns the heavy per-dataset dump (rarely needed — use describe_dataset). Schemas listed under schemas_without_data have no rows — skip them.
search_datasets (query, schemas?, limit?)	Keyword (not semantic) ranking of datasets across all schemas. The first discovery step — find the right schema + dataset_code.
describe_dataset (schema, dataset_code)	Full metadata for one dataset: topic, methodology, release dates, base-change notice, dimensions, example series. Call after search_datasets.
search_series (schema, terms, limit?, include_compound?)	Series-level title-substring search within one schema (terms is a list — prefer short stems). Includes COMPOUND:: series.
run_sql (schema, sql, question?, explore?, auto_retry?)	Read-only SELECT against one schema. The power tool for joins, pivots, aggregation.
get_series (schema, series_id, from_year?, to_year?)	Fetch one series — timeseries, tabular, or COMPOUND:: ids all work.
get_market_data (function, symbol?, interval?, outputsize?)	Quotes, daily/weekly/monthly series, fundamentals (OVERVIEW, INCOME_STATEMENT, EARNINGS), FX, commodities (WTI, BRENT, GOLD), SYMBOL_SEARCH.
get_geo_data (dataset, region, start_date, end_date, aggregation?)	Satellite-derived signals, fetched live rather than stored as series: fires_viirs (crop burning/wildfires, ~3h lag; aggregation="grid" maps the fire's footprint), no2_tropomi / so2_tropomi / co_tropomi (industrial, coal/smelting, and combustion activity), aerosol_index_tropomi (smoke/dust/haze), ndvi_s2 (crop condition), precip_chirps (rainfall), temperature_power, soil_moisture_power — aggregated over a country, state ("India/Punjab"), or bbox. Read references/data/satellite.md before first use — it covers windows (max 50 intervals; grid max 92 days), the valid_obs_share rule, and attribution.
search_earnings_transcripts (query, search_target?, company_filter?, quarter_filter?, claim_family?, section?, detail?, limit?)	Full-text (lexical, not semantic) search over earnings-call transcripts — atomic, quote-anchored rows decomposed from live calls, never a raw transcript dump. All query terms must match (websearch syntax; fuzzy fallback catches typos) — retry with the company's own vocabulary ("capital expenditure" vs "capex") before concluding silence. An empty query browses the newest rows: with company_filter that reads one call's claim spine in spoken order. search_target: "claims" (default — management's guidance/comparisons/quantified statements, with direction/value/unit where checkable plus assertion_status — never present an analyst_hypothesized row as management's claim), "pressure_points" (what analysts pressed for in Q&A and whether management confirmed/declined/deflected, incl. the specific number refused), "disclosure_profile" (ticker-only lookup of what a company routinely discloses vs. withholds), or "coverage" (which tickers/quarters exist, with claim counts — call it first when unsure a company is covered). Narrow with company_filter (ticker or comma-separated "MU,NVDA") / quarter_filter (e.g. "FY2026Q3") / claim_family (18-code vocabulary — invalid values return the list) / section ("prepared_remarks" vs "qa", claims only). detail=true adds structured_fields (the per-family quantified payload) and other normalized columns. limit is 1–50 (default 15). For filed XBRL financials use run_sql on the sec schema; for formally issued targets use sec_guidance — this tool covers what was said live. Full workflow: references/report-patterns/earnings-intelligence.md.
search_media_appearances (query, search_target?, company_filter?, person?, sort?, date_from?, date_to?, limit?)	What executives said outside earnings calls — podcasts, TV interviews, and conference appearances (9,000+ videos across 900+ listed companies, 2019–present). Unlike the other search tools this one is question-driven, not keyword-driven: a search agent on FactIQ's side takes your query, runs the transcript searches itself (iterating phrasings and synonyms), and returns paraphrased findings — person, date, one sentence on what was said, and a YouTube link that jumps to the moment it was said. So don't just pass a name — ask the actual question: "What has Jamie Dimon said about recession risk?", "Has any executive discussed GLP-1 drugs affecting their business?", "How has Jensen Huang's framing of export controls changed over time?". Cross-company theme questions work as well as person-specific ones. Findings come back newest-first with dates, so evolving-position timelines fall out naturally. Each call takes ~10–30 s (the agent is searching, not doing one lookup) — put what you want to know into one well-phrased question rather than several fragments. Transcript text is never returned; treat findings as sourced leads and cite the timestamped links. Narrow with company_filter (ticker), person (name substring — everyday media names like "Jensen Huang" work), date_from/date_to, and sort="newest" when you want the latest statements specifically. search_target="coverage" lists which executives are covered (video counts, date span, hours of footage) — call it first when unsure a person or company is in the corpus. Free-plan usage is capped (200 searches or $5 of search compute per calendar month). Pairs with search_earnings_transcripts: that tool covers what management said on the call; this one covers what they say off-script — comparing the two surfaces tone shifts and detail that never reaches an earnings call.
search_news (query?, tickers?, topic?, sources?, start_date?, end_date?, sort?, limit?)	Search FactIQ's curated business-news feed — public RSS headlines and summaries from Bloomberg, the Financial Times, and the Wall Street Journal, plus India-macro (Zerodha Daily Brief, ET HealthWorld) and global-health sources (WHO, ECDC, CDC, STAT News, KFF), aggregated and processed by FactIQ so each article carries the listed companies it names ({symbol, exchange, country}) and an analysis block: searchable keywords, a geography, and an angle — one sentence on why the story matters to an investor. Company stories get analysis too, not just macro ones; only content with no business read at all (sports, lifestyle, celebrity) comes back with analysis: null. Pivot from a story into data: company stories → get_market_data / search_earnings_transcripts / run_sql on sec; macro stories → search_series / run_sql. Results are headline + short publisher summary + link out, never full articles. query is lexical full-text over headline+summary — start with short concrete stems ("obesity drug", "rate cut"); if a multi-word query matches nothing in full, the tool automatically retries matching ANY of the words with rare words ranked first, flagged as meta.query_mode: "any_term", so one query attempt is usually enough. tickers matches share classes and cross-listings automatically (GOOG also finds GOOGL-tagged articles, TSM its Taiwan listing) — pass whichever symbol you know; most macro stories name no listed company, so zero ticker matches is a normal answer. topic is one of markets / economics / companies / technology / politics / world / energy / health / india / opinion — combined with a query it is a ranking preference (matching sections rank first, but strong matches from other sections still return, since stories often run outside their obvious feed); without a query it filters to the topic's feeds. sort is "latest" (default) or "relevance" (needs a query); limit 1–50 (default 20). Coverage is recent news (most feeds start late 2025), and the feed is continuously being expanded and improved — treat it as a current-events lens, not an archive.
get_style_guides (guides)	FactIQ's house-style guides ("chart", "report", "sql", "earnings", or "all"). Optional; this skill's references/ already cover the publishing JSON formats — use these guides for extra house-style detail. Fetch "earnings" before writing anything built from search_earnings_transcripts (quoting discipline, spoken-vs-filed sourcing).

Every row-returning tool (run_sql, get_series, search_earnings_transcripts, search_media_appearances) returns at most 50 rows. When a result comes back "truncated": true there is more data — your move is to aggregate or compute in SQL (a GROUP BY date_trunc(...), a SUM/AVG/rank/ratio) and fetch that, or window a single series with from_year / to_year. There is no "give me everything" option, by design — see Context budget below.

Publishing

Tool	Purpose
share_chart (chart, question?)	Publish a ChartSpec object (owned by your account, editable from the UI). Returns {share_id, share_url}.
share_report (question, report, model?)	Publish a multi-section report ({summary, sections, …}) as a public shared run. Returns the publish result incl. share_url.

Pass the spec/report as the tool argument directly — build the object in your context (or with the Write tool / local Python for large data arrays) and hand it to the tool. A validation failure comes back as a tool error naming the bad field; nothing is published until it validates.

Feedback

Tool	Purpose
send_feedback (message, category?)	Report a problem to the FactIQ team: category is "data_issue" (a value that contradicts the official source, wrong units/scale, duplicated or missing periods), "tool_error" (a tool that errors or returns malformed results), "missing_data" (advertised but empty, or coverage ends too early), or "other". Returns an acknowledgment.

Use it on your own initiative whenever something looks broken — don't wait for the user to complain. Write one short, specific message with the concrete identifiers (schema, dataset_code / series_id, the SQL you ran, expected vs. observed, the official source's value or URL if you have one). Don't include the user's personal details or your conversation. It's one-way — the team reviews every report, but nothing comes back — so file it and continue with the task; never block on it.

Terminal charts — term_chart.py

term_chart.py prints local ANSI/ASCII previews from normal FactIQ chart objects. It never calls FactIQ. For share_chart, build the ChartSpec from data you already fetched, save it to JSON, publish it, then render it:

python3 "{plugin_root}/scripts/term_chart.py" render --spec /tmp/factiq-chart.json --width 80 --charset ascii --color auto

For share_report, save the report object or the full share_report argument object ({"question": "...", "report": {...}}) to JSON, publish it, then render the report's charts:

python3 "{plugin_root}/scripts/term_chart.py" report --report /tmp/factiq-report.json --width 80 --charset ascii --color auto

Any time you create a shared chart or shared report, return both the share link and the terminal preview. After term_chart.py renders, paste the preview verbatim into your reply inside a triple-backtick code block; leaving it only in the tool result hides it behind a collapsed block in Claude Code. Use terminal-only output only when the user explicitly asks for no share link.

Supported terminal renderers:

Renderer	Use when
bar	Categorical comparisons and short ranked lists
line	Time-series trends (one or more series)
table	Fallback for unsupported chart types or dense data

Useful options:

Option	Purpose
--type auto|bar|line|table	Pick the terminal renderer; auto maps from ChartSpec.type
--width 80 / --width auto	Fixed width by default; auto reads the terminal size
--height N	Line-chart plot height
--charset ascii|unicode-block	Strict ASCII or denser Unicode block glyphs
--color auto|always|never	ANSI color control; auto respects TTY, NO_COLOR, and TERM=dumb
--max-charts N	Report previews only: cap the number of rendered charts; 0 means all
--out FILE	Also save the rendered text

Because agents often capture command output instead of streaming it directly to the user's terminal, use --charset ascii --color never for previews you paste into the final answer. Use ANSI color for real terminal stdout or saved .ansi previews.

Local viz — build_viz.py

build_viz.py save … / assemble … / render … saves raw tool results to disk (no retyping), builds, and screenshots a bespoke local HTML viz (see Bespoke local visualizations). Local-only; never calls the API.

Orchestration workflow

0. Interview before major forks. If the request is broad, vague, or about to become a high-commitment workflow — especially a detailed report, multi-panel dashboard, bespoke visualization, or a report that could follow multiple scopes — use an explorer-agent interview before fetching data or spawning research subagents. Read references/report-patterns/interview-step.md and ask only the few choices that would materially change the work: detail level, audience, user context or hypothesis, priority lens, required/excluded entities, and time window. Pass the answers into all downstream research and assembler prompts as hard context. Skip the interview for direct answers, narrow quick charts, or when the user already gave clear scope, audience, and detail level. If the user does not answer, proceed with the defaults in the interview guide.

1. Catalog first. Call get_data_catalog once to get the compact per-schema index and the table DDL. It tells you what each schema covers, not every dataset. Skip schemas under schemas_without_data. (You rarely need full=true; use describe_dataset for detail on one dataset.)

2. Find datasets, then drill in. Call search_datasets to rank datasets across all schemas by keyword — the primary discovery step. Survey every schema that could be relevant before committing: for India check both mospi and rbi; for the US check bls, bea, census; energy means eia. Once a dataset looks right, describe_dataset for its dimensions and example series, then find the exact series with search_series (substring — prefer short stems like rare, not rare earth) or exploration SQL (run_sql with explore=true) on the series and dimensions tables. For multi-source stories, actually fetch data from 2+ schemas. Satellite-derived series live in two schemas: portwatch (daily shipping — chokepoints, ports, country trade estimates) and satellite (nighttime lights by state, lake/reservoir water levels) — see references/data/schemas.md for routing and references/data/satellite.md for the on-demand geo tool.

   Eurostat Comext is the exception: country schemas contain millions of series, so do not explore their series or dimensions tables by text or dimension value. Read the Eurostat Comext country schemas section in references/data/sql-guide.md; it uses the small product lookup table and exact indexed series IDs.

   Domain report patterns. If the question is broad and analytical — policy, trade, revenue, investment analysis, "what's driving X" — read references/report-patterns/README.md before fetching. It teaches the dialectical method every report follows (thesis: the headline reading; antithesis: the strongest contradiction, fetched, not footnoted; synthesis: one claim that explains both) and routes covered domains (bilateral trade, bilateral economic policy, monetary policy, fiscal-policy revenue, business formation) to a playbook of that domain's canonical antitheses with ready SQL. For domains without a playbook, apply the method directly. Either way it changes what you fetch, not just how you write it up. The interview step does not replace this method: it sets the user's preferred scope and audience first, then the report-pattern method determines the thesis, antithesis, synthesis, and data work inside that scope.

   For report-mode questions covering multiple topics, companies, or data sources, consider decomposing the research into parallel subagents — see Subagent orchestration below.

3. Fetch in batches. Once you know which series you need, issue the fetch calls together (multiple tool calls in one turn). Use get_series for 1–2 known ids; run_sql with a CASE-WHEN pivot for 3+ series or joins. Keep results inside the 50-row cap — aggregate in SQL to the granularity a chart actually needs. For report tables, choose row granularity by context: monthly rows can work for shorter multi-year windows, roughly up to 3-5 years, when timing, seasonality, or turning points matter; for longer windows, especially 5+ years, usually summarize with annual totals, YTD comparisons, latest/prior snapshots, or selected turning points. Do not default categorically to monthly or yearly rows.

4. Compute yourself. YoY growth, rebasing to an index, per-capita, ratios — write your own Python locally on the fetched values. There is no server-side code interpreter in this loop.

5. Recent market data. The DB lags for very recent market/price data — use get_market_data for current quotes, commodities, and FX. For what the news is saying about a company, sector, or economy right now, use search_news — each business article carries keywords, a geography, and a one-sentence investor angle, plus the tickers it names; chase a company story with get_market_data or search_earnings_transcripts, and a macro story with search_series/run_sql.

6. Satellite signals. For crop burning, wildfires, air-quality-based activity (NO2/SO2/CO), smoke and dust plumes, crop condition (NDVI), monsoon rainfall, heatwaves, or agricultural drought — where satellite observation runs ahead of official statistics — use get_geo_data. Read references/data/satellite.md first: it covers the nine datasets, region syntax, the 50-interval window budget (seasonal comparisons = one call per season), the fires-only grid mode for mapping a fire's footprint, cloud-cover caveats, and required attribution. Satellite data complements warehouse series; prefer curated series where both exist.

7. Answer, publish, render, or build. Direct-answer mode: once you have the value, reply with a single sentence stating the number, its period, and the source — no ChartSpec, no share_chart, no terminal render. Quick-chart mode: build a ChartSpec object (see references/output/chart-spec.md) with wide-format data rows, save it to JSON, call share_chart, then run term_chart.py render; return the share_url and paste the terminal preview into your reply inside a triple-backtick code block. Terminal-chart-only mode: build the same ChartSpec, save it to JSON, run term_chart.py render, and paste the terminal output into your reply without publishing only if the user explicitly requested no share link. Report mode: build a report object (see references/output/report-spec.md and Detailed reports below), save it to JSON, call share_report, then run term_chart.py report; return the share_url and paste the terminal previews into your reply inside a triple-backtick code block. If a publish validation error occurs, fix and republish before rendering the final terminal preview from the corrected object. Bespoke-viz mode: save each fetched result to a JSON file with build_viz.py save (no retyping), author an HTML file, build_viz.py assemble, build_viz.py render to screenshot and iterate, then give the user the local file path (see Bespoke local visualizations).

Subagent orchestration

For report-mode questions that span multiple distinct topics, companies, or data sources, decompose the work into parallel subagents. This does two things: each research thread gets a full, focused context instead of competing for attention in one serial pass, and the report-assembly step gets the spec loaded directly in its prompt so it never guesses at field names.

Before spawning subagents for a broad or underspecified request, run the explorer-agent interview described in references/report-patterns/interview-step.md unless the user already gave clear scope, detail level, audience, and priority lens. Include the interview answers in every research-agent prompt and in the report-assembler prompt so the final artifact reflects the user's context instead of only the generic version of the question.

Explorer interview subagent

Use an explorer agent for the interview step, not a research subagent. Its job is to clarify the decision, audience, scope, output shape, and success criteria and return a compact brief. It should not fetch data, choose final chart schemas, or publish anything. Research subagents run only after the brief and the relevant report pattern are known.

Do NOT use subagents for quick-chart mode or single-topic questions — the overhead is not worth it. The decision point is right after step 2 of the orchestration workflow: once you have done the catalog lookup and initial dataset discovery, you know whether the question decomposes into 2+ independent research threads. If it does, fan out.

Research subagents

Spawn one Agent call per research thread. Each agent inherits the skill's FactIQ MCP tools, so it can discover, fetch, and compute on its own. Give each agent a tightly scoped prompt and tell it to return structured findings — not prose, not a published artifact.

Agent prompt template (adapt the specifics per thread):

You are a FactIQ research agent. Your job is to answer ONE sub-question and
return structured findings. Do NOT call share_chart or share_report — just
research and return data.

Sub-question: {sub_question}

Relevant schemas/datasets (from the parent's catalog step): {hints}

Steps:
1. search_datasets / describe_dataset / search_series to find the right series.
2. Fetch data with get_series or run_sql. Aggregate in SQL to stay under the
   50-row cap.
3. Save each fetched result to its own JSON file so a later charting step can
   load the exact numbers instead of re-querying FactIQ or retyping them.
   Right after each fetch, run (no retyping — it copies the payload from the
   transcript), giving each file a thread-unique name and a --match on a
   distinctive bit of your own SQL so a sibling agent's result can't be grabbed:
   `python3 {plugin_root}/scripts/build_viz.py save --tool run_sql --match "<distinctive SQL fragment>" --out /tmp/factiq-raw/{thread_label}-<name>.json`
4. Compute derived metrics (YoY, ratios, indices) yourself.
5. Return your findings as a structured block:

FINDINGS:
- sub_question: (echo it back)
- series_used: [{schema, series_id, title}, ...]
- sql_queries: [the exact SQL you ran, formatted multi-line]
- data: [{columns: [...], rows: [...]}, ...] — the actual fetched/computed values
- raw_data_files: [/tmp/factiq-raw/{thread_label}-*.json, ...] — the files you
  saved in step 3, so a downstream viz/report step loads exact data
- key_insights: [1-3 sentences stating what the data shows, with numbers]
- chart_suggestion: {chart_type, title, x_column, y_columns, units}

Launch the agents in parallel — multiple Agent tool calls in one turn:

Agent(prompt="<research prompt for thread 1>", label="research-supply-chain")
Agent(prompt="<research prompt for thread 2>", label="research-pricing")
Agent(prompt="<research prompt for thread 3>", label="research-demand")

Each agent runs independently and returns its findings block. Wait for all of them before proceeding to assembly.

Report assembler subagent

After all research is complete, spawn a single report-assembler agent. Its prompt must contain two things: (1) the full content of references/output/report-spec.md so the spec is in context, not behind a file read that might be skipped, and (2) all the research findings from the previous step.

Before spawning the assembler, read references/output/report-spec.md yourself with the Read tool. Then embed its entire content in the assembler's prompt.

Agent prompt template:

You are a FactIQ report assembler. Build a complete report object, publish it
with share_report, and render terminal previews for its charts. Do NOT do any
data discovery or fetching — all data is provided below.

USER QUESTION: {original_question}

=== REPORT SPEC (from references/output/report-spec.md) ===
{paste the full content of references/output/report-spec.md here}
=== END REPORT SPEC ===

=== RESEARCH FINDINGS ===
{paste all findings blocks from the research agents, labeled by thread}
=== END RESEARCH FINDINGS ===

Instructions:
1. Design 2-5 sections. Each section makes one claim its chart(s) prove.
2. Chart titles state the finding with numbers, not the topic.
3. Narratives are plain text — no markdown formatting.
4. Every chart must have columns, data (from the findings above), x_column,
   y_columns (for line/bar), sources, and lineage.
5. Lineage code must be formatted multi-line SQL/Python with real newlines.
   series_refs must list every series the step used.
6. Call share_report with question, report, and model. After it succeeds, save
   the full share_report argument object to JSON and run:
   `python3 {plugin_root}/scripts/term_chart.py report --report <json-file> --charset ascii --color never`
7. Return the share_url and paste the terminal previews into the reply inside a
   triple-backtick code block.

Launch the assembler:

Agent(prompt="<assembler prompt with spec + findings>", label="report-assembler")

The assembler has the full spec in context, so it builds the report object correctly on the first attempt. It calls share_report itself and returns the share_url plus terminal previews, which you relay to the user by pasting the previews inside a triple-backtick code block.

Example decomposition

Question: "How is the US EV market evolving — supply chain, pricing, and demand?"

After step 2 (catalog + discovery), you identify three independent threads:

Thread	Sub-question	Schemas
Supply chain	What does US EV battery/component production look like?	census, bea
Pricing	How have EV prices and average selling prices changed?	bls (CPI), market data
Consumer demand	What are EV sales and registration trends?	bts, bea, market data

Spawn three research agents in parallel. When all return, spawn one assembler agent with the spec and all three findings blocks. The assembler builds a 3-section report (one per thread), calls share_report, renders terminal previews, and returns both.

When NOT to use subagents

• Quick-chart mode (single metric, single chart).
• Single-topic questions even in report mode ("How has US unemployment evolved since 2020?" — one thread, no decomposition needed).
• Bespoke local visualizations — the build_viz loop is inherently iterative and does not benefit from fan-out.

For these cases, do the research and publishing in the main context as usual.

Detailed reports

A report is a public, fully rendered FactIQ research page: a bulleted summary up top, then sections that pair narrative with charts, then methodology notes. You author the whole thing — every chart's data rows, every narrative claim — from data you actually fetched in this session. The JSON format, per-chart fields, and a worked example live in references/output/report-spec.md. For reliable publishing, use a dedicated report-assembler subagent with the spec loaded in its prompt — see Subagent orchestration.

Ground rules:

• 2–5 sections, 1–2 charts each is the sweet spot (server caps: 12 sections, 16 charts). Each section should make one claim its charts prove.
• Chart titles state the finding ("Health care added 652k jobs in 2024 — triple tech's losses"), not the topic ("Jobs by sector").
• Narratives are plain text — markdown is not rendered on the report page, so **bold** shows up as literal asterisks.
• Cite sources and lineage. Every chart should carry sources (the datasets behind it) and lineage (the SQL/computation steps you actually ran). Charts without lineage get a generic "uploaded data" stub — fine, but real lineage makes the "How we built this" panel meaningful. Lineage code renders verbatim in a code block, so write it as formatted multi-line SQL/Python — never collapsed onto one line — and list every series the step touched in series_refs, not a single representative one.
• Don't pad. If the data only supports one chart, publish a quick chart instead of inflating a report.
• Broad analytical questions get the dialectic. Follow the thesis → antithesis → synthesis method in references/report-patterns/README.md: sections that only restate the headline reading are an unfinished report. Covered domains (bilateral trade, bilateral economic policy, monetary policy, fiscal-policy revenue, business formation) must additionally meet the required coverage in the playbook the README routes to — do not reduce them to the easiest single chart.

The share_report tool validates the report against FactIQ's real chart schemas server-side, stores it as a completed public run, and returns the share_url. After it succeeds, render the report object with term_chart.py report, then paste the previews into your reply so the user gets both the link and visible terminal previews. The report appears in your FactIQ history and can be forked by anyone who opens the share link.

Bespoke local visualizations

When the answer wants something the published ChartSpec can't express — a custom layout, a dashboard of several panels, a force/flow/chord diagram, an annotated narrative, a novel encoding, or just fine visual control — build it yourself as a self-contained local HTML file. There is no spec and no fixed chart-type list: you author the HTML/JS (ECharts, D3, Canvas, SVG, WebGL), inject the data you already fetched, then render and iterate. Read references/output/viz-guide.md before starting — it covers technique selection, the data contract, and the legibility checklist.

The tool is {plugin_root}/scripts/build_viz.py (local-only — it never calls the API):

Command	Purpose
save --out F.json [--tool run_sql] [--match STR] [--index N] [--list]	Copy a tool result's raw JSON from the harness transcript to F.json — the shell copies the bytes, you never retype the data. Feeds assemble --data. Stdlib only.
assemble --template T.html --data k1=f1.json k2=f2.json … --out O.html [--open]	Inject on-disk JSON into your HTML at the __FACTIQ_DATA__ marker; write one portable, self-contained file. Stdlib only. List all key=path pairs after the one --data flag.
render O.html [--out P.png] [--width N] [--height N] [--full-page] [--selector CSS] [--wait MS]	Screenshot the file in headless Chromium and report JS/console errors + failed asset loads. Installs Playwright + Chromium into ~/.factiq/viz-venv on first run (uses uv if available, else a stdlib venv).

The loop that makes this work — fetch → save → author → assemble → render → look → fix:

1. Fetch the data with the MCP tools, then save each result to a JSON file with build_viz.py save — do not retype it via Write. The MCP tools return their payload into your context, not to disk; save lifts that exact payload back out of the harness transcript so the shell copies the bytes (never re-emit a ~100-row result by hand — it double-pays the tokens and one mistyped digit ships a wrong chart with no error). Run one save per fetch, pinning the call with --tool/--match:

   python3 "{plugin_root}/scripts/build_viz.py" save --match "korea_customs" --out /tmp/korea.json
   

   The file holds the tool's own {columns, results, …} payload — see references/output/viz-guide.md (Saving data without retyping) for --list, --index, and the fallback when a transcript can't be found. Because the MCP caps results at 50 rows, this is context-cheap; aggregate or window in SQL to get exactly the rows the viz needs.
2. Copy assets/viz-shell.html, add any CDN library you need, and author the viz. Keep the __FACTIQ_DATA__ marker inside its <script id="factiq-data" type="application/json"> tag — that exact element is where the data lands and how the page reads it back. After assembly the page exposes a DATA global; rows are at DATA.<key>.results.
3. assemble the self-contained file, then render it and actually read the screenshot. render exits 5 when the page logged a JS error or a failed request — that usually means a blank page; fix it before judging the visual. One render pass is never enough; budget two or three.
4. Hand the user the local file path; offer --open to open it in a browser.

If the viz will instead be published as a claude.ai Artifact that calls FactIQ live from the page (window.claude.mcp), read references/output/viz-guide.md (Publishing as a claude.ai Artifact with live data) first. In short: declare the capability as factiq (the default connector name) and publish without asking; when you deliver the link, tell the user that if the page can't find their connector they should send you its exact name from claude.ai Settings → Connectors so you can republish with it. In the page's own JS, discover the callable server at runtime with listTools() rather than hardcoding a name.

Context budget — the 50-row cap

Every row-returning MCP tool (run_sql, get_series) returns at most 50 rows, and there is no "give me everything" option — by design. The cap keeps results context-sized, so you do not stage data to disk to protect your context; you take the tool result directly.

When a result comes back "truncated": true, there is more data and your move is to aggregate or compute it in SQL, not to try to fetch the raw rows:

• Roll a long daily/monthly series up with GROUP BY date_trunc('month', time) (or quarter/year) — a chart wants a few hundred points at most, and 50 aggregated points usually says everything.
• Return a SUM / AVG / rank / ratio instead of the underlying rows.
• For one series, window it with get_series(..., from_year=, to_year=), or make a few windowed calls and stitch them.

Whatever you chart or report has to be the aggregated result you bring back — which is also all it needs. For build_viz, persist that (already small) result to a JSON file with build_viz.py save before assembling — it copies the payload from the transcript so you never retype the rows.

Errors and limits

• MCP tool unavailable / auth error — the FactIQ MCP isn't connected. Tell the user to authorize it (Claude Code: /mcp → factiq; Codex: codex mcp login factiq). The same connection authorizes both the data tools and share_chart / share_report, so this fixes publishing failures too.
• 429 — either the 1 request/second rate limit or the monthly tool-call quota (the error says when it resets). Note that publishing counts against the same monthly tool quota as the data tools. Don't burn calls re-fetching data you already have.
• 403 — that schema is admin-restricted for this account; drop it.
• SQL errors come back in the tool result as an error (syntax errors, timeouts, bad column names). Revise the query and rerun.
• Zero rows — your filter was too narrow. Broaden it yourself (see references/data/sql-guide.md). auto_retry=true opts into a server-side LLM reviser, but you can usually revise better and cheaper yourself.
• SQL timeout — statements are capped at 30s. Filter on indexed columns (series_id, dataset_code) instead of scanning titles, and never pattern-match series_id on data_points — resolve ids from series first (see the pitfall in references/data/sql-guide.md). For eu_comext_*, do not retry a dimension scan; use eu_comext_lookup.product_codes and exact IDs as described in the Comext section of that guide.
• Publishing validation error — share_chart / share_report validate the payload against FactIQ's real chart schemas and return a tool error naming the failing field paths (e.g. sections[1].charts[0].x_column). Fix the named fields and call the tool again; nothing is published until it validates.
• Anything that looks broken on FactIQ's side — a value that contradicts the official source, wrong units, missing periods, an advertised dataset that returns nothing, a tool that keeps erroring — report it with send_feedback (see Feedback above), then work around it and continue. The FactIQ team reviews every report and fixes what it can.

References

references/data/ — the data layer:

• schemas.md — what lives in each schema. The get_data_catalog tool is the live, authoritative version; search_datasets / describe_dataset drill into individual datasets on demand.
• sql-guide.md — table structure, query idioms, pitfalls (frequency literals, national vs sub-national, pivots, tabular data).
• satellite.md — the get_geo_data satellite tool: datasets and their economic reading, region syntax and coverage, window budgeting, the fires-only grid footprint mode, cloud/quality caveats, attribution requirements.

references/output/ — the publishing formats:

• chart-spec.md — ChartSpec format, chart-type selection, a worked share_chart example.
• report-spec.md — report JSON format for share_report: sections, per-chart fields, sources/lineage authoring, limits, a worked example.
• viz-guide.md — bespoke local HTML visualizations with build_viz.py: the assemble/render loop, the DATA contract, technique selection (ECharts/D3/Canvas/WebGL), a legibility checklist, starter recipes.

references/report-patterns/ — how to think about broad analytical questions. Start at report-patterns/interview-step.md when the request is vague or high-commitment; it defines the explorer-agent interview that clarifies scope and audience before data work. Then read report-patterns/README.md: it teaches the dialectical method (thesis → antithesis → synthesis) that every report follows and routes covered domains (bilateral trade, bilateral economic policy, monetary policy, fiscal-policy revenue, business formation, and any added later) to a playbook of that domain's canonical antitheses with ready SQL. For uncovered domains — investment analysis, general macro — the README shows how to apply the method directly.