capabilities

NAME

capabilities — complete capability reference for SLICC agents

SYNOPSIS

Run commands for the full executable list, <cmd> --help for usage, or man <topic> for in-depth docs on any topic below.

BROWSER AUTOMATION

playwright-cli (aliases: playwright, puppeteer) — full browser control via Chrome DevTools Protocol.

Navigate: open, goto, go-back, go-forward, reload
Inspect: snapshot (accessibility tree with element refs), screenshot, eval <js>, eval-file <path>
Interact: click, fill, type, press, hover, select, check, drag
Tabs: tab-list, tab-new, tab-close. Requires --tab=<id> in scoops/.jsh.
Storage: cookie-*, localstorage-*, sessionstorage-*
Recording: record [url] [--filter], stop-recording <id> — HAR capture
Teleport: teleport --start <regex> --return <regex> — auth handoff to follower

See man playwright-cli for the full reference.

MULTI-RUNTIME (TRAYS)

Control tabs across multiple devices and runtimes — local browser, remote CLI, iOS app, Chrome extension.

host — show leader status, join URL, connected followers
host reset — disconnect all followers, fresh session
playwright-cli open <url> --runtime=<id> — open tab on remote tray
playwright-cli tab-list — shows tabs across all runtimes
playwright-cli teleport --list — list follower runtimes
rsync <src> <dest> — sync files between local and remote VFS over tray

Remote tabs appear as [follower-xxx:targetId] in tab-list. All playwright commands work on remote tabs. See man tray, man rsync.

FILES AND MOUNTS

mount <path> — local directory picker
mount --source s3://bucket /mnt/s3 — S3/R2/MinIO
mount --source da://org/repo /mnt/da — Adobe DA (da.live)
mount list, mount unmount, mount refresh
fswatch create --path <p> --pattern <glob> --scoop <name> — file change events as licks
ln -s — symlinks work transparently

MEDIA AND DOCUMENTS

convert / magick — ImageMagick (WASM): resize, crop, rotate, format conversion
pdftk / pdf — PDF merge, split, rotate, extract text
pptx-lib — create PowerPoint files programmatically
afplay <file> — play audio (Web Audio API)
say -l <lang> "<text>" — text-to-speech (Web Speech API)
chime — notification sound
open --view <path> — return image inline for agent vision
imgcat <path> — preview image/video in browser tab
screencapture <path> — capture user's screen (requires permission)

AUTH AND SECRETS

oauth-token <provider> — fresh OAuth access token (github, adobe, etc.). Auto-login if needed.
oauth-token --list — show available providers and status
secret set <name> <value> --domain <patterns> — credentials for fetch proxy / mounts
secret list, secret delete, secret test

See man oauth-token.

AUTOMATION (LICKS)

webhook create --scoop <name> [--filter <js>] — HTTP endpoint → scoop
crontask create --name <n> --scoop <name> --cron "<expr>" — scheduled → scoop
fswatch create --path <p> --pattern <glob> --scoop <name> — file changes → scoop

All three deliver structured lick events. See man lick, man webhook, man crontask.

UI

sprinkle open/close/send/list/chat — persistent panels. See man sprinkle.
sprinkle chat '<html>' — blocking inline card, waits for user click, returns action JSON
Inline ```shtml blocks in chat — render interactive widgets
serve <dir> — serve VFS directory in browser tab. See man serve.
open <path> — open file in tab. -d downloads. -v returns for vision. See man open.

AGENTS AND DELEGATION

agent <cwd> <allowed-cmds> <prompt> — one-shot sub-agent, blocks, returns stdout
scoop_scoop / feed_scoop / drop_scoop — persistent conversational scoops

See man delegation for strategy, man scoop for lifecycle.

SKILLS

skill list — installed skills
skill read <name> — full instructions
upskill search "<query>" — find skills on ClawHub + Tessl
upskill <owner/repo> --all — install from GitHub

See man skill.

DATA AND CODE

Languages: node, python3, sqlite3
Version control: git
Network: curl, html-to-markdown
Data: jq, base64
Text: grep, rg, sed, awk, sort, uniq, diff, wc, cut, tr

SESSION

models — list available LLMs and current selection
cost — session cost breakdown by agent
nuke — full environment reset (destructive, requires launch code)

CONSTRAINTS

No package managers (apt, npm install, pip). Tools are built-in or installable as skills.
No long-running servers. serve handles previewing via service worker.
fetch()/curl route through browser Fetch API — Cookie headers stripped, Origin is localhost. For cookie-based APIs, use playwright-cli eval from page context.
screencapture needs user permission (may require browser restart on macOS).