{"id":"5c2b7e5e-3874-4aad-8863-24867b2a33c6","slug":"behaviour-studio","category_slug":"workers","order_index":103,"published":true,"created_at":"2026-06-19T14:24:50.512Z","related_slugs":"[\"create-workers\",\"rules-for-workers\",\"workers-integrations\"]","title_fr":"Compétences, Behavior Studio & Discover","title_en":"Skills, Behavior Studio & Discover","content_en":"<p>For a skill to use in the Assistant without a Worker, see <a href=\"/en/articles/create-library-skill\">Create a skill in the Library</a>.</p>\n\n\n<blockquote>\n<p><strong>The right picture:</strong> if the Worker is a teammate, the <strong>skills</strong> are its procedure sheets, the <strong>initiative</strong> is its temperament (wait vs take the lead), and <strong>discover</strong> is the moment it \"pulls the right tool out of its box\" when it needs it.</p>\n</blockquote>\n\n<h2>Skills</h2>\n<p>A <strong>skill</strong> is a reusable unit of know-how attached to the Worker. It combines:</p>\n<ul>\n<li>a <strong>title</strong> (e.g. <em>Reply to an inbound support ticket</em>)</li>\n<li><strong>content</strong> (the procedure, the rules, examples — rich text)</li>\n<li>associated <strong>integrations</strong> (the tools useful to this skill)</li>\n<li>reference <strong>documents</strong> and <strong>links</strong></li>\n<li><strong>manager notes</strong> (context for the team)</li>\n</ul>\n<h3>How to create one</h3>\n<ol class=\"steps\">\n<li>Open the Worker's page → <strong>Behavior Studio</strong> tab</li>\n<li>Click <strong>Add a skill</strong></li>\n<li>Two modes:<ul>\n<li><strong>Assisted mode</strong> (default): you describe the skill in natural language, the AI drafts it</li>\n<li><strong>Manual mode</strong>: you write the title and content yourself</li>\n</ul>\n</li>\n<li>In the full-screen skill editor, browse the file tree on the left (<strong>SKILL.md</strong>, <strong>data</strong>, <strong>examples</strong>, <strong>references</strong>, <strong>templates</strong>), edit <strong>SKILL.md</strong> on the right, fill in <strong>Skill trigger</strong>, review <strong>Tests</strong>, then click <strong>Save version</strong>.</li>\n</ol>\n<figure class=\"screenshot\"><img src=\"/docs_mcp/worker_skill_editor.png\" alt=\"Full-screen skill editor: SKILL.md, data, examples, references and templates tree on the left; SKILL.md content, Skill trigger, Save version and Tests on the right\"><figcaption>Full-screen skill editor: SKILL.md, data, examples, references and templates tree on the left; SKILL.md content, Skill trigger, Save version and Tests on the right</figcaption></figure><p><em>Skill editor: file tree on the left, SKILL.md and Skill trigger on the right, with Save version and Tests.</em></p>\n<blockquote>\n<p><strong>What happens under the hood:</strong> a skill <strong>is not code</strong> nor an executable tool. It's a knowledge document <strong>injected into the Worker's context</strong>. The Worker \"reads\" it when working in order to follow the right procedure.</p>\n</blockquote>\n<blockquote>\n<p><strong>Limit:</strong> up to <strong>50 skills</strong> per Worker. A few clear skills beat fifty vague ones.</p>\n</blockquote>\n\n<h2>Initiative (the autonomy level)</h2>\n<p>The <strong>initiative</strong> defines how much the Worker acts on its own. Four levels:</p>\n<table>\n<thead>\n<tr>\n<th>Level</th>\n<th>Worker behavior</th>\n</tr>\n</thead>\n<tbody><tr>\n<td><strong>Passive</strong></td>\n<td>Acts only on an explicit request. Drafts, but never sends.</td>\n</tr>\n<tr>\n<td><strong>Reactive</strong></td>\n<td>Responds to clear triggers. Asks before committing.</td>\n</tr>\n<tr>\n<td><strong>Proactive</strong></td>\n<td>Surfaces next steps. Confirms before high-impact actions.</td>\n</tr>\n<tr>\n<td><strong>Autonomous</strong></td>\n<td>Decides and acts, unless told otherwise.</td>\n</tr>\n</tbody></table>\n<p>Set this level in the Worker’s <strong>Agent</strong> tab, under <strong>Tone &amp; style</strong>, by choosing <strong>Passive</strong>, <strong>Reactive</strong>, <strong>Proactive</strong> or <strong>Autonomous</strong>.</p>\n<figure class=\"screenshot\"><img src=\"/docs_mcp/worker_initiative_levels.png\" alt=\"Worker Agent tab, Tone &amp; style section: initiative choices Passive, Reactive, Proactive and Autonomous\"><figcaption>Worker Agent tab, Tone &amp; style section: initiative choices Passive, Reactive, Proactive and Autonomous</figcaption></figure><p><em>Initiative levels in Agent &gt; Tone &amp; style: Passive, Reactive, Proactive and Autonomous.</em></p>\n<blockquote>\n<p><strong>Under the hood:</strong> the initiative level is baked directly into the Worker's system instructions. It influences each of its \"do I act or do I ask?\" decisions.</p>\n</blockquote>\n\n<h2>Rules (guardrails)</h2>\n<p>In addition to initiative, <strong>rules</strong> set precise limits per action type: <em>always</em>, <em>ask</em> or <em>never</em>. They're detailed in the dedicated <strong>Rules &amp; guardrails</strong> article — just remember here that they live alongside skills in the Behavior Studio.</p>\n\n<h2>\"Discover\": how the Worker activates a tool</h2>\n<p>This is the most confusing mechanism on first contact, so let's clarify it.</p>\n<p>The Worker does <strong>not</strong> load all the tools of all your integrations permanently — that would be heavy and confusing. Instead:</p>\n<ol class=\"steps\">\n<li>The Worker sees the <strong>list of connected integrations</strong> (e.g. Slack is connected).</li>\n<li>When a task requires it, it <strong>\"discovers\" the corresponding skill</strong> — internally, it calls a tool such as <code>discover_slack_skill</code>.</li>\n<li>This discover <strong>checks that the connection exists</strong>:<ul>\n<li>if it's missing → message <em>\" Integration not connected. Connect it in Settings &gt; Integrations.\"</em></li>\n<li>if it's there → the skill <strong>activates</strong> and the real tools (<code>slack_post_message</code>, etc.) become available.</li>\n</ul>\n</li>\n<li>From there, the Worker can actually act on the tool.</li>\n</ol>\n<figure class=\"screenshot\"><img src=\"/docs_mcp/worker_discover_activity.png\" alt=\"Worker Activity tab: Today log listing the Worker’s actions with time and status\"><figcaption>Worker Activity tab: Today log listing the Worker’s actions with time and status</figcaption></figure><p><em>Activity &gt; Today: each Worker action is listed with its time and status.</em></p>\n<blockquote>\n<p><strong>What you see:</strong> the Worker’s <strong>Activity</strong> tab lists its actions under <strong>Today</strong>, each with a time and status.</p>\n</blockquote>\n<blockquote>\n<p> <strong>Common blocker:</strong> if the Worker replies <em>\"this integration is not connected\"</em> when you thought you had connected it, it's the discover failing. Check in the <strong>Integrations</strong> tab that the connection is indeed <strong>active</strong> (toggle on) — see the <em>Integrations</em> article.</p>\n</blockquote>\n\n<h2>Skill vs integration vs tool vs rule — the right mental model</h2>\n<table>\n<thead>\n<tr>\n<th>Concept</th>\n<th>What it is</th>\n<th>What you do</th>\n</tr>\n</thead>\n<tbody><tr>\n<td><strong>Skill</strong></td>\n<td>A know-how document (a \"playbook\")</td>\n<td>You write it in the Behavior Studio</td>\n</tr>\n<tr>\n<td><strong>Integration</strong></td>\n<td>A connection to a third-party service (OAuth / API key)</td>\n<td>You connect it in the Integrations tab</td>\n</tr>\n<tr>\n<td><strong>Discover</strong></td>\n<td>The moment the Worker activates an integration at runtime</td>\n<td>You do nothing — you see it in the log</td>\n</tr>\n<tr>\n<td><strong>Tool</strong></td>\n<td>A concrete action (send a Slack message…)</td>\n<td>Invisible: the Worker calls it once the integration is activated</td>\n</tr>\n<tr>\n<td><strong>Rule</strong></td>\n<td>A behavior limit (always / ask / never)</td>\n<td>You set it in the Behavior Studio</td>\n</tr>\n</tbody></table>\n<blockquote>\n<p><strong>Remember:</strong> you configure <strong>skills</strong> (knowledge) and <strong>integrations</strong> (access); the Worker takes care, on its own, of <strong>discovering</strong> and calling the right <strong>tools</strong> at the right time, within its <strong>rules</strong>.</p>\n</blockquote>","tags_fr":null,"tags_en":null,"intro_fr":"Le Behavior Studio est l'endroit où tu façonnes comment ton Worker travaille : ce qu'il sait faire (compétences), à quel point il agit seul (initiative) et ce qu'il a le droit de faire (règles).","intro_en":"The Behavior Studio is where you shape how your Worker works: what it can do (skills), how much it acts on its own (initiative) and what it's allowed to do (rules).","prerequisites_fr":null,"prerequisites_en":null}