@carllee1983/dbcli 1.48.0 → 1.49.0

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "dbcli-agent",
-  "version": "1.48.0",
+  "version": "1.49.0",
   "description": "Database CLI skill and command reference for AI agents.",
   "author": {
     "name": "Carl Lee",

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "dbcli-agent",
-  "version": "1.48.0",
+  "version": "1.49.0",
   "description": "Database CLI skill and command reference for AI agents",
   "author": {
     "name": "Carl Lee",
@@ -26,11 +26,7 @@
     "longDescription": "Installs the dbcli agent skill and reference guide so Codex can inspect schemas, query databases, respect blacklist boundaries, and recover from database errors through the dbcli command workflow.",
     "developerName": "Carl Lee",
     "category": "Productivity",
-    "capabilities": [
-      "Database",
-      "Local CLI",
-      "Agent Skill"
-    ],
+    "capabilities": ["Database", "Local CLI", "Agent Skill"],
     "websiteURL": "https://github.com/CarlLee1983/dbcli",
     "defaultPrompt": [
       "Inspect my database with dbcli.",

package/.cursor/rules/dbcli.mdc

@@ -363,7 +363,7 @@ Full flags and edge cases: see [reference.md](reference.md) `init` section.
 | `assert` | query-only+ | **(v1.25)** SQL only. Verify an invariant; exit 1 on failure unless `--no-fail`. `--expect "rows>0\|value==X\|col:c not null\|unique\|between a and b\|>= n"`, `--vs <query> --compare rows\|value` (reconcile), `--against <snapshot> --tolerance <pct>`. |
 | `verification` | n/a | Inspect and manage local verification artifacts. `list` / `show <id-or-path>` / `summary` are read-only; `prune` is dry-run by default and deletes only with `--execute --force`. Reads `<cwd>/.dbcli/verification/`; no DB connection, no audit writes. |
 | `backfill artifact` | n/a | Build a bounded, reviewable source-to-SQL backfill artifact from JSON. Includes source/target identity, blacklist/schema preflight, read-back verification, and rollback hints; dry-run only and never executes writes. |
-| `proxy` | n/a | **(v1.26)** MySQL/MariaDB/PostgreSQL only. Local-dev observability proxy — relays app traffic to the real DB and appends query/latency/byte/error events to `.dbcli/proxy/events.jsonl`. Subcommands: `mysql` \| `mariadb` \| `postgresql`. `--listen`, `--target`, `--events`, `--slow-ms` (default `1000`), `--redact none\|literals`. Observe-only. **(v1.27)** `proxy analyze` aggregates the event log offline into a JSON/text report (summary, byFingerprint with suggestedCommands, slowest, errors, hotTables, N+1) — errors out if no events exist yet. |
+| `proxy` | n/a | **(v1.26)** MySQL/MariaDB/PostgreSQL only. Local-dev observability proxy — relays app traffic to the real DB and appends query/latency/byte/error events to `.dbcli/proxy/events.jsonl`. Subcommands: `mysql` \| `mariadb` \| `postgresql`. `--listen`, `--target`, `--events` (default `.dbcli/proxy/events.jsonl`), `--slow-ms` (default `1000`), `--redact none\|literals` (default `none`). Observe-only. **(v1.27)** `proxy analyze` aggregates the event log offline into a JSON/text report (summary, byFingerprint, slowest, errors, hotTables, N+1) — `--format`, `--top`, `--slow-ms`, `--n-plus-one`; errors out if no events exist yet. Actionable blocks carry `suggestedCommands` + `hints` so an agent can act: SELECT hotspots/N+1 → `explain` / `guide missing-index-for`, errors → `schema <table>` (verify names, never guess), N+1 → batch (JOIN / `IN (...)`). After analyzing, run each finding's `suggestedCommands`, read its `hints`, then propose the fix. |
 | `status` | query-only+ | Safe JSON/text summary (no credentials). |
 | `inspect` | query-only+ | Read-only context snapshot (connection, permission, blacklist, objects, snippets, context-aware `suggestedCommands`, and **(v1.23)** human-readable `hints`). `--for-agent` / `--brief` / `--no-connect` / `--require-schema-cache`. Supports `--recovery`. |
 | `report` | query-only+ | Diagnostic report built from `@diag/*` snippets. `--section <health\|capacity\|perf>` (comma-separated to combine), `--brief`, `--for-agent`, `--no-connect`. |

package/.cursor/skills/dbcli/reference.md

@@ -1202,6 +1202,15 @@ dbcli proxy analyze --slow-ms 200 --n-plus-one 5  # custom thresholds
 
 **`proxy analyze`** — offline aggregation of the event log (no DB). Flags: `--events <path>` (default `.dbcli/proxy/events.jsonl`), `--format json|text` (default `json`), `--top <n>` (default 20; text rows + suggestedCommands depth), `--slow-ms <ms>` (default 1000; recomputes slowCount), `--n-plus-one <n>` (default 10), `--no-include-rotated`. JSON report blocks: `summary`, `byFingerprint` (sorted by total time; SELECT entries in the top-N carry `suggestedCommands` for `explain` / `guide missing-index-for`), `slowest`, `errors`, `hotTables`, `repetition` (N+1 suspects). Reads the current log plus the rotated `.1` segment by default.
 
+Every actionable block carries machine-readable next steps so an agent can move from "what is wrong" to "what to run":
+- `byFingerprint[]` — top-N SELECT entries get `suggestedCommands`: `dbcli explain "<sql>"` and `dbcli guide missing-index-for "<sql>"`.
+- `errors[]` — carries `tables`; emits `suggestedCommands` of `dbcli schema <table>` (capped at the first 3 tables) plus a `hints` note to verify table/column names before fixing (never guess column names). No tables known (e.g. a syntax error) → no `suggestedCommands`, but the hint still appears.
+- `repetition[]` — carries `statement` and a runnable `exampleSql` (the slowest occurrence). SELECT N+1 groups get `explain` / `guide missing-index-for` `suggestedCommands`; every group carries a `hints` note suggesting batching (JOIN / `IN (...)`) or caching.
+
+`suggestedCommands` are emitted as strings only — `proxy analyze` never executes them. When the proxy ran with `--redact literals`, `exampleSql` (and therefore the suggested commands) contains `?` placeholders; fill in real values before running them.
+
+**Acting on the report (agent loop):** after `proxy analyze`, for each block read `hints` for the diagnosis, run the entry's `suggestedCommands` to gather schema/plan/index evidence, then propose a concrete fix — add an index (`guide missing-index-for`), rewrite a slow SELECT (`explain`), batch an N+1 (`repetition`), or correct a column/table name (`errors` → `schema`). The text format mirrors this with aggregated `SUGGESTED COMMANDS` and `HINTS` sections; JSON keeps the suggestions attached per-finding.
+
 **Engines:** MySQL / MariaDB / PostgreSQL
 **Permission:** n/a (acts as a TCP relay; does not use dbcli's SQL permission model)
 

package/.cursor-plugin/plugin.json

@@ -2,7 +2,7 @@
   "name": "dbcli-agent",
   "displayName": "dbcli Agent",
   "description": "Database CLI skill and command reference for AI agents.",
-  "version": "1.48.0",
+  "version": "1.49.0",
   "author": {
     "name": "Carl Lee",
     "url": "https://github.com/CarlLee1983"

package/.github/skills/dbcli/SKILL.md

@@ -363,7 +363,7 @@ Full flags and edge cases: see [reference.md](reference.md) `init` section.
 | `assert` | query-only+ | **(v1.25)** SQL only. Verify an invariant; exit 1 on failure unless `--no-fail`. `--expect "rows>0\|value==X\|col:c not null\|unique\|between a and b\|>= n"`, `--vs <query> --compare rows\|value` (reconcile), `--against <snapshot> --tolerance <pct>`. |
 | `verification` | n/a | Inspect and manage local verification artifacts. `list` / `show <id-or-path>` / `summary` are read-only; `prune` is dry-run by default and deletes only with `--execute --force`. Reads `<cwd>/.dbcli/verification/`; no DB connection, no audit writes. |
 | `backfill artifact` | n/a | Build a bounded, reviewable source-to-SQL backfill artifact from JSON. Includes source/target identity, blacklist/schema preflight, read-back verification, and rollback hints; dry-run only and never executes writes. |
-| `proxy` | n/a | **(v1.26)** MySQL/MariaDB/PostgreSQL only. Local-dev observability proxy — relays app traffic to the real DB and appends query/latency/byte/error events to `.dbcli/proxy/events.jsonl`. Subcommands: `mysql` \| `mariadb` \| `postgresql`. `--listen`, `--target`, `--events`, `--slow-ms` (default `1000`), `--redact none\|literals`. Observe-only. **(v1.27)** `proxy analyze` aggregates the event log offline into a JSON/text report (summary, byFingerprint with suggestedCommands, slowest, errors, hotTables, N+1) — errors out if no events exist yet. |
+| `proxy` | n/a | **(v1.26)** MySQL/MariaDB/PostgreSQL only. Local-dev observability proxy — relays app traffic to the real DB and appends query/latency/byte/error events to `.dbcli/proxy/events.jsonl`. Subcommands: `mysql` \| `mariadb` \| `postgresql`. `--listen`, `--target`, `--events` (default `.dbcli/proxy/events.jsonl`), `--slow-ms` (default `1000`), `--redact none\|literals` (default `none`). Observe-only. **(v1.27)** `proxy analyze` aggregates the event log offline into a JSON/text report (summary, byFingerprint, slowest, errors, hotTables, N+1) — `--format`, `--top`, `--slow-ms`, `--n-plus-one`; errors out if no events exist yet. Actionable blocks carry `suggestedCommands` + `hints` so an agent can act: SELECT hotspots/N+1 → `explain` / `guide missing-index-for`, errors → `schema <table>` (verify names, never guess), N+1 → batch (JOIN / `IN (...)`). After analyzing, run each finding's `suggestedCommands`, read its `hints`, then propose the fix. |
 | `status` | query-only+ | Safe JSON/text summary (no credentials). |
 | `inspect` | query-only+ | Read-only context snapshot (connection, permission, blacklist, objects, snippets, context-aware `suggestedCommands`, and **(v1.23)** human-readable `hints`). `--for-agent` / `--brief` / `--no-connect` / `--require-schema-cache`. Supports `--recovery`. |
 | `report` | query-only+ | Diagnostic report built from `@diag/*` snippets. `--section <health\|capacity\|perf>` (comma-separated to combine), `--brief`, `--for-agent`, `--no-connect`. |

package/.github/skills/dbcli/reference.md

@@ -1202,6 +1202,15 @@ dbcli proxy analyze --slow-ms 200 --n-plus-one 5  # custom thresholds
 
 **`proxy analyze`** — offline aggregation of the event log (no DB). Flags: `--events <path>` (default `.dbcli/proxy/events.jsonl`), `--format json|text` (default `json`), `--top <n>` (default 20; text rows + suggestedCommands depth), `--slow-ms <ms>` (default 1000; recomputes slowCount), `--n-plus-one <n>` (default 10), `--no-include-rotated`. JSON report blocks: `summary`, `byFingerprint` (sorted by total time; SELECT entries in the top-N carry `suggestedCommands` for `explain` / `guide missing-index-for`), `slowest`, `errors`, `hotTables`, `repetition` (N+1 suspects). Reads the current log plus the rotated `.1` segment by default.
 
+Every actionable block carries machine-readable next steps so an agent can move from "what is wrong" to "what to run":
+- `byFingerprint[]` — top-N SELECT entries get `suggestedCommands`: `dbcli explain "<sql>"` and `dbcli guide missing-index-for "<sql>"`.
+- `errors[]` — carries `tables`; emits `suggestedCommands` of `dbcli schema <table>` (capped at the first 3 tables) plus a `hints` note to verify table/column names before fixing (never guess column names). No tables known (e.g. a syntax error) → no `suggestedCommands`, but the hint still appears.
+- `repetition[]` — carries `statement` and a runnable `exampleSql` (the slowest occurrence). SELECT N+1 groups get `explain` / `guide missing-index-for` `suggestedCommands`; every group carries a `hints` note suggesting batching (JOIN / `IN (...)`) or caching.
+
+`suggestedCommands` are emitted as strings only — `proxy analyze` never executes them. When the proxy ran with `--redact literals`, `exampleSql` (and therefore the suggested commands) contains `?` placeholders; fill in real values before running them.
+
+**Acting on the report (agent loop):** after `proxy analyze`, for each block read `hints` for the diagnosis, run the entry's `suggestedCommands` to gather schema/plan/index evidence, then propose a concrete fix — add an index (`guide missing-index-for`), rewrite a slow SELECT (`explain`), batch an N+1 (`repetition`), or correct a column/table name (`errors` → `schema`). The text format mirrors this with aggregated `SUGGESTED COMMANDS` and `HINTS` sections; JSON keeps the suggestions attached per-finding.
+
 **Engines:** MySQL / MariaDB / PostgreSQL
 **Permission:** n/a (acts as a TCP relay; does not use dbcli's SQL permission model)
 

package/CHANGELOG.md

@@ -5,6 +5,34 @@ All notable changes to dbcli are documented here.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.49.0] - 2026-08-06 - 欄位遮罩涵蓋攤平與陣列列，且不再為每條規則掃過整個結果集
+
+延續 1.48.0 的 blacklist 主題：那一版修的是「哪些物件受保護」，這一版修的是「受保護的欄位到底有沒有真的被遮掉」，以及遮罩本身的成本。兩條安全性修復都屬於 fail-open —— 資料原樣回傳，其中一條連安全通知都不會發。
+
+### Added
+
+- **`dbcli proxy analyze` — 各區塊行動化(commands + hints),銜接 AI agent 介入。** `errors[]` 新增 `tables`,並附 `suggestedCommands`(`dbcli schema <table>`,最多前 3 表)+ `hints`(先核對表名/欄名再修正,勿臆測欄名);`repetition[]` 新增 `statement` 與可執行的 `exampleSql`(取最慢一筆),SELECT 群組附 `explain` / `guide missing-index-for` 的 `suggestedCommands`,每組附 N+1 批次化/快取的 `hints`。text 輸出彙整為 `SUGGESTED COMMANDS` 與 `HINTS` 區段;JSON 將建議附在各發現上。沿用 inspect 的 `suggestedCommands` + `hints` 雙軌慣例。skill 與使用者文件同步說明「analyze 後的 agent 行動流程」。
+
+### Security
+
+- **黑名單指定父欄位時，攤平後的子欄位未被遮蔽，而且不發通知。** Elasticsearch adapter 會把 `_source` 遞迴攤平成帶點的頂層鍵（`{profile:{ssn}}` → 鍵 `profile.ssn`，文件裡根本沒有 `profile`），而遮蔽判斷是以欄位名等值比對，因此把 `profile` 列入欄位黑名單完全沒有作用：資料原樣回傳，且因為「已遮蔽欄位」清單是空的，連安全通知都不會發出 —— 使用者不會知道有東西本來該被藏起來。影響 `query` / `q` / `export` 三條 Elasticsearch 路徑。現在任何位於黑名單祖先之下的欄位都會被遮蔽（`profiles`、`profile_name` 這類僅前綴相似的欄位不受影響）。已知天花板（規則指定葉節點名、或祖先不從路徑開頭起算）記於 `docs/security-threat-model.md`。
+
+- **欄位遮罩把「列本身是陣列」當成 record，遮蔽通知發了但資料沒被拿掉。** 判斷「這個欄位在不在」的 `readPath` 把陣列當透明容器往裡面找，但實際執行遮蔽的 `cloneRecord` 把它當 record，於是索引變成鍵、元素裡的敏感欄位原封不動 —— 一列被回報為已遮蔽，卻是完整回傳的。兩邊現在都看穿陣列（含巢狀陣列）。目前沒有任何 adapter 會回傳陣列列，所以這是把「偵測」與「移除」兩半釘在同一個答案上，而不是修一個可觸發的洩漏。
+
+### Performance
+
+- **欄位遮罩對每一列的每一個欄位各複製一次整列。** `omitFieldPaths` 逐一路徑呼叫 `omitPath`，而後者每次都重建整個 record，因此成本是 O(列數 × 遮罩欄位數) 次完整複製 —— 100 列對上 50 個遮罩欄位就是 5000 次。沒有點的路徑只會刪掉一個頂層鍵，而欄位黑名單絕大多數就是這種名稱，現在合併成一趟處理，只有真正的巢狀路徑才遞迴。輸出完全相同，實測 30.37ms → 1.73ms。任何回傳大量資料又設有欄位黑名單的查詢或匯出都會受益。
+
+- **黑名單規則多、命中少時，遮罩層對每條規則掃過整個結果集。** 判斷一條規則有沒有東西要遮，會對每一列呼叫一次 `hasFieldPath`，而它每次都重跑一遍 `path.split('.')`。掃不到表名的 fail-safe 分支會把設定裡的**每一條**規則都套上來，所以「規則多、命中少」正是它的常態形狀：60 條帶點且完全不命中的規則對上 1000 列，實測 12.10ms，而真正遮掉的只有一個欄位。現在沒有點的規則直接由既有的欄位集合精確回答，帶點的規則只在「開頭那一段在某列真的是物件或陣列」時才走訪列 —— 這與 `readPath` 本來就會判斷的條件相同，只是每條規則判一次而不是每列判一次；而讀取欄位值（比只列出名稱貴約 4 倍，也會觸發列物件上的 getter）只在黑名單裡真的有帶點規則時才做，所以常見的純欄位名黑名單反而比修改前更快。輸出完全相同，實測 12.10ms → 0.70ms，一般查詢的遮罩 0.69ms → 0.53ms。
+
+### Changed
+
+- **效能基準改為阻擋 CI。** `ci.yml` 的 `continue-on-error: true` 是那兩條基準能失敗四個月沒人發現的原因，已移除。同時每條基準都改為取多次量測的中位數並印出實測值，預算也依**實際 runner** 的數字重訂（最慢的 windows-latest 1.3.3 量到 3.25ms，原本 5ms 的預算只剩 35% 餘裕，放寬到 8ms）。一個會被強制執行的寬預算，勝過一個被忽略的嚴預算。
+
+### Fixed
+
+- **兩條效能基準自 2026-03-26 加入起就沒通過過。** 它們被 `ci.yml` 的 `continue-on-error: true` 蓋住，所以 CI 從未因此變紅，超標 6 倍也沒人看見。門檻本身是合理的（修正後餘裕 3 倍），問題在上面那條實作。基準也改為取多次量測的中位數並印出實測值：單次 `performance.now()` 加硬門檻約每三次就會誤報一次，當不了 gate。
+
 ## [1.48.0] - 2026-08-05 - blacklist 涵蓋語句中的每一張表，以及所有執行路徑（安全性修復）
 
 對應 issue [#23](https://github.com/CarlLee1983/dbcli/issues/23)。與 1.47.1 修掉的六個繞過不同，這一批洩漏的是**讀取內容**，不是寫入能力。

package/assets/SKILL.md

@@ -363,7 +363,7 @@ Full flags and edge cases: see [reference.md](reference.md) `init` section.
 | `assert` | query-only+ | **(v1.25)** SQL only. Verify an invariant; exit 1 on failure unless `--no-fail`. `--expect "rows>0\|value==X\|col:c not null\|unique\|between a and b\|>= n"`, `--vs <query> --compare rows\|value` (reconcile), `--against <snapshot> --tolerance <pct>`. |
 | `verification` | n/a | Inspect and manage local verification artifacts. `list` / `show <id-or-path>` / `summary` are read-only; `prune` is dry-run by default and deletes only with `--execute --force`. Reads `<cwd>/.dbcli/verification/`; no DB connection, no audit writes. |
 | `backfill artifact` | n/a | Build a bounded, reviewable source-to-SQL backfill artifact from JSON. Includes source/target identity, blacklist/schema preflight, read-back verification, and rollback hints; dry-run only and never executes writes. |
-| `proxy` | n/a | **(v1.26)** MySQL/MariaDB/PostgreSQL only. Local-dev observability proxy — relays app traffic to the real DB and appends query/latency/byte/error events to `.dbcli/proxy/events.jsonl`. Subcommands: `mysql` \| `mariadb` \| `postgresql`. `--listen`, `--target`, `--events`, `--slow-ms` (default `1000`), `--redact none\|literals`. Observe-only. **(v1.27)** `proxy analyze` aggregates the event log offline into a JSON/text report (summary, byFingerprint with suggestedCommands, slowest, errors, hotTables, N+1) — errors out if no events exist yet. |
+| `proxy` | n/a | **(v1.26)** MySQL/MariaDB/PostgreSQL only. Local-dev observability proxy — relays app traffic to the real DB and appends query/latency/byte/error events to `.dbcli/proxy/events.jsonl`. Subcommands: `mysql` \| `mariadb` \| `postgresql`. `--listen`, `--target`, `--events` (default `.dbcli/proxy/events.jsonl`), `--slow-ms` (default `1000`), `--redact none\|literals` (default `none`). Observe-only. **(v1.27)** `proxy analyze` aggregates the event log offline into a JSON/text report (summary, byFingerprint, slowest, errors, hotTables, N+1) — `--format`, `--top`, `--slow-ms`, `--n-plus-one`; errors out if no events exist yet. Actionable blocks carry `suggestedCommands` + `hints` so an agent can act: SELECT hotspots/N+1 → `explain` / `guide missing-index-for`, errors → `schema <table>` (verify names, never guess), N+1 → batch (JOIN / `IN (...)`). After analyzing, run each finding's `suggestedCommands`, read its `hints`, then propose the fix. |
 | `status` | query-only+ | Safe JSON/text summary (no credentials). |
 | `inspect` | query-only+ | Read-only context snapshot (connection, permission, blacklist, objects, snippets, context-aware `suggestedCommands`, and **(v1.23)** human-readable `hints`). `--for-agent` / `--brief` / `--no-connect` / `--require-schema-cache`. Supports `--recovery`. |
 | `report` | query-only+ | Diagnostic report built from `@diag/*` snippets. `--section <health\|capacity\|perf>` (comma-separated to combine), `--brief`, `--for-agent`, `--no-connect`. |

package/assets/reference.md

@@ -1202,6 +1202,15 @@ dbcli proxy analyze --slow-ms 200 --n-plus-one 5  # custom thresholds
 
 **`proxy analyze`** — offline aggregation of the event log (no DB). Flags: `--events <path>` (default `.dbcli/proxy/events.jsonl`), `--format json|text` (default `json`), `--top <n>` (default 20; text rows + suggestedCommands depth), `--slow-ms <ms>` (default 1000; recomputes slowCount), `--n-plus-one <n>` (default 10), `--no-include-rotated`. JSON report blocks: `summary`, `byFingerprint` (sorted by total time; SELECT entries in the top-N carry `suggestedCommands` for `explain` / `guide missing-index-for`), `slowest`, `errors`, `hotTables`, `repetition` (N+1 suspects). Reads the current log plus the rotated `.1` segment by default.
 
+Every actionable block carries machine-readable next steps so an agent can move from "what is wrong" to "what to run":
+- `byFingerprint[]` — top-N SELECT entries get `suggestedCommands`: `dbcli explain "<sql>"` and `dbcli guide missing-index-for "<sql>"`.
+- `errors[]` — carries `tables`; emits `suggestedCommands` of `dbcli schema <table>` (capped at the first 3 tables) plus a `hints` note to verify table/column names before fixing (never guess column names). No tables known (e.g. a syntax error) → no `suggestedCommands`, but the hint still appears.
+- `repetition[]` — carries `statement` and a runnable `exampleSql` (the slowest occurrence). SELECT N+1 groups get `explain` / `guide missing-index-for` `suggestedCommands`; every group carries a `hints` note suggesting batching (JOIN / `IN (...)`) or caching.
+
+`suggestedCommands` are emitted as strings only — `proxy analyze` never executes them. When the proxy ran with `--redact literals`, `exampleSql` (and therefore the suggested commands) contains `?` placeholders; fill in real values before running them.
+
+**Acting on the report (agent loop):** after `proxy analyze`, for each block read `hints` for the diagnosis, run the entry's `suggestedCommands` to gather schema/plan/index evidence, then propose a concrete fix — add an index (`guide missing-index-for`), rewrite a slow SELECT (`explain`), batch an N+1 (`repetition`), or correct a column/table name (`errors` → `schema`). The text format mirrors this with aggregated `SUGGESTED COMMANDS` and `HINTS` sections; JSON keeps the suggestions attached per-finding.
+
 **Engines:** MySQL / MariaDB / PostgreSQL
 **Permission:** n/a (acts as a TCP relay; does not use dbcli's SQL permission model)