Postgresql CLI

psql — PostgreSQL Interactive Terminal

psql is PostgreSQL's feature-rich interactive terminal. It lets you write and execute queries, inspect database objects, import/export data, script batch operations, and customize output formatting — all from the command line.

Prerequisites

Before using psql, verify it is installed and available:

# Check if psql is installed
psql --version

# If not found, install PostgreSQL client tools:

# macOS (Homebrew)
brew install libpq
brew link --force libpq

# Ubuntu / Debian
sudo apt install postgresql-client

# CentOS / RHEL
sudo yum install postgresql

# Alpine
apk add postgresql-client

# Windows — install PostgreSQL via the official installer or use WSL

psql ships as part of the postgresql-client package. The server (postgresql) is not required — you only need the client to connect to a remote PostgreSQL instance.

Quick Reference

Connecting

# 1. CLI flags
psql -h host -p port -U user -d dbname

# 2. Connection URI
# WARNING: Password in URI is visible in shell history and process listings.
#          Prefer ~/.pgpass for production use (see method 4 below).
psql "postgresql://user:YOUR_PASSWORD@host:port/dbname"

# 3. Environment variables (no flags needed)
export PGHOST=localhost
export PGPORT=5432
export PGDATABASE=mydb
export PGUSER=postgres
# WARNING: PGPASSWORD is visible in process listings (e.g. `ps aux`).
#          Use ~/.pgpass in production instead.
export PGPASSWORD=YOUR_PASSWORD
psql                       # picks up all params from env

# 4. ~/.pgpass file (RECOMMENDED for passwords)
#    Format: hostname:port:database:username:password
touch ~/.pgpass && chmod 600 ~/.pgpass
# Then manually edit ~/.pgpass and add entries (avoids password in shell history):
# hostname:port:database:username:password
# Example: localhost:5432:mydb:postgres:YOUR_PASSWORD
psql -h localhost -U postgres -d mydb   # no password prompt

# 5. Execute and exit
psql -f script.sql dbname                        # execute file then exit
psql -c "SELECT 1" dbname                        # run single command then exit
psql -1 -f migration.sql dbname                  # run in single transaction

# 6. Service connection (reads from pg_service.conf)
psql service=mydb_prod

# 7. Reconnect within a session
\c dbname                                       # reconnect to different db
\c -reuse-previous=on sslmode=require           # change only sslmode
\c "host=newhost port=5432 dbname=mydb"         # conninfo string

On connection failure: interactive mode keeps the previous connection; script mode closes it and all subsequent database commands fail until the next successful \c.

Key flags: -h host, -p port, -U user, -d database, -w no password prompt, -W force password prompt, -1 single transaction, -f execute file, -c execute command, -t tuples only, -x expanded, -A unaligned, -E echo hidden queries (\d internals), -L log file, -X skip ~/.psqlrc.

Connection precedence: CLI flags > environment variables > pg_service.conf > defaults. Password precedence: connection string/password flag > PGPASSWORD env > ~/.pgpass. Use ~/.pgpass instead of PGPASSWORD in production — PGPASSWORD is visible in process listings (ps aux).

Object Inspection (\d family)

| \dA | Access methods | | \dAc / \dAf / \dAo / \dAp | Operator classes, families, operators, support functions | | \dC | Type casts | | \dconfig | Server configuration parameters (\dconfig * for all, PostgreSQL 16+) | | \dd | Object descriptions (comments) | | \ddp | Default privileges | | \dL | Procedural languages | | \do | Operators (accepts arg type patterns) | | \dO | Collations | | \dP[itn] | Partitioned tables (t=tables, i=indexes, n=nested) | | \drg | Granted role memberships | | \dRp / \dRs | Replication publications / subscriptions | | \dX | Extended statistics | | \dx | Installed extensions | | \dy | Event triggers | | \sf[+] | Show function definition | | \sv[+] | Show view definition | | \z | Privileges (alias for \dp) |

Modifiers (append to most \d commands):

• + — extra info (size, description): \dt+, \l+, \du+
• S — include system objects: \dtS, \dfS+
• x — expanded display mode: \dt+x (note: \dx is a different command; x must follow S or +)

Provide a name for details: \d table_name shows columns, types, indexes, constraints, foreign keys.

Pattern matching in \d commands:

• * = any sequence of characters, ? = single character
• . separates schema from object: \dt public.* or \dt my_schema.users
• .. separates database.schema.object: \dt mydb.public.* (db must match current db)
• Double quotes stop case folding and wildcard expansion: \dt "FOO" matches FOO not foo
• $ is matched literally (not regex anchor)
• Regex chars like [0-9] work: \dt user[0-9]* matches user1, user2
• No pattern: shows all objects visible in current search_path (not all objects in DB)
• Use *.* to see all objects in all schemas regardless of visibility

Query Execution

Data Import/Export

-- Server-side (requires superuser for file access, uses server filesystem)
COPY table TO '/path/file.csv' WITH (FORMAT csv, HEADER true);
COPY table FROM '/path/file.csv' WITH (FORMAT csv, HEADER true);

-- Client-side (runs with client permissions, no superuser needed) — preferred
\copy table TO '/path/file.csv' WITH (FORMAT csv, HEADER true)
\copy table FROM '/path/file.csv' WITH (FORMAT csv, HEADER true)
\copy (SELECT ...) TO '/path/output.csv' WITH (FORMAT csv, HEADER true)

-- Advanced: specific columns, NULL handling, custom delimiter
\copy table (col1, col2) FROM 'data.csv' WITH (FORMAT csv, HEADER true, NULL 'N/A')

\copy is the go-to for day-to-day work — it uses the client's filesystem and permissions, not the server's.

\copy syntax detail:

-- FROM (import): sources are 'filename', program 'command', stdin, pstdin
\copy table FROM 'file.csv' WITH (FORMAT csv, HEADER true) [ WHERE condition ]

-- TO (export): destinations are 'filename', program 'command', stdout, pstdout
\copy table TO 'file.csv' WITH (FORMAT csv, HEADER true)

For \copy ... FROM stdin, data rows continue until a line containing only \. is read or EOF is reached. Use pstdin/pstdout to always read/write psql's actual stdin/stdout regardless of \o setting.

WARNING: The program option executes a shell command. If constructed from user input, it can lead to command injection. Avoid string concatenation with untrusted data.

Tip: \copy takes the entire rest of the line as arguments (no variable interpolation). When you need variable interpolation or multi-line queries, use SQL COPY ... TO STDOUT with \g instead:

-- This allows variable interpolation and multi-line queries
COPY (SELECT * FROM :table WHERE id > :min_id) TO STDOUT WITH (FORMAT csv, HEADER true) \g /tmp/output.csv

Output Formatting

\a                  Toggle aligned/unaligned output
\x                  Toggle expanded display (vertical vs table)
\t                  Toggle tuples only (no headers/footers)
\pset format FORMAT  Set output format: aligned, asciidoc, csv, html, latex, latex-longtable, troff-ms, unaligned, wrapped
\pset border N       Set border style (0-2; 3 for latex data-row lines)
\pset null STRING    Display NULL as STRING
\pset pager [off]    Control pager usage
\pset title 'TEXT'   Set table title
\pset recordsep SEP  Set record separator for unaligned mode
\pset fieldsep SEP   Set field separator for unaligned mode (default: |)
\pset footer [on|off] Toggle row count footer
\pset columns N      Set target width for wrapped format
\pset csv_fieldsep C  Set CSV field separator (default: comma)
\pset numericlocale [on|off]  Toggle locale-specific number formatting
\pset linestyle STYLE Set border style: ascii, old-ascii, unicode
\pset pager_min_lines N  Minimum lines before pager activates
\pset xheader_width MODE  Expanded header width: full, column, page, or N (PostgreSQL 17+)
\H                   Toggle HTML output (shortcut)
\C [title]           Set table title (shortcut for \pset title)
\f [string]          Set field separator (shortcut for \pset fieldsep)
\T table_options     Set HTML table attributes (shortcut for \pset tableattr)

Large Objects

\lo_import filename [comment]   Import file as large object, returns OID
\lo_export loid filename        Export large object to file
\lo_list[x+]                    List all large objects
\lo_unlink loid                 Delete large object

Large object OIDs are persistent references. Always associate a human-readable comment on import. Use \lo_list to find OIDs.

Scripting & Control Flow

\i filename         Execute file (relative to current working directory)
\ir filename        Execute file (relative to the script being processed)
\o [filename]       Redirect query output to file (or pipe with |cmd)
\o                   Stop output redirection
\qecho TEXT          Output text to redirected output
\echo TEXT           Output text to stdout (-n suppresses trailing newline)
\warn TEXT           Output text to stderr
\! command           Execute shell command
\cd [dir]            Change working directory
\set NAME VALUE      Set psql variable
\unset NAME          Unset psql variable
\prompt [TEXT] NAME  Prompt user for variable value
\getenv psql_var env_var   Copy environment variable into psql variable
\setenv name [value]       Set or unset environment variable
\p                  Print current query buffer
\w filename         Write query buffer to file (or pipe with |cmd)

-- Conditional execution (useful in scripts)
\if EXPR
  \echo 'true branch'
\else
  \echo 'false branch'
\endif

\elif EXPR           Else-if inside \if block

\if and \elif evaluate their argument as a boolean. Valid values (case-insensitive, unambiguous prefix matching): true, false, 1, 0, on, off, yes, no. Expressions that don't evaluate to true/false generate a warning and are treated as false. Variable references in skipped lines are NOT expanded.

Variables in SQL: :'varname' (quoted string value, escapes embedded quotes), :"varname" (double-quoted identifier), :'varname'::type (with cast), :varname (unquoted — can break SQL), :{?varname} (tests existence, expands to TRUE/FALSE).

Session Management

\c [dbname [user]]  Connect to database (or reconnect)
\conninfo           Display connection info (includes SSL info)
\encoding [ENC]     Set or show client encoding
\password [USER]    Change password (does NOT appear in command history or server log)
\q                   Quit psql. In a script file, only that script is terminated. In interactive mode, the entire program exits.
\r                   Reset (clear) the query buffer
\e                   Edit query buffer in external editor
\ef [FUNCNAME]       Edit function definition
\ev [VIEWNAME]       Edit view definition
\sf[+] FUNCNAME      Show function definition (read-only)
\sv[+] VIEWNAME      Show view definition (read-only)
\s [FILE]            Print command history (or save to file)
\restrict KEY        Enter restricted mode (only \unrestrict allowed)
\unrestrict KEY      Exit restricted mode
\timing [on\|off]    Toggle query execution time display (milliseconds)
\errverbose          Repeat last error at maximum verbosity
\? [topic]           Help: commands, options, or variables
\h [command]         SQL syntax help (use * for all: \h *)
\copyright           Show PostgreSQL copyright

Pipeline Mode (PostgreSQL 14+)

\startpipeline
  SELECT $1 \bind 42 \sendpipeline
  SELECT $1 \bind 100 \sendpipeline
  \getresults
\endpipeline

Pipeline mode sends multiple queries without waiting for each result, reducing round-trip latency. All queries use the extended query protocol.

Pipeline commands:

• \startpipeline — begin pipeline block
• \endpipeline — end pipeline block and process remaining results
• \sendpipeline — append current query buffer to pipeline without waiting
• \syncpipeline — send sync message without ending pipeline
• \flushrequest — request server flush without sync
• \flush — manually push unsent data to server
• \getresults [N] — read pending results (N=0 or omitted means all)

Pipeline limitations:

• COPY is not supported in pipeline mode
• Meta-commands like \g, \gx, \gdesc are not allowed inside a pipeline
• All queries use the extended query protocol
• Use \bind, \bind_named, \parse, \close_prepared, or \sendpipeline within pipelines
• A %P prompt variable shows pipeline status (on, off, or abort)

\watch Syntax

\watch [i[nterval]=SECONDS] [c[ount]=TIMES] [m[in_rows]=ROWS] [SECONDS]

count and min_rows require PostgreSQL 17+.

• interval — seconds between executions (default: 2, overridable via WATCH_INTERVAL variable)
• count — stop after N executions
• min_rows — stop if query returns fewer than N rows

If the query buffer is empty, \watch re-executes the most recently sent query.

Examples:

SELECT * FROM pg_stat_activity WHERE state = 'active';
\watch interval=5 count=10      -- every 5s, stop after 10 runs

SELECT count(*) FROM queue WHERE status = 'pending';
\watch i=1 min_rows=1            -- every 1s, stop when queue is empty

Exit Codes

Security Considerations

Destructive Operations Checklist

Before running any destructive SQL, verify impact first:

-- BEFORE DELETE: check how many rows are affected
SELECT count(*) FROM users WHERE condition;  -- verify scope
BEGIN;
DELETE FROM users WHERE condition RETURNING *;  -- see what was deleted
-- ROLLBACK if wrong; COMMIT only after verification

-- BEFORE DROP TABLE: verify no foreign keys depend on it
\d table_name  -- check "Referenced by" section
-- Consider renaming first: ALTER TABLE old RENAME TO old_backup;

Dangerous Commands Requiring Extra Caution

Variable Interpolation Safety

psql variables are plain text substitution, not parameterized queries. This means:

-- UNSAFE: if :name contains "Robert'); DROP TABLE users;--" it will execute the injection
SELECT * FROM users WHERE name = :'name';

-- SAFER: use \prompt for interactive input (user sees what they typed)
\prompt 'Enter name: ' search_name
SELECT * FROM users WHERE name = :'search_name';

-- SAFEST: use \bind for programmatic parameter passing (truly parameterized)
SELECT * FROM users WHERE name = $1;
\bind 'Robert' \g

The :'varname' form (quoted) is always safer than :varname (unquoted), because unquoted substitution can break SQL syntax or enable injection. Use :"varname" for identifiers (table/column names) — it properly escapes embedded double quotes.

When to Use What

• references/meta-commands-core.md — Core meta-commands: query buffer behavior, argument parsing rules, connection management, query execution, \copy syntax, and scripting commands (\if, \i, \o, backquote expansion). Read this when you need exact syntax or behavioral details for any backslash command.
• references/meta-commands-inspection.md — Full \d command reference: all object inspection commands, modifiers (S, +, x), and pattern matching rules. Read this when exploring database schema or when the user needs to inspect tables, indexes, views, functions, privileges, etc.
• references/meta-commands-formatting.md — Output formatting (\pset options and all format descriptions), pipeline mode, \watch, \crosstabview, and session management (\e, \ef, \ev, \timing, etc.). Read this when the user needs to control output format or use pipeline mode.
• references/cli-options-and-variables.md — All CLI flags, environment variables, psql internal variables (AUTOCOMMIT, ON_ERROR_STOP, ECHO, FETCH_COUNT, etc.), prompt customization, ~/.psqlrc configuration, and SQL interpolation syntax. Read this when configuring psql startup behavior, writing scripts that depend on variable state, or customizing prompts.
• references/tips-workflows.md — Practical workflows (exploring a new database, understanding table structure), scripting patterns (safe scripts, conditional execution, \gexec), output control for automation, and data import/export patterns. Read this when the user asks how to accomplish a specific task with psql.
• references/tips-advanced.md — Performance tips, debugging/introspection (EXPLAIN, lock analysis, ECHO_HIDDEN), safety best practices (ON_ERROR_STOP, transaction patterns, search_path safety), and common gotchas. Read this for lock analysis, query plan inspection, and troubleshooting.

Important Notes

psql handles two comment styles differently:

• C-style block comments (/* ... */): Passed to the server for processing and removal.
• SQL-standard comments (--): Removed by psql itself, before sending to the server.

This distinction matters when writing scripts that rely on comment behavior — only SQL-standard comments are stripped client-side.

Variable Variables (Soft References)

psql allows indirect variable references through \set:

\set foo 'my_table'
\set bar :foo         -- copies the value of foo into bar
\echo :bar            -- outputs: my_table

While constructs like \set :foo 'something' are syntactically valid, they produce "soft links" that have limited practical use. For straightforward variable copying, use \set new_var :old_var.

Version Compatibility

psql works best with servers of the same or an older major version. Backslash commands (especially \d family) may fail with newer server versions. When connecting to multiple server versions, use the newest available psql client. The \d commands generally work with servers back to version 9.2.

External References

• PostgreSQL Client Applications
• Official PostgreSQL Documentation
• The SQL Language
• SQL Syntax - The SQL Language
• SQL Command
• PostgreSQL Wiki

Limitations

• Use this skill only when the task clearly matches its upstream source and local project context.
• Verify commands, generated code, dependencies, credentials, and external service behavior before applying changes.
• Do not treat examples as a substitute for environment-specific tests, security review, or user approval for destructive or costly actions.

Source: sickn33/agentic-awesome-skills → skills/postgresql-cli/SKILL.md

Also appears in: sickn33/agentic-awesome-skills/plugins/agentic-awesome-skills/skills/postgresql-cli/SKILL.md, sickn33/agentic-awesome-skills/plugins/agentic-awesome-skills-claude/skills/postgresql-cli/SKILL.md