Give bots skills: runbooks they load when a job matches

A skill is a runbook: a name, a unique slug, a one-line description of when
it applies, and the markdown body. A run only ever sees the slug and
description (one load_skill tool carries them), and reads the whole body
only when it calls the tool, so a long runbook costs nothing until needed.

Skills are the third extension kind beside MCP servers and custom tools, on
the same botExtensionsSql path, so per-bot, per-workspace and all-bots
grants work the same way, and attachDefaultExtensions hands an every-bot
skill to new bots. load_skill is a reserved tool name, so a custom tool
cannot shadow it, and a built-in wins a clash.

Managed on Settings > Skills; ExtensionRow links each kind to its own page
and a skill never reports a failure, since nothing connects or runs.
This commit is contained in:
2026-10-22 12:31:42 +01:00
parent 379ef35e08
commit 08d02e6033
62 changed files with 1438 additions and 79 deletions
@@ -17,8 +17,9 @@ type Props = {
};
/**
* One MCP server or tool the bot can use here, in the composer's tools
* modal. One that isn't working says so, one that's off is struck out.
* One MCP server, tool or skill the bot can use here, in the composer's
* tools modal. One that isn't working says so, one that's off is struck
* out. A skill has nothing to connect or run, so it never reports one.
*/
export default function ComposerToolRow({ item, busy, onRemove }: Props) {
const Icon = extensionIcon({ kind: item.kind });
@@ -27,11 +28,13 @@ export default function ComposerToolRow({ item, busy, onRemove }: Props) {
item.source == "workspace" ? "From this workspace" : "",
!item.enabled
? "Off for every bot"
: item.state == "error"
? "Not working"
: item.state == "untested"
? "Not checked yet"
: "",
: item.kind == "skill"
? ""
: item.state == "error"
? "Not working"
: item.state == "untested"
? "Not checked yet"
: "",
].filter(Boolean);
return (