ibmi

name: ibmi description: "Core skill for working with IBM i systems through the ibmi CLI. Covers text-to-SQL methodology, Db2 for i conventions, schema discovery, multi-system configuration, and — critically — agent scripting patterns (automatic JSON-when-piped, semantic exit codes, NDJSON streaming, dry-run planning, watch mode, multi-system workflows). Use this skill as the foundation for ANY IBM i task: running queries, exploring the database, configuring systems, writing bash/agent scripts that target IBM i, or composing pipelines that need structured output and reliable error handling."

IBM i Core — CLI, Text-to-SQL, and Agent Scripting

Foundation skill for working with IBM i systems. Covers CLI usage, Db2 for i SQL conventions, and the scripting patterns that make ibmi a reliable building block for agent workflows, bash automation, and CI pipelines.

Preflight

Before running any ibmi command, verify the CLI is installed, recent enough, and has a system configured.

command -v ibmi                  # CLI present on PATH
ibmi --version                   # expect >= 0.5 for all features in this skill
ibmi system list                 # shows configured systems and which is default

If the CLI is missing, ask the user whether to install it, then run one of:

# Recommended — daily use
npm i -g @ibm/ibmi-cli

# One-shot via npx (no install; first call is slower while npx caches)
npx -y @ibm/ibmi-cli --help

If the CLI is outdated (a flag or subcommand used below is rejected), upgrade:

npm update -g @ibm/ibmi-cli

If no systems are configured, ibmi system list returns no rows. Walk the user through the System connections section below (ibmi system add → ibmi system default → ibmi system test) before issuing queries. For ephemeral / CI environments, set DB2i_HOST, DB2i_USER, and DB2i_PASS as environment variables instead of using the config file.

Never execute destructive or mutating SQL (INSERT/UPDATE/DELETE/CALL, or any --no-read-only) without confirming scope and system with the user first.

CLI at a glance

The CLI exposes first-class subcommands for every common discovery/execution task. Prefer them over hand-written SQL when they exist.

Global flags useful on any command: --system <name>, --format {table,json,csv,markdown}, --raw (json shorthand), --stream (NDJSON), --watch <s>, --output <path>, --no-color.

System connections

ibmi system list                         # Show configured systems + default
ibmi system add prod --host ... --user ...
ibmi system test prod                    # Connectivity check — use in CI
ibmi system default dev                  # Set default
ibmi --system prod sql "SELECT ..."      # Override per command

ibmi config show prints the resolved config and which file each value came from — useful when a script misbehaves because of a stale env var or config file.

Text-to-SQL methodology

Iterate in four steps. Never guess names — always discover first.

1. Discover the schema

ibmi schemas                             # What libraries exist
ibmi tables QSYS2 --filter '%JOB%'       # Tables matching a pattern
ibmi columns QSYS2 ACTIVE_JOB_INFO       # Column metadata
ibmi describe QSYS2.ACTIVE_JOB_INFO      # Full DDL

2. Validate before executing

ibmi validate "SELECT JOB_NAME, CPU_TIME FROM TABLE(QSYS2.ACTIVE_JOB_INFO())"

For YAML tools, use --dry-run to see the resolved SQL without hitting the database:

ibmi tool list_active_jobs --tools "$SKILL_DIR/tools/" --dry-run

3. Start small, iterate

Begin narrow, inspect, then expand:

ibmi sql "SELECT * FROM QSYS2.JOB_QUEUE_INFO FETCH FIRST 3 ROWS ONLY"
ibmi sql "SELECT JOB_QUEUE_NAME, NUMBER_OF_JOBS FROM QSYS2.JOB_QUEUE_INFO ORDER BY NUMBER_OF_JOBS DESC FETCH FIRST 10 ROWS ONLY"

4. Search the service catalog when you don't know which view to use

# Find services by keyword
ibmi sql "SELECT SERVICE_CATEGORY, SERVICE_NAME, SQL_OBJECT_TYPE FROM QSYS2.SERVICES_INFO WHERE UPPER(SERVICE_NAME) LIKE '%JOB%' ORDER BY SERVICE_CATEGORY"

# Usage example for a known service
ibmi sql "SELECT EXAMPLE FROM QSYS2.SERVICES_INFO WHERE SERVICE_NAME = 'ACTIVE_JOB_INFO'"

The helper tools search_services and list_service_categories (in tools/ibmi.yaml) wrap these for convenience.

Db2 for i SQL conventions

Table functions require TABLE()

-- GOOD
SELECT * FROM TABLE(QSYS2.ACTIVE_JOB_INFO(SUBSYSTEM_LIST_FILTER => 'QBATCH')) X

-- WRONG (SQL0204 / parse error)
SELECT * FROM QSYS2.ACTIVE_JOB_INFO()

Use UDTF filter parameters, not WHERE

-- GOOD: filter is pushed inside the table function
SELECT * FROM TABLE(QSYS2.JOB_INFO(
  JOB_STATUS_FILTER => '*ACTIVE',
  JOB_USER_FILTER   => 'MYUSER'))

-- BAD: scans everything, filters afterward
SELECT * FROM TABLE(QSYS2.JOB_INFO()) X
WHERE JOB_STATUS = '*ACTIVE' AND AUTHORIZATION_NAME = 'MYUSER'

Always FETCH FIRST — views can return millions of rows.

Case sensitivity — names are uppercase by default. UPPER() user-supplied parameters.

Two naming conventions — SQL naming SCHEMA.TABLE (default) vs system naming LIBRARY/FILE. Use SQL naming everywhere.

Common optional-filter pattern

WHERE (:filter = '' OR COLUMN = UPPER(:filter))
WHERE (:filter = '*ALL' OR COLUMN = UPPER(:filter))

Agent & scripting best practices

The CLI is built for programmatic use: piped output auto-switches to JSON, exit codes are semantic, and --dry-run lets an agent plan without hitting the database.

1. JSON automatically when piped — no --format json needed

# Not a TTY → JSON output
count=$(ibmi sql "SELECT COUNT(*) AS CNT FROM SAMPLE.EMPLOYEE" | jq '.data[0].CNT')
echo "Employee count: $count"

Force JSON explicitly with --raw (or --format json) when you need to be certain — e.g. inside bash -c where stdin/stdout detection varies.

2. Semantic exit codes — branch logic without parsing output

ibmi sql "SELECT 1 FROM SYSIBM.SYSDUMMY1" --system prod
case $? in
  0) echo "Connection OK" ;;
  1) echo "Connection failed — check host/port" ;;
  3) echo "SQL error — check query syntax" ;;
  5) echo "Auth failed — check credentials" ;;
esac

JSON errors also carry a machine-readable error.code alongside the message:

GENERAL_ERROR, USAGE_ERROR, CONNECTION_ERROR, QUERY_ERROR, SQL_ERROR,
SECURITY_VIOLATION, AUTH_FAILURE, NOT_FOUND, TIMEOUT

out=$(ibmi sql "SELECT * FROM SAMPLE.BADTABLE" 2>/dev/null)
code=$(echo "$out" | jq -r '.error.code // empty')
[ "$code" = "SQL_ERROR" ] && echo "Recognised SQL error; continuing..."

3. --stream for large result sets (NDJSON, one object per row)

ibmi sql "SELECT * FROM SAMPLE.EMPLOYEE" --stream | while IFS= read -r row; do
  name=$(echo "$row" | jq -r '.FIRSTNME')
  echo "Processing: $name"
done

--stream is memory-safe for multi-million-row exports and works with any SQL statement (including UDTF-backed views).

4. --dry-run for safe agent planning

--dry-run resolves parameters and prints the final SQL without opening a database connection. Use it when an agent is deciding whether to commit to a query:

# Preview SQL for a direct query
ibmi sql "SELECT * FROM SAMPLE.EMPLOYEE WHERE SALARY > 50000" --dry-run

# Preview YAML-tool SQL with bound parameters
ibmi tool monthly_report --tools ./tools.yaml --schema SAMPLE --month 3 --dry-run

Combine with ibmi validate for a no-connection "will this parse + reference real objects?" check.

5. --output writes results to a file

ibmi sql "SELECT * FROM SAMPLE.EMPLOYEE" --format csv --output /tmp/employees.csv
ibmi sql "SELECT * FROM SAMPLE.EMPLOYEE" --raw        --output /tmp/employees.json

6. --watch <seconds> for monitoring loops

# Sample active jobs every 10s
ibmi sql "SELECT JOB_NAME, CPU_TIME FROM TABLE(QSYS2.ACTIVE_JOB_INFO()) WHERE JOB_STATUS='RUN'" \
  --watch 10

# Write a CSV time-series to disk
ibmi sql "SELECT * FROM TABLE(QSYS2.SYSTEM_STATUS())" \
  --format csv --output /tmp/status.csv --watch 30

Ctrl+C stops the loop; --output re-writes the file each tick (it does not append — redirect --stream output to append).

7. Multi-system workflows — --system <name> per command

dev_count=$(ibmi sql  "SELECT COUNT(*) AS C FROM MYLIB.ORDERS" --system dev  | jq '.data[0].C')
prod_count=$(ibmi sql "SELECT COUNT(*) AS C FROM MYLIB.ORDERS" --system prod | jq '.data[0].C')
echo "Dev: $dev_count, Prod: $prod_count"

8. Put it all together — a text-to-SQL agent step

#!/bin/bash
set -euo pipefail

SYSTEM="${IBMI_SYSTEM:-dev}"
SCHEMA="SAMPLE"
TABLE="EMPLOYEE"

# 1. Discover
ibmi tables  "$SCHEMA" --filter "${TABLE}%" --system "$SYSTEM" > /dev/null
ibmi columns "$SCHEMA" "$TABLE" --system "$SYSTEM" > /dev/null

# 2. Plan — no DB connection needed
query="SELECT EMPNO, SALARY FROM $SCHEMA.$TABLE WHERE SALARY > 50000 FETCH FIRST 100 ROWS ONLY"
ibmi validate "$query" --system "$SYSTEM"

# 3. Execute, branching on exit code
if ibmi sql "$query" --system "$SYSTEM" --raw --output ./result.json; then
  rows=$(jq '.meta.rows' ./result.json)
  echo "Retrieved $rows rows"
else
  case $? in
    3) echo "SQL error — inspect ./result.json"; exit 3 ;;
    5) echo "Auth failure — re-authenticate"; exit 5 ;;
    *) echo "Unexpected failure ($?)"; exit 1 ;;
  esac
fi

For deeper recipes (cross-system diffs, monitoring alert loops, a full text-to-SQL agent), see references/agent-scripting.md.

Gotchas

• --format is ignored when --output has an extension — the extension wins (.csv → CSV regardless of --format).
• --read-only is default-on; mutation SQL (INSERT/UPDATE/DELETE/CALL) requires --no-read-only AND a writable system entry.
• Positional key=value passed to ibmi tool is silently ignored. Use --kebab-case-name <value> — flag names come from the YAML param name with underscores converted to hyphens.
• --watch + --output rewrites the same file each tick; to build a time series pipe --stream to tee -a.
• ibmi tools with no --tools emits "No tools path specified" — there are no globally-installed YAML tools; every skill ships its own tools/ directory.

Error diagnosis cheat sheet

Helper YAML tools

tools/ibmi.yaml ships a small set of tools that go beyond what the CLI exposes natively. Native-command duplicates (list_tables_in_schema, get_column_info, validate_query, sample_rows) have been removed — use the CLI subcommand instead.

ibmi tool get_table_statistics --tools "$SKILL_DIR/tools/" --schema-name QSYS2 --table-name SYSTABLES
ibmi tool search_services      --tools "$SKILL_DIR/tools/" --keyword JOB
ibmi tool list_service_categories --tools "$SKILL_DIR/tools/"

Reference documentation

• Agent scripting cookbook — deep recipes: cross-system diff, monitoring loop, text-to-SQL agent
• Service catalog overview — category inventory
• SQL patterns — reusable query snippets
• IBM i CLI docs (upstream) — complete documentation index; key pages:
  • CLI commands reference
  • Agent integration
  • Output formats
  • YAML tools
• IBM i SQL Services — official service reference