Dolphindb

DolphinDB Skill

One skill, two modes:

• Runtime — run bash/Python snippets against the user's live DolphinDB server (patched connection info below).
• Reference — offline knowledge base for syntax, engines, plugins, error codes, and best practices (Routing Table below).

All content targets DolphinDB Server 3.00+ and its official client APIs.

---

⚠️ Authoritative connection info — USE THESE VALUES VERBATIM

Do not invent defaults like localhost:8848 or 127.0.0.1:8848. The values in the table below were written into this file by the dolphindb-agent-skills installer and are the user's real DolphinDB server. Every s.connect(...) call in the Runtime Patterns section is already hard-coded with these same 4 values — copy a snippet as-is, do not rewrite it.

Field	Value
Host	{{DDB_HOST}}
Port	{{DDB_PORT}}
User	{{DDB_USER}}
Password	{{DDB_PASSWD}}

If the table above still shows literal {{DDB_HOST}} / {{DDB_PORT}} / {{DDB_USER}} / {{DDB_PASSWD}} placeholders, the user never ran the dolphindb-agent-skills installer (or ran it non-interactively). Tell them to re-run it in a real terminal and enter their real server info.

---

Decision tree — runtime vs reference

1. User wants to run / query / execute against their DolphinDB (e.g. "what databases do I have", "show me 10 rows", "跑一下这个脚本", "我的 dolphindb 里有哪些库") → use the Runtime Patterns section below.
2. User shows a .dos file or inline DolphinDB script and asks "does this work?" / "what does this return?" → run it via Runtime Pattern 2 or Pattern 3 and report the real result.
3. User asks "what databases/tables exist?" / "how big is this table?" → Runtime Pattern 6 (getClusterDFSDatabases, getTables(database(…)), getTableDiskUsage).
4. User has a local CSV / pandas DataFrame to push into DolphinDB → Runtime Pattern 4 (upload + query) or Pattern 7 (bulk insert).
5. Long batch job / many calls → start from Runtime Pattern 8 (robust connect) or Pattern 10 (reusable DDBClient), wrap each call with Runtime Pattern 9's run_safely helper.
6. User only wants explanation / syntax / design / error-code lookup → jump to the Routing Table and pull the right docs/ or reference/ file.

Safety rules (for runtime execution)

• Read-only by default. Do not run drop*, dropPartition, delete from, truncate, rename*, or DDL that mutates the cluster unless the user explicitly asked for it.
• Start small. Probe with select top 10 … / select count(*) … before running heavy aggregations.
• Echo the script you ran in your reply so the user can audit.
• Partition column in where. Always filter on the partition column (usually a date/time) to avoid full-cluster scans.

---

Runtime Patterns

Pattern 1 — One-liner sanity check

python3 -c "import dolphindb as ddb; s=ddb.session(); s.connect('127.0.0.1', 8848, 'admin', '123456'); print(s.run('version()'))"

If this prints a version string, the connection is healthy.

---

Pattern 2 — Run a .dos script file

python3 << 'PYEOF'
import dolphindb as ddb
s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456")
script = open("/path/to/your/script.dos").read()
result = s.run(script)
print(result)
PYEOF

Replace /path/to/your/script.dos with the actual file path (use Glob to find it if the user didn't say).

---

Pattern 3 — Run an inline DolphinDB script

python3 << 'PYEOF'
import dolphindb as ddb
s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456")

script = """
symVec = `AAPL`MSFT`GOOG
n=50; ts = 2024.01.02T09:30:00.000 + (0..(n-1))*60000
syms=array(SYMBOL,0); times=array(TIMESTAMP,0)
opens=array(DOUBLE,0); highs=array(DOUBLE,0); lows=array(DOUBLE,0)
closes=array(DOUBLE,0); vols=array(LONG,0)
for(sym in symVec){
    bp=100.0+rand(100.0,1)[0]; bv=5000.0+rand(3000.0,1)[0]
    for(t in ts){
        o=bp+rand(2.0,1)[0]; h=o+rand(1.5,1)[0]; l=o-rand(1.5,1)[0]
        c=l+rand(h-l,1)[0]; v=round(bv+rand(2000.0,1)[0],0)
        syms.append!(sym); times.append!(t)
        opens.append!(o); highs.append!(h); lows.append!(l)
        closes.append!(c); vols.append!(v)
    }
}
bars=table(syms as symbol, times as tradetime, opens as open,
           highs as high, lows as low, closes as close, vols as volume)
bars=select * from bars order by symbol, tradetime

f=select symbol, tradetime, close,
       mavg(volume,5)/mavg(volume,20) as volRatio,
       close/mavg(close,20)-1 as priceMom,
       (mavg(volume,5)/mavg(volume,20))*(close/mavg(close,20)-1) as pvpFactor
from bars context by symbol csort tradetime

print(select top 5 symbol,tradetime,close,volRatio,priceMom,pvpFactor from f)
print(select symbol,count(*) as n, avg(pvpFactor) as meanPvp from f group by symbol)
"""
r = s.run(script)
print(r)
PYEOF

---

Pattern 4 — Upload a pandas DataFrame, then query it

python3 << 'PYEOF'
import dolphindb as ddb
import pandas as pd

s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456")

df = pd.DataFrame({
    "symbol": ["AAPL"] * 5,
    "close":  [100.0, 101.0, 102.0, 101.5, 103.0],
    "volume": [1000, 1100, 1050, 1150, 1200],
})

s.upload({"myDF": df})
result = s.run("""
    select * from myDF
    context by symbol csort rowNo
""")
print(result)
PYEOF

---

Pattern 5 — Parameterized query (safe against SQL injection)

python3 << 'PYEOF'
import dolphindb as ddb
import pandas as pd

s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456")

# Upload filter values as a table, then reference by name.
local_df = pd.DataFrame({"sym": ["AAPL", "MSFT"], "d": ["2024.01.02", "2024.01.02"]})
s.upload({"filter": local_df})

result = s.run("""
    select count(*) as cnt from loadTable('dfs://demo',`trades)
    where sym in filter.sym and date in filter.d
""")
print(result)
PYEOF

Prefer this over f-string interpolation of user input.

---

Pattern 6 — DFS catalog & disk usage (canonical ops)

These 4 operations cover most "what's in this DolphinDB?" questions. Prefer these exact calls over show databases / show tables (the latter are OLAP-era aliases and don't always work on newer clusters).

python3 << 'PYEOF'
import dolphindb as ddb
s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456",
          keepAliveTime=3600, reconnect=True)

# (1) List all DFS databases on the cluster
print(s.run("getClusterDFSDatabases()"))

# (2) List tables in a specific DFS database
print(s.run('getTables(database("dfs://trades"))'))

# (3) Disk usage for one DFS table (requires the 'ops' module)
print(s.run('use ops; getTableDiskUsage("dfs://trades", "trade", byNode=false)'))

# (4) Run an arbitrary script — the universal escape hatch
print(s.run("select top 10 * from loadTable('dfs://trades', `trade)"))
PYEOF

Tip: getTableDiskUsage returns a per-chunk breakdown by default. Pass byNode=true if you want it rolled up per datanode, or wrap the call to aggregate yourself (select sum(diskSize) from …).

---

Pattern 7 — Bulk-append rows to a DFS table

python3 << 'PYEOF'
import dolphindb as ddb
import pandas as pd

s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456")

df = pd.DataFrame({
    "sym":    ["AAPL", "MSFT"],
    "date":   pd.to_datetime(["2024-01-02", "2024-01-02"]).date,
    "price":  [189.5, 370.1],
    "volume": [1000, 2000],
})
s.upload({"chunk": df})

# `tableInsert` returns the number of rows inserted.
print(s.run("""
    tableInsert(loadTable('dfs://trades', `trade), chunk)
"""))
PYEOF

For high-throughput ingestion use MultithreadedTableWriter — see docs/60-api/python-api.md via the Routing Table below.

---

Pattern 8 — Robust connect (long-running scripts / notebooks)

For anything longer than a one-shot query, pass keepAliveTime and reconnect=True so a dropped TCP connection is auto-recovered.

python3 << 'PYEOF'
import dolphindb as ddb

s = ddb.session()
s.connect(
    "127.0.0.1", 8848, "admin", "123456",
    keepAliveTime=3600,   # seconds; suppresses idle-disconnect
    reconnect=True,       # auto-reconnect on transient network errors
)

# ... many calls over hours ...
print(s.run("now()"))
s.close()
PYEOF

Other useful kwargs on session() / connect():

• enableSSL=True — if the server listens with TLS.
• highAvailability=True, highAvailabilitySites=["ip1:port", "ip2:port"] — cluster mode with failover across controllers.
• compress=True — compress result payloads for large frames.

---

Pattern 9 — Capture errors (don't crash the whole run)

Wrap every s.run(...) in try/except so a bad script does not kill the whole workflow.

python3 << 'PYEOF'
import dolphindb as ddb

def run_safely(s, script: str):
    try:
        return True, s.run(script)
    except Exception as e:
        return False, str(e)

s = ddb.session()
s.connect("127.0.0.1", 8848, "admin", "123456",
          keepAliveTime=3600, reconnect=True)

scripts = [
    "version()",
    "getClusterDFSDatabases()",
    "select top 5 * from loadTable('dfs://no_such_db', `x)",  # will fail
]
for sc in scripts:
    ok, result = run_safely(s, sc)
    status = "OK " if ok else "ERR"
    print(f"[{status}] {sc}\n  -> {result}\n")
s.close()
PYEOF

If you see RefId: Sxxxxx in the error string, look it up via the Routing Table below (reference/error-codes/Sxxxxx.md).

---

Pattern 10 — Reusable session context manager

Copy this class into a standalone script when you need to run many snippets against the same server — it guarantees close() runs even on exceptions.

# dolphindb_client.py
import dolphindb as ddb

class DDBClient:
    def __init__(self, host="127.0.0.1", port=8848,
                 user="admin", passwd="123456",
                 keep_alive_time=3600, reconnect=True):
        self.host, self.port, self.user, self.passwd = host, port, user, passwd
        self.keep_alive_time, self.reconnect = keep_alive_time, reconnect
        self.session = ddb.session()

    def __enter__(self):
        self.session.connect(
            self.host, int(self.port), self.user, self.passwd,
            keepAliveTime=self.keep_alive_time, reconnect=self.reconnect,
        )
        return self

    def __exit__(self, exc_type, exc_value, tb):
        self.session.close()

    def run(self, script: str):
        try:
            return True, self.session.run(script)
        except Exception as e:
            return False, str(e)


if __name__ == "__main__":
    with DDBClient() as c:   # defaults match this skill's patched values
        print(c.run("version()"))
        print(c.run("getClusterDFSDatabases()"))

Then:

python3 dolphindb_client.py

---

Connection troubleshooting

ConnectionRefusedError / "Connection refused"

nc -zv 127.0.0.1 8848

If that fails: DolphinDB is not running on that host:port, or a firewall is blocking it.

ModuleNotFoundError: No module named 'dolphindb'

pip install dolphindb
# or, in an externally-managed env (macOS Homebrew / PEP 668):
uv pip install dolphindb
# or to run without installing globally:
uvx --with dolphindb python3 -c "import dolphindb; print(dolphindb.__version__)"

"Server response: Authentication failed"

Credentials are wrong. Re-run the dolphindb-agent-skills installer and enter the correct user/password, or edit this file.

Script runs but s.run(...) returns None

Many DolphinDB scripts print nothing when they have no tail expression. Either add an explicit print(...) inside the script, or end the script with an expression (e.g. a variable name) whose value should be returned.

---

Reference Library

How to use this skill (reference mode)

1. Identify the user's intent and consult the Routing Table below to pick the right file.
2. For an unknown DolphinDB function, look it up in reference/functions/INDEX.md → then read the theme file it points to.
3. For a runtime error message containing RefId: Sxxxxx, look it up in reference/error-codes/INDEX.md → then read reference/error-codes/Sxxxxx.md.
4. For "how do I do X" tasks, check patterns/ first; for runnable end-to-end scripts check examples/.
5. Only fall back to the Quick snippets section below for the most common one-liners.

---

Routing Table

If the user is asking about...	Go to
What DolphinDB is / architecture / node types	docs/00-overview.md
Install, connect, first script	docs/01-quickstart.md
Data types (INT, LONG, DECIMAL, SYMBOL, TIMESTAMP, …)	docs/10-language/data-types.md
Data forms (vector, matrix, table, dict, tuple, set, pair, tensor)	docs/10-language/data-forms.md
Dict (creation, ANY values, syncDict, missing-key, merge) ★	docs/10-language/dict.md
Operators, assignment = / <-, in-place !	docs/10-language/operators.md
Control flow (if, for, do..while, try..catch)	docs/10-language/control-flow.md
Named / anonymous / lambda / partial application / higher-order	docs/10-language/functions.md
Metaprogramming, sqlCol, makeCall, sql()	docs/10-language/metaprogramming.md
Modules (use, module)	docs/10-language/modules.md
SELECT ... WHERE basics	docs/20-sql/select-where.md
group by aggregation	docs/20-sql/group-by.md
context by (per-group vectorized calc) ★	docs/20-sql/context-by.md
Time types (DATE/TIMESTAMP/NANOTIMESTAMP/… — 10 variants, join-empty trap) ★	docs/10-language/time-types.md
NULL handling (typed nulls, isValid, nullFill, window propagation) ★	docs/10-language/null-handling.md
Error handling (try/catch, RefIds, job errors, streaming poison-pill)	docs/10-language/error-handling.md
pivot by ★	docs/20-sql/pivot-by.md
Window functions / analytic functions	docs/20-sql/window-functions.md
Joins: equi / left / full / cross / asof / window / prefix	docs/20-sql/joins-overview.md
asof join (aj) / wj — time-series alignment ★	docs/20-sql/asof-join.md
update / insert into / delete / alter	docs/20-sql/update-insert-delete.md
Create DFS database, database(...), createPartitionedTable	docs/30-database/dfs-database.md
Partitioning schemes (VALUE / RANGE / HASH / LIST / COMPO)	docs/30-database/partitioning.md
TSDB engine specifics (sortColumns, keepDuplicates)	docs/30-database/tsdb-engine.md
OLAP engine specifics	docs/30-database/olap-engine.md
Primary-key engine (PKEY) — upsert semantics	docs/30-database/pkey-engine.md
DFS limits & best practices	docs/30-database/limits-and-best-practices.md
streamTable, share, persist	docs/40-streaming/stream-table.md
subscribeTable, handler, msgAsTable	docs/40-streaming/subscribe.md
Stream engines (reactiveState / timeSeries / cross / asof / session / anomaly)	docs/40-streaming/engines.md
Stream engine selection — decision tree ★	docs/40-streaming/engine-selection.md
CEP	docs/40-streaming/cep-overview.md
Historical replay	docs/40-streaming/replay.md
loadText, ploadText, schema inference	docs/50-ingestion/loadText-ploadText.md
HDF5 / Parquet / Arrow	docs/50-ingestion/hdf5-parquet.md
Kafka / MQTT ingestion	docs/50-ingestion/kafka-mqtt.md
Python API (dolphindb, ddb.session, .run, .upload)	docs/60-api/python-api.md
Java API	docs/60-api/java-api.md
C++ API	docs/60-api/cpp-api.md
Cross-language type mapping	docs/60-api/type-mapping.md
Query optimization, EXPLAIN, hints	docs/70-perf/query-optimization.md
Partition pruning	docs/70-perf/partition-pruning.md
Memory & threading tuning	docs/70-perf/memory-threading.md
Slow-query diagnosis checklist ★	docs/70-perf/slow-query-diagnosis.md
JIT (@jit) compilation guide	docs/70-perf/jit-guide.md
Cluster ops	docs/90-admin/cluster.md
Backup / restore	docs/90-admin/backup-restore.md
Users / ACL	docs/90-admin/security.md
Look up any built-in function by name	python scripts/lookup.py fn <name> (or reference/functions/INDEX.md)
Look up any runtime error RefId: Sxxxxx	python scripts/lookup.py error S00012 (or reference/error-codes/INDEX.md)
Jump to curated reads for a topic	`python scripts/lookup.py topic <backtest
One-page top-traps cheatsheet ★	docs/cheatsheet.md
Chinese ↔ English keyword map (中文提问)	docs/cn-keywords.md
Plugin quick catalog (one-line per plugin)	reference/plugins-catalog.md
Any specific plugin manual (amdQuote / Arrow / Kafka / ODBC / Parquet / CTP / INSIGHT / …)	docs/plugins/README.md (hub) + docs/plugins/<name>/ or docs/plugins/<name>.md
Worked tutorials (OHLC, backtest, IoT anomaly, scheduledJob, …)	docs/tutorials/README.md (curated index of 281 tutorials)
Built-in modules (ta, wq101alpha, gtja191Alpha, mytt, MarketHoliday, …)	docs/modules/README.md
Deployment guides / license fingerprint	docs/deploy/
DolphinDB MCP	docs/mcp/
O&M troubleshooting (connection lost / server hang / slow I/O)	docs/90-admin/omc/
Web console admin UI (user mgmt, config, stream graph, querybuilder, Shell)	docs/90-admin/web/README.md (18 pages)
Client IDE & editor integrations (VSCode, Jupyter, DBeaver, Grafana, PowerBI, Superset)	docs/60-api/{vscode,jupyter,gui,terminal,clients}.md + docs/60-api/tools/
Configuration parameter reference	docs/90-admin/cfg/
Version release notes	docs/release-notes/
Upstream top-level index & 3rd-party integrations list	docs/upstream-index.md, docs/third_party.md
Functions by topic (categorical index of all 1721 built-ins)	reference/functions/funcs_by_topics.md (55 KB) + funcs_intro.md + appendix.md
Backtest / simulated matching (Backtest plugin, MatchingEngineSimulator, OME, SimulatedExchangeEngine) ★	docs/backtest/README.md (hub) + docs/backtest/{backtest-plugin-guide,matching-engine-guide,assets,traps,factors,tutorials-index}.md
Factor / alpha computation (@state, reactive state engine, lookahead, WQ101, GTJA191) ★	docs/backtest/factors.md
Runnable end-to-end scripts	examples/ (backtest-quickstart/-future/-option, parquet-roundtrip, stream-reactive-engine, tick-to-ohlc, python-api-quickstart)
"How do I do X" recipes	patterns/ (signal-to-order, stream-ingestion-to-dfs, stream-recovery-after-restart, scheduled-job-template, python-roundtrip-type-safety, asof-join, partition-design, tick-to-ohlc, upsert-via-pkey)
Eval battery — 10 representative tasks	evals/README.md
How to measure hit-rate / uplift	evals/HOW-TO-MEASURE.md + scripts/run_evals.py

---

Common traps (read before writing DolphinDB code)

These are the most frequent mistakes agents make. Follow the linked page for details.

• context by ≠ group by. group by collapses rows; context by keeps all rows and computes per-group vectors. Use context by for rolling/cumulative per-symbol calculations. → docs/20-sql/context-by.md
• Partition column must appear in where, otherwise the query scans all partitions. Always filter on the partition column first (typically a date/time). → docs/70-perf/partition-pruning.md
• share before subscribe. A stream table must be shared (or persisted) before subscribeTable can attach. → docs/40-streaming/stream-table.md
• = vs ==. In DolphinDB, = is assignment and equality comparison inside where clauses. Use == for equality in script expressions; use = inside SQL predicates. → docs/10-language/operators.md
• <- is assignment in function definitions and also appears in some stream APIs; it is NOT a comparison operator.
• Symbol literals use backticks. `AAPL is a SYMBOL literal; "AAPL" is STRING. Mixing them changes partition routing and join behavior.
• Date literals have no quotes. Write 2024.01.01, not "2024-01-01". Use date("2024-01-01") to convert from string.
• append! mutates; append does not exist for tables. The ! suffix means in-place mutation.
• loadTable(...) is lazy. Operations are lazily planned; only fully materialized when the query is executed or the result is touched.
• Python API returns numpy-backed DataFrames. SYMBOL/STRING become object, TIMESTAMP becomes datetime64[ns]. Check docs/60-api/type-mapping.md before comparing values.
• Dict is NOT Python-style. No {"a": 1} literal — use dict(STRING, INT) (empty) or dict(keys, vals). Missing-key read returns null (not an error); use d.contains(k). Concurrent writes need syncDict, otherwise the node can crash. → docs/10-language/dict.md
• Backtest lookahead bias. mavg(close, 5) and any same-bar factor include the current bar, which is only valid if you execute at bar close. For next-bar execution, lag signals by one bar. matchingRatio=1, zero slippage, and unmodeled queue position all flatter results. → docs/backtest/traps.md
• show engines, getStreamingStat() and getPerformance() are your first debugging tools — check them before assuming a bug.

---

Quick snippets

Kept intentionally minimal. For more, read examples/.

Connect from Python

import dolphindb as ddb

s = ddb.session()
s.connect("localhost", 8848, "admin", "123456")

df = s.run("select top 100 * from loadTable('dfs://trades', `trade)")

Create a partitioned DFS table (TSDB engine)

db = database("dfs://trades", VALUE, 2024.01.01..2024.12.31, engine="TSDB")

schema = table(
    1:0,
    `sym`date`price`volume,
    [SYMBOL, DATE, DOUBLE, INT]
)

db.createPartitionedTable(
    table       = schema,
    tableName   = `trade,
    partitionColumns = `date,
    sortColumns = `sym`date
)

Append rows

t = table(
    take(`AAPL`MSFT, 10) as sym,
    take(2024.01.01..2024.01.10, 10) as date,
    rand(100.0, 10) as price,
    rand(1000, 10)   as volume
)
loadTable("dfs://trades", `trade).append!(t)

context by vs group by

// group by: 1 row per sym
select sym, avg(price) as avgPx from t group by sym

// context by: keep all rows, add per-sym 5-row moving avg
select sym, date, price, mavg(price, 5) as ma5
from t context by sym

Stream table + subscription

share streamTable(1000:0, `time`sym`price, [TIMESTAMP, SYMBOL, DOUBLE]) as trades

def myHandler(msg) { /* msg is a table when msgAsTable=true */ }

subscribeTable(
    tableName = `trades,
    actionName = `printAction,
    handler    = myHandler,
    msgAsTable = true
)

Diagnosing an error from a script

If the user shows a log line like ... RefId: S02006, read reference/error-codes/S02006.md — every error code ships with 报错信息 / 错误原因 / 解决办法.

---

Maintenance

Every file in this skill is either hand-authored or auto-mirrored from the upstream DolphinDB documentation. They coexist flatly — there is no separate _source/ layer.

Auto-mirrored files begin with the HTML comment <!-- Auto-mirrored from upstream ... -->. Do not edit them by hand; rerun the build script and they will be overwritten. Hand-authored files have no such marker and are never touched by the build.

Auto-mirrored tree (regenerated by scripts/build_from_docs.py):

• reference/functions/ — INDEX, by-theme, by-name (1718 function pages).
• reference/error-codes/ — every RefId: Sxxxxx page in full.
• reference/plugins-catalog.md — one-line summary per plugin.
• docs/**/*.md except the hand-authored files listed below.

Hand-authored (never auto-touched):

• docs/00-overview.md, docs/01-quickstart.md.
• docs/<area>/README.md in every numbered area.
• docs/10-language/{data-types,data-forms,dict,time-types,null-handling,error-handling,operators,control-flow,functions,metaprogramming,modules}.md.
• docs/20-sql/{select-where,group-by,context-by,pivot-by,window-functions,joins-overview,asof-join,update-insert-delete}.md.
• docs/30-database/{dfs-database,partitioning,tsdb-engine,olap-engine,pkey-engine,limits-and-best-practices}.md.
• docs/40-streaming/{stream-table,subscribe,engines,engine-selection,cep-overview,replay}.md.
• docs/50-ingestion/{loadText-ploadText,hdf5-parquet,kafka-mqtt}.md.
• docs/60-api/{python-api,java-api,cpp-api,type-mapping}.md.
• docs/70-perf/{partition-pruning,query-optimization,memory-threading,slow-query-diagnosis,jit-guide}.md.
• docs/90-admin/{cluster,backup-restore,security}.md.
• docs/backtest/{README,backtest-plugin-guide,matching-engine-guide,assets,traps,factors,tutorials-index}.md.
• docs/tutorials/README.md, docs/plugins/README.md, docs/modules/README.md — curated navigation indexes.
• patterns/*.md — "how do I do X" recipes.
• examples/*.dos, examples/*.py — runnable end-to-end scripts.
• evals/{README,scoring,run}.md + evals/tasks/*.md — regression battery.
• docs/cheatsheet.md — compressed top-traps.
• docs/cn-keywords.md — CN↔EN keyword map.
• scripts/lookup.py — agent-invokable CLI for error codes / functions / topics.
• SKILL.md — this file.

Rebuild after upstream changes:

python skills/dolphindb/scripts/build_from_docs.py

Only files carrying the auto-mirror marker are deleted/rewritten; anything you wrote manually is preserved across rebuilds.