Drush Token Killer
This module provides compressed output for Drush commands to reduce token usage by AI agents. It allows you to selectively enable these optimizations on a per-command, per-session, or permanent basis, without altering the output for standard terminal use. Compression works by modifying existing commands to output fewer fields and use CSV formatting, significantly shrinking the data returned.
Drush DTK - the "Drush Token Killer" - provides token-saving output defaults for common Drush commands. Inspired by RTK (the "Rust Token Killer"), it shrinks the output of frequently-run, read-only Drush commands so that AI agents (and humans) spend far fewer tokens reading them.
Opt-in compression. Enabling the module changes nothing for humans at a terminal - you (or your agent) switch it on per command, per session, or permanently. See "Enabling compression" below.
It works with lightweight Drush hooks that, when you have not asked for specific output yourself, select a compact set of fields and force CSV to drop the wide ASCII table chrome. No commands are added for everyday use - the module only alters the output of existing core and contrib Drush commands.
How it works
Each supported command has an INITIALIZE hook that, when compression is enabled, sets sensible defaults (--fields and/or --format=csv) before the command runs. config:status also has an ALTER_RESULT hook that rewrites its verbose state strings to short tokens. The original commands and their behaviour are otherwise untouched.
Enabling compression
Three opt-in surfaces, highest precedence first:
--ai-compressflag - one command, e.g.drush pml --ai-compressDTK_COMPRESS=1environment variable - one session; recommended for agents, since they invoke Drush through wrappers where injecting flags is awkward- Config -
drush config:set dtk.settings compress 1for always-on
DTK_COMPRESS=0 (also false, no, off) explicitly disables compression, overriding the config. --no-ai-compress on any command beats everything and restores native output. There is deliberately no TTY or "am I an agent?" detection - output only changes when something asked for it.
Agent hooks
One command wires DTK_COMPRESS=1 into an agent's project-level config:
drush dtk:install claude # merges into .claude/settings.json drush dtk:install codex # adds to .codex/config.toml drush dtk:install opencode # prints the manual step drush dtk:install copilot # prints the manual step
Compressed commands
Defaults applied when compression is enabled:
Command Default applied Savingspm:list (pml)
--status=<wbr>enabled|<wbr>disabled → --fields=<wbr>name; otherwise --fields=<wbr>name,<wbr>status; CSV
up to 90%
core:requirements (rq)
--severity=1 (warnings + errors only), --fields=<wbr>title,<wbr>severity, CSV
up to 97%
config:status (cst)
CSV + state tokens (see below)
up to 45%
watchdog:show (ws)
--fields=<wbr>type,<wbr>severity,<wbr>message, CSV
up to 50%
role:list (rls)
--fields=<wbr>rid,<wbr>label, CSV (native default is YAML)
up to 50%
queue:list
CSV
small
migrate:status (ms)
CSV
varies
migrate:messages (mmsg)
--fields=<wbr>level,<wbr>source_ids,<wbr>message, CSV
up to 40%
field:info (fi)
CSV
varies
field:base-info (fbi)
--fields=<wbr>field_name,<wbr>field_type,<wbr>required,<wbr>cardinality, CSV
up to 70%
field:types
--fields=<wbr>id,<wbr>label (native YAML kept — result can't render as CSV)
up to 80%
field:formatters
--fields=<wbr>id,<wbr>label,<wbr>field_types (native YAML kept — result can't render as CSV)
up to 70%
field:widgets
--fields=<wbr>id,<wbr>label,<wbr>field_types (native YAML kept — result can't render as CSV)
up to 70%
updatedb:status (updbst)
--fields=<wbr>module,<wbr>update_id,<wbr>description, CSV
up to 50%
user:information (uinf)
--fields=<wbr>uid,<wbr>name,<wbr>mail,<wbr>user_status, CSV
up to 60%
deploy:hook-status
CSV
small
Contrib module compressed commands
Compressed only when the owning module is installed; with it absent the hook never fires. The Devel rows also require Drush 13.7+ (see Requirements).
devel:services (dcs)
Devel
CSV (single id column)
~67%
devel:token (token)
Devel
--fields=group,token, CSV
~85%
search-api:list (sapi-l)
Search API
--fields=id,serverName,status, CSV
Varies
search-api:status (sapi-s)
Search API
--fields=id,complete,indexed,total, CSV
Varies
search-api:server-list (sapi-sl)
Search API
--fields=id,status, CSV
Varies
search-api:search (sapi-search)
Search API
CSV
Small
views:bulk-operations:list (vbo-list)
Views Bulk Operations
--fields=id,entity_type_id, CSV
Small
config:status state tokens
The verbose state strings are collapsed to single tokens:
The --state filter still works with the original strings (e.g. drush config:status --state='Only in DB'), because the tokenisation runs after the command has already filtered.
Opting out
Even when compression is enabled, add --no-ai-compress to any command to get its native, uncompressed output:
drush pml --no-ai-compressCompression is also skipped automatically whenever you choose your own output, i.e. you pass --field or --fields (and, for format, any --format other than the command's native default).
Limitations
- Pre-bootstrap commands cannot be compressed. Commands that run before Drupal's modules load - notably
core:statusandcache:rebuild(cr) - never receive this module's hooks.core:status --no-ai-compresseven reports "option does not exist". This is inherent to module-provided Drush hooks. config:import/config:exportpreviews are not reformatted. Those commands print their change preview as a hardcoded Symfony Console table straight to the terminal, mid-command, before the confirmation prompt - there is no formatter hook to retarget it. To preview pending config changes as CSV, useconfig:statusinstead (itsonly_sync/only_db/differenttokens are the same information framed as create / delete / update).
Requirements
- Drupal ^11
- Drush 13+
Installation
drush en dtk drush cr
Contributing
The Drush Token Killer module's issue queue is intended for reporting bugs, submitting feature requests, and planning.
Additionally, you can join the #ai-contrib or #ai-learners channels on Drupal Slack.