feat: query JSON/recipes, golden & benchmark tooling, docs hub

CLI (codemap query)
- Add --json (JSON array; {"error"} on failure) and --recipe <id> with bundled SQL.
- Wire query-recipes, bootstrap/main/cmd-query, index-engine error handling.

Golden query regression
- scripts/query-golden.ts + Zod schema; --corpus minimal|external, --update, --strict-budget.
- fixtures/golden scenarios + minimal goldens; external example + gitignore.

Benchmark
- Split benchmark into benchmark-common/defaults/config; CODEMAP_BENCHMARK_CONFIG JSON.
- fixtures/benchmark/scenarios.example.json; validate CODEMAP_ROOT on benchmark.

QA & tooling
- scripts/qa-external-repo.ts; fixtures/qa prompts template; package.json scripts.

Docs & agents
- docs/README.md hub: single-source rows for golden, benchmark config, qa:external.
- Root README, benchmark.md, architecture, plan-query-golden, CONTRIBUTING, templates.

Tests & CI
- cli/cmd-query tests; test:golden in check; changesets for query/json/recipe/golden.

Author: SutuSebastian

.agents/rules/codemap.mdc

@@ -16,6 +16,8 @@ A local database (default **`.codemap.db`**) indexes structure: symbols, imports
 | ------- | ----------------- | ----- |
 | **Default** — from this clone | `bun src/index.ts` | `bun src/index.ts query "<SQL>"` |
 | Same entry | `bun run dev` | (same as first row) |
+| JSON output | — | `bun src/index.ts query --json "<SQL>"` |
+| Recipe | — | `bun src/index.ts query --recipe fan-out` (see **`bun src/index.ts query --help`**) |
 
 After **`bun run build`**, **`node dist/index.mjs`** matches the published **`codemap`** binary (same flags). **`bun link`** / global **`codemap`** also work when testing the packaged CLI.
 
@@ -66,6 +68,10 @@ If the question looks like any of these → use the index:
 bun src/index.ts query "<SQL>"
 ```
 
+**Row count:** The CLI does **not** impose a maximum number of rows. Add **`LIMIT`** (and **`ORDER BY`**) in SQL when you need a bounded list. For automation and multi-row answers, prefer **`bun src/index.ts query --json`** — stdout is a JSON array; on failure, stdout is **`{"error":"<message>"}`** and the process exits **1**.
+
+**Verbatim answers:** When the user asks for lists, counts, or enumerated structural data from the index, **paste or summarize from the query output without inventing rows** — do not substitute a prose “summary” that omits rows the user asked to see. Prefer **`--json`** so the full result set is unambiguous.
+
 ## Quick reference queries
 
 | I need to... | Query |

.agents/skills/codemap/SKILL.md

@@ -16,6 +16,44 @@ bun src/index.ts query "<SQL>"
 
 After **`bun run build`**, **`node dist/index.mjs query …`** or a linked **`codemap`** binary matches the published CLI. Use **`--root`** / **`CODEMAP_ROOT`** to index another tree.
 
+## Query output and agents
+
+- **`bun src/index.ts query --json`** prints a **JSON array** of row objects to stdout. On SQL error, stdout is **`{"error":"<message>"}`** and the process exits **1**.
+- The CLI **does not cap** how many rows SQLite returns — add **`LIMIT`** and **`ORDER BY`** in SQL when you need a bounded list.
+- When answering with structural facts from the index (lists of paths, symbols, dependency edges), **ground the answer in the query rows** — do not invent or silently drop rows. Prefer **`--json`** for large or multi-column results.
+
+## Agent-friendly SQL recipes
+
+Replace placeholders (`'...'`) with your module path, file glob, or symbol name.
+
+**CLI shortcuts:** **`bun src/index.ts query --recipe <id>`** runs bundled SQL (optional **`--json`**). Ids include **`fan-out`**, **`fan-out-sample`** (**`GROUP_CONCAT`** samples), **`fan-out-sample-json`** (same, but **`json_group_array`** — needs SQLite JSON1), **`fan-in`**, **`index-summary`**, **`files-largest`**, **`components-by-hooks`**, **`markers-by-kind`** — see **`bun src/index.ts query --help`**. The fan-out rows match the SQL below; others align with “Conditional aggregation”, “Codebase statistics”, and component sections later in this skill.
+
+**Top files by dependency fan-out:**
+
+```sql
+SELECT from_path, COUNT(*) AS deps
+FROM dependencies
+GROUP BY from_path
+ORDER BY deps DESC
+LIMIT 10;
+```
+
+**Same ranking, plus up to five sample targets per file** (uses a correlated subquery; adjust **`LIMIT 5`** as needed):
+
+```sql
+SELECT d.from_path,
+  COUNT(*) AS deps,
+  (SELECT GROUP_CONCAT(to_path, ' | ')
+   FROM (SELECT to_path FROM dependencies d2 WHERE d2.from_path = d.from_path LIMIT 5))
+    AS sample_targets
+FROM dependencies d
+GROUP BY d.from_path
+ORDER BY deps DESC
+LIMIT 10;
+```
+
+**JSON array samples (JSON1):** replace **`GROUP_CONCAT`** with **`json_group_array(to_path)`** in that subquery if your SQLite build has JSON1 — or use **`bun src/index.ts query --recipe fan-out-sample-json`**.
+
 ## Schema
 
 ### `files` — Every indexed file
@@ -196,6 +234,8 @@ SELECT name, file_path, hooks_used
 FROM components WHERE hooks_used LIKE '%useTheme%';
 
 -- Components with most hooks (complexity indicator)
+-- `json_array_length` requires SQLite JSON1. For a portable ranking, use
+-- `bun src/index.ts query --recipe components-by-hooks` (comma-based count on the stored JSON array).
 SELECT name, file_path,
   json_array_length(hooks_used) as hook_count
 FROM components ORDER BY hook_count DESC LIMIT 15;

.changeset/codemap-query-golden.md

@@ -0,0 +1,5 @@
+---
+"@stainless-code/codemap": patch
+---
+
+Add **`bun run test:golden`**: index **`fixtures/minimal`**, run scenarios from **`fixtures/golden/scenarios.json`**, and compare query JSON to **`fixtures/golden/minimal/`**. Use **`bun scripts/query-golden.ts --update`** after intentional fixture or schema changes. Documented in **benchmark.md** and **CONTRIBUTING**.

.changeset/codemap-query-json.md

@@ -0,0 +1,7 @@
+---
+"@stainless-code/codemap": patch
+---
+
+Add **`codemap query --json`**: print a JSON array of result rows to stdout (and **`{"error":"…"}`** on SQL errors) for agents and automation, matching the ergonomics of other CLIs that expose structured output.
+
+Document that the query subcommand does **not** cap rows — use SQL **`LIMIT`** for bounded results. Update bundled agent rule and skill with **`--json`** preference, verbatim structural answers, and generic SQL recipes (fan-out + sample targets).

.changeset/codemap-query-recipe.md

@@ -0,0 +1,9 @@
+---
+"@stainless-code/codemap": patch
+---
+
+Add **`codemap query --recipe <id>`** for bundled read-only SQL so agents and scripts can run common structural queries without embedding SQL on the command line. **`--json`** works with recipes the same way as ad-hoc SQL.
+
+Bundled ids include dependency **`fan-out`** / **`fan-out-sample`** / **`fan-out-sample-json`** (JSON1 **`json_group_array`**) / **`fan-in`**, index **`index-summary`**, **`files-largest`**, React **`components-by-hooks`** (comma-based hook count, no JSON1), and **`markers-by-kind`**.
+
+Benchmark scenario 8 uses the **`fan-out`** recipe SQL for the indexed path; docs clarify that recipes add no extra query cost vs pasting the same SQL.

.github/CONTRIBUTING.md

@@ -11,7 +11,9 @@ Codemap is in **bootstrap / extraction** phase. Before large PRs, please open an
 bun install   # runs `prepare` → Husky git hooks
 bun run dev   # same as `bun src/index.ts` — CLI from source
 bun test
-bun run check   # format + lint + tests + typecheck + build
+bun run test:golden   # golden SQL vs fixtures/minimal (also runs at end of `bun run check`)
+bun run test:golden:external   # Tier B: local tree via CODEMAP_ROOT / --root (not in CI)
+bun run check   # format + lint + tests + typecheck + build + test:golden
 bun run clean   # remove untracked/ignored build artifacts (keeps `.env`, `.codemap.db*`)
 bun run check-updates   # interactive dependency updates (`bun update -i --latest`)
 ```
@@ -36,6 +38,8 @@ Then open a PR on GitHub into **`main`**.
 - **Public API** — Anything exported from the package entry (`src/index.ts` → `src/api.ts`, `config.ts`, shared types) should have **JSDoc** that reads well in hovers and in published typings.
 - **Layers** — Keep boundaries clear: [architecture.md](../docs/architecture.md) (`cli` → `application` → infrastructure). Don’t let CLI concerns leak into parsers or the DB layer.
 - **Before you open / update a PR** — `bun run check` (or at least `bun run test` + `bun run typecheck` while iterating).
+- **Golden queries (Tier A)** — If you change `fixtures/minimal/` or schema/query behavior expected by [fixtures/golden/](../fixtures/golden/), run `bun scripts/query-golden.ts --update`, review diffs, and commit updated JSON under `fixtures/golden/minimal/`. Prefer **fixing the indexer** when output changes for the wrong reason; only refresh goldens when the new rows are correct. See [docs/plan-query-golden-scenarios.md](../docs/plan-query-golden-scenarios.md).
+- **Golden queries (Tier B)** — Against a **local** clone, use `bun run test:golden:external` with `CODEMAP_ROOT` / `--root`. Copy [fixtures/golden/scenarios.external.example.json](../fixtures/golden/scenarios.external.example.json) to `scenarios.external.json` if you need custom scenarios; goldens under `fixtures/golden/external/` are gitignored — do not commit snapshots from proprietary trees.
 - **Style** — Match Oxfmt/Oxlint; prefer **straight-line code** and extracted helpers over long nested blocks.
 
 **Editor (VS Code):** [`.vscode/extensions.json`](../.vscode/extensions.json) lists recommended extensions (Bun, Oxc, TypeScript native preview, etc.). [`.vscode/settings.json`](../.vscode/settings.json) enables Oxc format on save and `tsgo`. Formatting and lint rules live in [`.oxfmtrc.json`](../.oxfmtrc.json) and [`.oxlintrc.json`](../.oxlintrc.json) (no framework-specific options beyond defaults).
@@ -46,6 +50,8 @@ Then open a PR on GitHub into **`main`**.
 
 Do **not** add Codemap as a dependency to the bench repo. In **this** repo, copy `.env.example` to `.env` and set **`CODEMAP_TEST_BENCH`** to an **absolute path** to the other clone, then run `bun src/index.ts` as usual. See [docs/benchmark.md § Indexing another project](../docs/benchmark.md#indexing-another-project).
 
+**One-shot QA (index + disk checks + benchmark):** `CODEMAP_ROOT=/absolute/path/to/app bun run qa:external` (or set **`CODEMAP_TEST_BENCH`** in `.env`; optional `--root` overrides). Validates indexed paths exist, spot-checks symbol lines vs files, prints sample SQL rows, then runs `src/benchmark.ts`. Do **not** add external app source into this repository.
+
 Releases: **[@changesets/cli](https://github.com/changesets/changesets)** — run **`bun run changeset`** when your PR should bump the version; see [docs/packaging.md § Releases](../docs/packaging.md#releases).
 
 **Issues:** use [GitHub issue templates](https://github.com/stainless-code/codemap/issues/new/choose) — **Core bug** vs **Adapter proposal** (see `.github/ISSUE_TEMPLATE/`).

.github/workflows/ci.yml

@@ -92,6 +92,9 @@ jobs:
       - name: Run unit tests with coverage
         run: bun run test:coverage
 
+      - name: Golden query regression (fixtures/minimal)
+        run: bun run test:golden
+
   build:
     name: 🧰 Build
     needs: skip-ci

.gitignore

@@ -8,3 +8,11 @@ dist/
 .env.*
 !.env.example
 tsconfig.lint-staged.json
+
+# Tier B golden query outputs (local / private trees — see docs/plan-query-golden-scenarios.md)
+fixtures/golden/external/
+fixtures/golden/scenarios.external.json
+
+# QA chat prompts tied to a private/local index (paths + product names)
+fixtures/qa/*.local.md
+fixtures/benchmark/*.local.json

README.md

@@ -5,7 +5,7 @@
 - **Not** full-text search or grep on arbitrary strings — use those when you need raw file-body search.
 - **Is** a fast, token-efficient way to navigate **structure**: definitions, imports, dependency direction, components, and other extracted facts.
 
-**Documentation:** [docs/README.md](docs/README.md) is the hub (topic index + single-source rules). Topics: [architecture](docs/architecture.md), [agents](docs/agents.md) (`codemap agents init`), [benchmark](docs/benchmark.md), [packaging](docs/packaging.md), [roadmap](docs/roadmap.md), [why Codemap](docs/why-codemap.md). **Bundled rules/skills:** [`.agents/rules/`](.agents/rules/), [`.agents/skills/codemap/SKILL.md`](.agents/skills/codemap/SKILL.md). **Consumers:** [.github/CONTRIBUTING.md](.github/CONTRIBUTING.md).
+**Documentation:** [docs/README.md](docs/README.md) is the hub (topic index + single-source rules). Topics: [architecture](docs/architecture.md), [agents](docs/agents.md) (`codemap agents init`), [benchmark](docs/benchmark.md), [golden query plan](docs/plan-query-golden-scenarios.md), [packaging](docs/packaging.md), [roadmap](docs/roadmap.md), [why Codemap](docs/why-codemap.md). **Bundled rules/skills:** [`.agents/rules/`](.agents/rules/), [`.agents/skills/codemap/SKILL.md`](.agents/skills/codemap/SKILL.md). **Consumers:** [.github/CONTRIBUTING.md](.github/CONTRIBUTING.md).
 
 ---
 
@@ -37,6 +37,13 @@ codemap --full
 
 # SQL against the index (after at least one index run)
 codemap query "SELECT name, file_path FROM symbols LIMIT 10"
+# JSON array on stdout (agents / scripts); errors: {"error":"..."}
+codemap query --json "SELECT name, file_path FROM symbols LIMIT 10"
+# Query is not row-capped — add LIMIT in SQL for large selects
+# Bundled SQL (same as skill examples): fan-out rankings
+codemap query --recipe fan-out
+codemap query --json --recipe fan-out-sample
+# `components-by-hooks` ranks by hook count without SQLite JSON1 (comma-based count on the stored JSON array).
 
 # Another project
 codemap --root /path/to/repo --full
@@ -81,12 +88,15 @@ const rows = cm.query("SELECT name FROM symbols LIMIT 5");
 
 Tooling: **Oxfmt**, **Oxlint**, **tsgo** (`@typescript/native-preview`).
 
-| Command                              | Purpose                                                          |
-| ------------------------------------ | ---------------------------------------------------------------- |
-| `bun run dev`                        | Run the CLI from source (same as `bun src/index.ts`)             |
-| `bun run check`                      | Build, format check, lint, tests, typecheck — run before pushing |
-| `bun run fix`                        | Apply lint fixes, then format                                    |
-| `bun run test` / `bun run typecheck` | Focused checks                                                   |
+| Command                              | Purpose                                                                    |
+| ------------------------------------ | -------------------------------------------------------------------------- |
+| `bun run dev`                        | Run the CLI from source (same as `bun src/index.ts`)                       |
+| `bun run check`                      | Build, format check, lint, tests, typecheck — run before pushing           |
+| `bun run fix`                        | Apply lint fixes, then format                                              |
+| `bun run test` / `bun run typecheck` | Focused checks                                                             |
+| `bun run test:golden`                | SQL snapshot regression on `fixtures/minimal` (included in `check`)        |
+| `bun run test:golden:external`       | Tier B: local tree via `CODEMAP_*` / `--root` (not in default `check`)     |
+| `bun run qa:external`                | Index + sanity checks + benchmark on `CODEMAP_ROOT` / `CODEMAP_TEST_BENCH` |
 
 ```bash
 bun install
@@ -100,11 +110,13 @@ bun run fix      # oxlint --fix, then oxfmt
 
 ## Benchmark
 
+Use a **real** project path (the repo must exist on disk). See [docs/benchmark.md § Indexing another project](docs/benchmark.md#indexing-another-project).
+
 ```bash
-CODEMAP_ROOT=/path/to/indexed-repo bun src/benchmark.ts
+CODEMAP_ROOT=/absolute/path/to/indexed-repo bun src/benchmark.ts
 ```
 
-Details: [docs/benchmark.md](docs/benchmark.md).
+Optional **`CODEMAP_BENCHMARK_CONFIG`** for repo-specific scenarios: [docs/benchmark.md § Custom scenarios](docs/benchmark.md#custom-scenarios-codemap_benchmark_config).
 
 ---