Build Kb

Build Knowledge Base

You are building the project's vector-indexed knowledge base — semantic_models, metrics, and reference_sql (the LanceDB-backed stores) — from the project's files and database metadata. This is the heavy companion to the lightweight /init: /init already produces the AGENTS.md inventory and the file-based stores (./knowledge/*.md, memory); this skill owns the expensive generation that writes the vector stores and then refreshes AGENTS.md's KB index sections.

This is an orchestration skill running in the main agent context, so you may call task, todo_write/todo_list/todo_read/todo_update, ask_user, add_memory/edit_memory, the filesystem tools (glob, grep, read_file, write_file, edit_file), and the database tools (list_databases, list_tables, describe_table, search_table, execute_sql).

Routing authority is storage-classify. This skill decides what to scan and explore within scope; the storage-classify skill owns which content goes to which store, written with which mechanism. Do NOT re-invent storage routing rules here — load storage-classify in Step 0 and follow it through Steps 2-4 (Step 2 inlines its rules into explore prompts; Step 3 routes every item by them).

Two-phase contract (important). Heavy generation (semantic_modeling / gen_sql_summary / gen_skill / extract-knowledge) costs tokens and writes real artifacts. So there is a hard turn boundary after exploration: you produce a Generation Manifest and then end your turn. Do NOT call ask_user for this confirmation — just present the manifest and stop. The user confirms or corrects it in their next message; only then do you run Step 3 and Step 4.

Exception — auto-run (confirmation skip). When the user explicitly waived confirmation in Step 0 (auto-run = true), there is no turn boundary: still assemble and print the full manifest (so the choices stay on the record and the dual-route knowledge rows stay visible), but instead of stopping, continue straight into Step 3 in the same turn. This is the only way the gate is skipped — never infer it from a bare scope hint or from impatience; default is always to confirm.

Step 0 — Resolve Scope (inferred, no questions)

Do NOT call ask_user in this step. The user invokes this skill as /build-kb <free-text scope hints> — the hints arrive as an "Additional context from the user" block. Parse them into a concrete scope; when no hints are given, default to the whole project.

1. Parse the scope hints into any of: in-scope files (globs / paths, e.g. queries/*.sql), datasources, tables (e.g. orders, order_items), and business domains (e.g. "only the sales domain"). Free text — interpret generously, e.g. /build-kb orders + order_items tables and queries/*.sql, sales domain only.
   • Also detect a confirmation-skip opt-out. If the current invocation's hints explicitly waive the manifest confirmation gate (e.g. "skip confirmation", "no confirm", "auto", "don't stop for confirmation", "直接执行" / "跳过确认" / "不用确认"), set auto-run = true. Absent such an explicit signal, auto-run = false (the default — always confirm). A bare scope hint is NOT an opt-out; only an explicit waive counts.
2. Infer the goal and datasource defaults the same way /init does: read README.md (first 3000 chars) or derive a 1-2 sentence goal. For datasources, default to the currently active datasource of the session (the one pinned via --datasource / /datasource, i.e. the project's default_datasource); when multiple datasources are configured, scope to that active one only — do NOT cover all of them unless the Step 0.1 hints explicitly name other datasources. Hints from Step 0.1 override these defaults.
3. Reuse /init's inventory when present. If ./AGENTS.md exists, read it to reuse the directory map / data-assets inventory rather than re-scanning the whole tree — narrow your scan to the in-scope subset.
4. Use glob to scan the in-scope directory tree (top 3 levels). Skip hidden dirs and __pycache__ / node_modules / .venv.
5. Load routing rules now: call load_skill("storage-classify") and extract its Decision Tree and Per-Store Reference into a local summary. This must happen before Step 2 so the routing rules can be inlined into every explore subagent prompt — explore subagents run in isolated contexts and cannot load skills themselves.

Record the resolved goal, in-scope datasources, the file/table/domain scope (with a one-line reason for each), and the auto-run flag — they become the first rows of the Generation Manifest so the user can correct them at the single confirmation point. If the scope is empty after parsing, state plainly that you are defaulting to the whole project.

Step 1 — Scan & Classify (within scope)

Gather the raw material inside the resolved scope, then classify it into a multi-level taxonomy of business domains / subtopics (e.g. sales/orders, sales/refunds, infra/etl).

File side:

• Use glob / grep to collect candidate files within scope — scan all text files, judged by content rather than extension. Skip binary files (images, executables, archives, compiled artifacts, parquet/db blobs, etc.) and skip oversized files (> ~1 MB) — if a large text file is clearly relevant, read its head/batches rather than the whole thing. When unsure whether a file is text, peek at the first bytes (read_file head) before committing to it.
• Validated-query corpus is special — treat it as enumerable, never as a sample. When the in-scope material holds a corpus of validated (question, SQL) pairs (a queries file, a golden/benchmark set, a saved-query catalog, a dbt/analysis SQL folder), count the total and plan to index every pair. Unlike tables (where representative sampling is fine), each validated query is a future few-shot example: one omitted pair is one the runtime can never retrieve. Do not curate "representative" patterns here.
• Enumerate the SQL corpus directly here — do NOT route it through an explore subagent. You are already reading these files' full text in this step, so build each (question, SQL) pair's reference_sql manifest row yourself: one row per pair, with the prompt-seed inlining the original natural-language question + the complete SQL + any mandatory filter. Handing a .sql file to explore only makes that subagent re-read the SQL verbatim and echo it back in its prompt-seed — a full round-trip of the SQL text through a subagent context for zero added insight. Large corpora: read and enumerate in batches rather than truncating.

Database side (for each in-scope datasource):

• list_databases → list_tables to enumerate tables/views; restrict to in-scope tables when the scope names them.
• For representative in-scope tables: describe_table for desc (column names/types/comments), search_table (its sample_data) or execute_sql("SELECT * FROM <t> LIMIT 5") for sample, and execute_sql("SELECT COUNT(*) AS rows, COUNT(DISTINCT <key>) AS card FROM <t>") for key statistics (row count, key-column cardinality). There is no dedicated statistics tool — compute it with execute_sql.
• For large databases (>50 in-scope tables), sample representative tables per naming pattern rather than describing every table.

Classify every in-scope file and table into the domain taxonomy. A single domain may contain both files and tables. Split the classified material into two routes:

• Validated-query SQL corpus → handle here. Enumerate each (question, SQL) pair into a reference_sql manifest row directly (per the bullet above). These rows skip Step 2.
• Tables + documentation → hand to Step 2 explore. Group them under their domain; these are what the explore subagents investigate.

Record with todos only when Step 2 will fan out more than one explore subagent: call todo_write once, with one todo per to-be-explored domain — title = domain name (≤ 8 words), content = the files + tables it covers plus exploration focus notes. Track each domain's exploration with todo_update (pending → in_progress → completed / failed) in Step 2. This is the only todo_write in this skill — Step 3 tracks generation through the confirmed manifest, never through new todos. Skip todos entirely when zero or one domain needs exploring (e.g. the scope is only the SQL corpus) — a single explore call is not worth a sidebar list.

Step 2 — Explore Each Domain in Parallel (concurrency ≤ 3)

Explore only what needs investigating — database tables and documentation. The validated-query SQL corpus was already enumerated into reference_sql rows in Step 1; do NOT launch an explore subagent for .sql files or query catalogs (that just round-trips the SQL text through a subagent for no gain). A domain whose only in-scope material is the SQL corpus needs no explore call at all.

For each domain that has tables or docs to investigate, delegate a read-only exploration to an explore subagent:

task(type="explore", prompt=..., description="explore <domain>")

The explore subagent runs in an isolated context — it sees nothing you gathered unless you inline it. The prompt must carry: the inferred project goal and the datasource (+ dialect) the domain lives in, the domain's file list + table list, the already-gathered desc / sample / statistic summaries, and the full Decision Tree + Per-Store Reference extracted from storage-classify in Step 0. Inlining the routing rules ensures the subagent's store assignments match what Step 3 will actually execute. It must instruct the subagent to explore read-only and summarize, returning a structured result in the storage-classify taxonomy:

subject (the domain)
  → store: one of semantic_models | metrics | reference_sql | knowledge | skills | memory | AGENTS.md | none
    → ref: file path / table name / column name
       rationale: one line — why this store (cite the storage-classify decision-tree branch)
       prompt-seed: the self-contained seed to hand the downstream generator — not just a bare ref but the context it needs (e.g. table names + the column encodings/intent for semantic_modeling; for a SQL example *discovered in a doc*, the full SQL + the business question + any mandatory filter for gen_sql_summary)

Coverage focuses on semantic_models, metrics, and knowledge (plus reference_sql only for SQL newly discovered in docs — the validated-query corpus was already enumerated in Step 1) — /init already wrote the AGENTS.md inventory and the initial knowledge/memory, so here the explorer should surface vector-store candidates plus any additional knowledge atoms the corpus reveals (do not re-propose facts /init already filed; do not propose other stores beyond memory/AGENTS.md notes).

The validated-query corpus is NOT explored here — you enumerated it directly in Step 1. The only reference_sql refs an explorer should return are SQL it newly discovers while investigating tables or docs (e.g. an example query embedded in a Markdown doc). For any such discovered SQL, instruct the explorer to carry the original natural-language question (if the doc states one) in the prompt-seed — it is the best retrieval key for future questions.

Concurrency rule: issue at most 3 task calls per batch (3 tool calls in one message), wait for the batch to return, then start the next batch. If Step 1 created todos, set the domain's todo to in_progress when you launch it and completed when it returns. Tell the user (briefly) how many domains were dropped if you cap anything.

Turn Boundary — Emit the Generation Manifest, then STOP

This manifest is the single user confirmation point — it must lead with the resolved scope (Step 0) so the user can correct the goal or scope here, not via an earlier question.

1. Lead with the resolved scope so it is explicitly confirmable:

   Inferred goal: <1-2 sentences> — (from README.md / directory name / table names) In-scope datasources: — (from hints / agent.yml) Scope: <files / tables / domains, or "whole project (no scope hints given)">

2. Aggregate every explore result together with the reference_sql rows you enumerated directly in Step 1, dedupe refs that appear under multiple domains, and build a Generation Manifest grouped by store, rendered as a Markdown table:

   Subject	Store	Refs	Mechanism	Summary
   sales/orders	semantic_models	orders, order_items	task(semantic_modeling)	core order facts
   sales/orders	metrics	GMV, AOV	task(semantic_modeling)	built on orders measures
   …	reference_sql	queries/top_skus.sql	task(gen_sql_summary)	reusable ranking query
   …	knowledge	status enum on orders	extract-knowledge (lite)	atomic field-encoding fact

   Dual-route rows must be explicit. Every reference_sql row sourced from a (question, SQL) pair that also carries a schema-non-inferable rule must appear a SECOND time as its own knowledge row here — that is the dual-route second write (see Step 3 point 4). Listing it makes its extra extract-knowledge cost visible so the user can strike it; Step 3 must NOT mine knowledge for any pair not listed as a knowledge row.

3. STOP here — unless auto-run was set in Step 0.

   • Default (auto-run = false): after printing the resolved scope + manifest, end your turn. Do NOT call any generation task, extract-knowledge, gen_skill, add_memory, or write any store yet. Do NOT call ask_user. State plainly: "Reply to confirm, or correct the goal / scope / any manifest row, and I'll run the generation." Wait for the user's next message.
   • Auto-run (auto-run = true): do not stop. Print the same scope + manifest, add one line noting confirmation was skipped per the user's instruction (so the manifest is still auditable), then proceed directly into Step 3 in this same turn.

Step 3 — Route & Generate (next turn, after confirmation; concurrency ≤ 3)

Once the user confirms or corrects the manifest (or immediately, in the same turn, when auto-run was set in Step 0):

1. Use the Decision Tree + Per-Store Reference + Context Handoff to Subagents from the storage-classify rules already loaded in Step 0 as the routing authority for every item.
   • Before routing semantic assets, resolve the active semantic adapter. Invoke semantic_modeling only for a Dosi project. MetricFlow and OSI projects are query-only: do not attempt semantic writes; report that the project must be migrated to Dosi before semantic authoring can continue.
2. Make every delegated prompt self-contained (see storage-classify's Context Handoff). The explore subagents that produced these refs are gone, and each generator runs in a fresh context — so inline the datasource (+ dialect), the business intent, the prompt-seed the explorer returned, and the rules/encodings already gathered that the artifact must honor. Route each manifest item to its store with the prescribed mechanism:
   • Light items → write directly: memory via add_memory (≤ 2000 bytes); small AGENTS.md notes via write_file / edit_file.
   • Heavy items → delegate (the placeholders below are the minimum each prompt must carry):
     • semantic_models / metrics → for Dosi only, task(type="semantic_modeling", prompt="<datasource> · table(s) · metric definitions if any · intent · column encodings and mandatory filters>"); combine dependent model and metric rows for the same business domain when practical. For MetricFlow or OSI, skip generation and report the query-only migration requirement.
     • reference_sql → task(type="gen_sql_summary", prompt="<datasource/dialect> · the original natural-language question (if known) · the complete SQL · why it is written this way>") — one call per SQL, enumerate the whole corpus: each query becomes its own reference_sql entry; index every (question, SQL) pair, do not select representatives (recall is driven by coverage). If a manifest row lists several SQLs, expand it into one gen_sql_summary call each; never pass multiple queries in a single prompt (they collapse into one mixed, unsearchable entry). Always pass the original question when the example came from one — it is the retrieval key future questions match against. The prompt must also instruct the generator explicitly: "set search_text to the original natural-language question verbatim (trim whitespace, keep its language); only fall back to keyword phrases when no original question exists" — search_text is the vector key the runtime embeds, and a user's question matches another question far better than it matches SQL keywords.
     • skills → task(type="gen_skill", prompt="<skill intent + the concrete steps observed>")
     • knowledge → run extract-knowledge in lite mode (do NOT trigger its deep blind-SQL flow); pass the source (the SQL/doc/table) and the specific fact to mine, plus the datasource it applies to. Only mine atoms /init did not already file — do not duplicate existing ./knowledge/*.md entries.
3. Grouping: in a Dosi project, send one coherent semantic_modeling request per business domain by default. If the user requests another grouping, follow it. Include dependent datasets, relationships, and metrics in the same request.
   • SQL-backed handoff contract: append the original natural-language question when present and the complete original SQL to the semantic_modeling prompt. Do not pass a rewritten summary query, selected CTE fragments, or an inferred replacement.
   • Do not assume a success-story row has external knowledge beyond its question, SQL, and stored metadata. Missing external knowledge is not a blocker and must not be synthesized.
4. Dual-route every (question, SQL) pair that appears as a knowledge row in the manifest — this is required, not optional. For each such pair: (a) send it to gen_sql_summary so the example (with its original question) lands in reference_sql, AND (b) feed the same pair to extract-knowledge (lite) to mine the non-inferable rule. The example teaches answer shape (retrieved later for few-shot); the mined atom teaches why (encodings, mandatory filters, term→column mappings). One source, two stores — neither replaces the other. Mine knowledge ONLY for pairs listed as knowledge rows in the manifest — whether the user confirmed them, or auto-run carried the manifest through unconfirmed (see Turn Boundary).
5. Concurrency ≤ 3: dispatch heavy task calls in batches of at most 3, waiting for each batch. Do NOT create or update todos in this step — the confirmed manifest is the item-level record; narrate per-batch progress briefly and list any failed items in the closing summary.

Do not hand-write semantic_models / metrics / reference_sql YAML yourself — use semantic_modeling for Dosi semantic assets and the matching reference-SQL owner (per storage-classify's Forbidden rules). Legacy MetricFlow and OSI projects remain query-only until migrated to Dosi.

Step 4 — Refresh the AGENTS.md KB Index

After all generation completes, update ./AGENTS.md last, following the AGENTS.md Section Ownership from storage-classify. Do not rewrite the inventory sections /init owns (# title · ## Architecture · ## Directory Map · ## Services · ## Data Assets · ## Recommended Tools · ## SQL Conventions). Use a scoped edit_file that touches only the KB index:

• ## Semantic Models / ## Metrics / ## Reference SQL — the vector-index sections you just populated. /init does not write these sections, so insert each one (in canonical order) with what it covers + how many + which tool retrieves it (these stores are queried by retrieval, not read as files). Only insert a section if you actually generated content for it — never write a "none yet" placeholder for a store you produced nothing for:
  • ## Semantic Models — N models (schools, satscores, frpm); retrieve with search_semantic_model.
  • ## Metrics — N metrics (county_avg_sat_math, avg_frpm_rate, …); retrieve with search_metrics.
  • ## Reference SQL — N validated queries; retrieve similar (question → SQL) examples with search_reference_sql before writing new SQL.
• ## Knowledge — append only. This section is owned by extract-knowledge and /init already filed the initial atoms. If Step 3 mined additional knowledge, append one bullet per new ./knowledge/*.md (- [<Domain>](knowledge/<slug>.md) — <one-line scope>), writing the scope line to convey unguessable specifics (exact thresholds, literal filter codes, enum spellings, term→column mappings). Never overwrite existing entries.

If ./AGENTS.md does not exist (the user never ran /init), create the full file per the AGENTS.md Section Ownership — the same skeleton /init would have written — then fill the KB index as above. In that fallback only, also induce a ## SQL Conventions section from the validated-SQL corpus (see below).

## SQL Conventions (only when creating AGENTS.md from scratch)

If /init already wrote ## SQL Conventions, leave it untouched. Only when you are creating AGENTS.md from scratch and the project has a validated-SQL corpus, induce its recurring output conventions and write them as a short bullet list. This section rides in <project_context>, so it nudges every downstream gen_sql toward this project's answer shape.

• Induce, do not hardcode. Read a representative slice of the corpus and state only patterns you actually observe, phrased schema-free (no specific table/column/code names).
• Every rule carries its trigger phrasing — write each as "when the question says/asks ⟨observable phrasing⟩ → ⟨output shape⟩". A rule whose trigger you cannot state as question wording is not a rule — drop it.
• Counter-example scan before persisting. For each candidate rule, scan the corpus for pairs whose question matches the trigger but whose SQL has a different shape; any counter-example means narrow or drop the rule.
• If the corpus is absent or too small to induce any convention, omit the ## SQL Conventions section entirely — do not write a placeholder.

Make the KB reachable at runtime

The KB you just built is useless if gen_sql can't reach its retrieval tools. With the default tool-permission behavior, a gen_sql node whose agentic_nodes.gen_sql block omits tools: inherits node defaults (which include context_search_tools.*), so no per-project tools: list is required. Only if the project's agent.yml pins an explicit tools: for gen_sql must it include context_search_tools.*. Mention this in your closing note only when the config visibly restricts tools; otherwise leave config untouched.

Hard constraints:

• AGENTS.md is a top-level overview, not a data dictionary. Target ≤ 200 lines (only the first ~200 lines are injected into <project_context>).
• For the semantic/metric/reference_sql index lines, state the count and the retrieval tool so a downstream agent knows the KB exists and how to consult it — do NOT inline their contents.
• Prefer a scoped edit_file over a full rewrite; ask via ask_user before overwriting an existing file wholesale.

Tell the user the KB is built and AGENTS.md's index is refreshed.

Important Notes

• Routing lives in storage-classify, not here. When in doubt about which store an item belongs to, defer to its decision tree and disambiguation table.
• The explore subagent's subject → store → ref output is exactly storage-classify's input contract — they dovetail.
• Scope is the whole point of this skill vs /init. Honor the resolved scope in every step — never scan, explore, or generate outside it unless the user gave no hints (whole-project default).
• /init owns the file-based stores; this skill owns the vector stores. Do not duplicate the AGENTS.md inventory or re-file knowledge/memory /init already wrote — only add what the heavy generation produces.
• Use placeholder comments when you cannot determine something rather than inventing facts.
• Do not ask the user anything before the manifest. Scope is resolved in Step 0 and confirmed at the turn boundary; the manifest is the single confirmation gate, not an ask_user. (The only later ask_user allowed is the Step 4 guard before wholesale-overwriting an existing AGENTS.md.)