Commands

Skiller adds a chat participant, @skiller, that runs declarative workflows. You drive it with slash commands: address @skiller, then type a command. A message that is neither a slash command nor a skill launch gets a short hint pointing you back to the commands below — @skiller does not do free-form chat.

You can also launch a skill straight from the editor — a code action, the editor context menu, or the Command Palette — without typing the command yourself. See Launching from the editor.

All commands

Command	What it does	Notes
/help	Lists the available commands.	Start here if you forget a command.
/skills	Lists every discovered skill with its id, name, and source tier.	Source is workspace, user, or built-in.
/skill <id> [params]	Runs the skill with that id.	Accepts named (name=Ada) and positional launch arguments — see below.
/tools	Lists the tools available to skills.	Includes the built-in file tools and any configured MCP tools.
/models	Lists the language models available for skill configuration.	The ids you reference in models.default / models.aliases.
/reload	Re-scans skills and tools and reports what changed.	Shows an added / removed / parse-error diff (see below).
/status	Reports whether a skill is awaiting input or confirmation.	Control command — works mid-skill.
/cancel	Aborts the skill currently awaiting input or confirmation.	Control command — works mid-skill.
/reset	Clears all Skiller state — pending interactions and graph highlights.	Control command — the blunt recovery hatch.

Launching a skill: /skill <id> [params]

/skill <id> runs a discovered skill. Any input the skill declares but you do not supply on the command line is collected interactively before the skill starts.

You can pass launch arguments two ways, and mix them:

• Named — name=value. Matches the input by its declared name.
• Positional — a bare value. Fills the remaining inputs in declaration order.

@skiller /skill greeter name=Ada
@skiller /skill greeter "Ada"

Both forms set the name input to Ada. Quote a value that contains spaces.

Launching from the editor

Skills don’t have to be typed into chat. Skiller registers a VS Code command, skiller.runSkill, reachable three ways:

Surface	How
Command Palette	Skiller: Run Skill → pick a skill from the list
Editor context menu	Right-click in a file → Run Skill
Code action	The lightbulb / quick-fix menu, for skills that match the current file or selection

Either way, Skiller captures the editor state at that moment — the selection, file, diff, or diagnostics — so a skill’s from: inputs are filled from where you launched, and an output.to sink writes back there even after chat takes focus. It then opens chat to run the skill; the skiller.skills.runSurface setting controls whether the command is prefilled for you to submit (adaptive, default) or submitted immediately (chat).

See Editor-native skills for the whole model.

The /reload diff

/reload re-scans all three discovery tiers and the configured tools, then prints what changed rather than a flat list:

• Added — skills or tools that now exist.
• Removed — skills or tools that disappeared (their graph highlights are cleared; a pending interaction tied to a removed skill is dropped).
• Parse errors — skills whose skill.yaml now fails validation (and any that were previously broken and are now fixed).

Use it after editing a manifest on disk so the registry and any open live execution graph panels pick up the change.

Control commands work mid-skill

/status, /cancel, and /reset are control commands. They are dispatched before the pending-interaction gate, so they always run — even while a skill is paused waiting for input or a confirmation. Without this, a wedged pending state could block its own escape hatch.

• /status — tells you whether something is paused, and at which step.
• /cancel — aborts only the skill awaiting input or confirmation.
• /reset — clears everything, when state is thoroughly stuck.

Pending state is isolated per conversation: a skill paused in one chat will not consume a turn you send in a different chat. Use /cancel or /reset from the chat that started the skill.

No free-form chat

Skiller is skills-only by design. A message to @skiller that is not a recognized command and does not launch a skill returns a short hint listing /skills, /skill <name>, and /help — it does not attempt an open-ended conversation.

Next

• Run a built-in skill — try /skills and /skill greeter end to end.
• Settings — tune verbose output and the agentic-loop limits.