@cursor/july 0.1.47 → 0.1.49

package/dist/docs/assets/{storage.md.CJYd5YmN.js → storage.md.D8NQ389l.js}

@@ -1,4 +1,4 @@
-import{_ as e,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const c=JSON.parse(`{"title":"Storage","description":"Point the Agent SDK's durable storage at a backend you own with defineStorage.","frontmatter":{"title":"Storage","description":"Point the Agent SDK's durable storage at a backend you own with defineStorage."},"headers":[],"relativePath":"storage.md","filePath":"storage.md"}`),n={name:"storage.md"};function o(r,s,d,h,l,p){return a(),t("div",null,[...s[0]||(s[0]=[i(`<h1 id="storage" tabindex="-1">Storage <a class="header-anchor" href="#storage" aria-label="Permalink to &quot;Storage&quot;">​</a></h1><p>The Agent SDK owns durable storage for sessions, continuation tokens, reminders, playground eval history, and live A/B samples. It chooses the keys (under <code>agentkit/v1/</code>), when to read and write, and how to restore after restart.</p><p>Keys have bounded length: caller-controlled segments (channel ids, continuation tokens) are URI-encoded, and any segment past 256 encoded bytes is replaced by its <code>sha256:…</code> digest — deterministically, so writes and lookups always agree. Backends can rely on this instead of imposing their own key-length caps (which would silently drop writes, since a throwing <code>put</code> is at-most-once).</p><p>By default that storage lives under <code>--state-root</code> on local disk. Fine for one machine; it does not survive replacing the host.</p><p>To keep the same framework storage across hosts, plug in a key-value backend with <code>agent/storage.ts</code>. You provide <code>put</code> / <code>get</code> / <code>delete</code> / <code>list</code>, plus the <code>cas</code> group when the deployment uses coordination features (conditional writes — see <a href="#conditional-writes-the-cas-group">Conditional writes</a>). The Agent SDK does the rest.</p><div class="language-ts vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">ts</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">// agent/storage.ts</span></span>
+import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const k=JSON.parse(`{"title":"Storage","description":"Point the Agent SDK's durable storage at a backend you own with defineStorage.","frontmatter":{"title":"Storage","description":"Point the Agent SDK's durable storage at a backend you own with defineStorage."},"headers":[],"relativePath":"storage.md","filePath":"storage.md"}`),n={name:"storage.md"};function o(r,e,d,l,h,p){return a(),t("div",null,[...e[0]||(e[0]=[i(`<h1 id="storage" tabindex="-1">Storage <a class="header-anchor" href="#storage" aria-label="Permalink to &quot;Storage&quot;">​</a></h1><p>The Agent SDK owns durable storage for sessions, continuation tokens, reminders, playground eval history, and live A/B samples. It chooses the keys (under <code>agentkit/v1/</code>), when to read and write, and how to restore after restart.</p><p>Keys have bounded length: caller-controlled segments (channel ids, continuation tokens) are URI-encoded, and any segment past 256 encoded bytes is replaced by its <code>sha256:…</code> digest — deterministically, so writes and lookups always agree. Backends can rely on this instead of imposing their own key-length caps (which would silently drop writes, since a throwing <code>put</code> is at-most-once).</p><p>By default that storage lives under <code>--state-root</code> on local disk. Fine for one machine; it does not survive replacing the host.</p><p>To keep the same framework storage across hosts, plug in a key-value backend with <code>agent/storage.ts</code>. You provide <code>put</code> / <code>get</code> / <code>delete</code> / <code>list</code>, plus the <code>cas</code> group when the deployment uses coordination features (conditional writes — see <a href="#conditional-writes-the-cas-group">Conditional writes</a>). The Agent SDK does the rest.</p><div class="language-ts vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">ts</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">// agent/storage.ts</span></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">import</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> { defineStorage } </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">from</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &quot;@cursor/july/storage&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">;</span></span>
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">export</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> default</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;"> defineStorage</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">({</span></span>
@@ -12,7 +12,7 @@ import{_ as e,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const c
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">    putIfVersion</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">: (</span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">key</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">, </span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">value</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">, </span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">expected</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">) </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">=&gt;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> db.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">casUpdate</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(key, value, expected),</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">    listKeys</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">: (</span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">prefix</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">) </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">=&gt;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> db.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">listKeysByPrefix</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(prefix),</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">  },</span></span>
-<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><div class="note custom-block github-alert"><p class="custom-block-title">NOTE</p><p>Import paths here use <code>@cursor/july/storage</code>. On projects still using <code>@anysphere/agent-serve</code>, swap the import. See <a href="/docs/#run-the-cli">Run the CLI</a> for the full rename table.</p></div><h2 id="which-fields-to-provide" tabindex="-1">Which fields to provide <a class="header-anchor" href="#which-fields-to-provide" aria-label="Permalink to &quot;Which fields to provide&quot;">​</a></h2><table tabindex="0"><thead><tr><th>Field</th><th>Required</th><th>Role</th></tr></thead><tbody><tr><td><code>put</code></td><td>Yes</td><td>Write or update a value</td></tr><tr><td><code>cas</code></td><td>For coordination</td><td>Conditional writes; see <a href="#conditional-writes-the-cas-group">Conditional writes</a></td></tr><tr><td><code>get</code></td><td>For restore</td><td>Look up one key</td></tr><tr><td><code>list</code></td><td>For restore</td><td>Return entries under a prefix, in key order</td></tr><tr><td><code>delete</code></td><td>For cleanup</td><td>Remove a key</td></tr><tr><td><code>name</code></td><td>No</td><td>Label surfaced on <code>GET /v1/info</code> diagnostics</td></tr><tr><td><code>policy</code></td><td>No</td><td>Timing knobs; see <a href="#policy">Policy</a></td></tr><tr><td><code>evals</code></td><td>No</td><td>Dedicated eval-runs table; see <a href="#eval-and-a-b-tables">Eval and A/B tables</a></td></tr><tr><td><code>abs</code></td><td>No</td><td>Dedicated A/B metrics table; see <a href="#eval-and-a-b-tables">Eval and A/B tables</a></td></tr></tbody></table><p>A throwing <code>put</code> is logged and dropped. It never fails a turn. When resolving a missing continuation token, a throwing <code>get</code> fails the follow-up so a store outage does not open a new session. Return <code>undefined</code> only for a real miss.</p><h2 id="eval-and-a-b-tables" tabindex="-1">Eval and A/B tables <a class="header-anchor" href="#eval-and-a-b-tables" aria-label="Permalink to &quot;Eval and A/B tables&quot;">​</a></h2><p>Two dedicated tables carry structured rows instead of opaque KV values. Both are optional and independent of <code>put</code> / <code>get</code> / <code>list</code>.</p><p><code>evals</code> keeps playground eval batches across restarts. Provide <code>put</code>, <code>delete</code>, and <code>list</code> over run snapshots keyed by <code>runId</code>. The Agent SDK upserts a snapshot as a batch starts, progresses, and finishes, prunes runs past the playground history window, and lists everything back at serve start. Without this table, eval history lives in process memory and a restart clears it. See <a href="./evals.html#configure-eval-runs">Evals</a>.</p><p><code>abs</code> exports live A/B metrics. Provide <code>putSample</code> to append one cumulative metric sample per enrolled experiment on each completed or failed turn. Optional <code>putSnapshot</code> and <code>getSnapshot</code> store and serve back the latest aggregate snapshot, so a replacement host with no local sessions can still serve the A/Bs surface. Without this table, samples only go where each experiment&#39;s <code>onSample</code> sends them; session event logs remain the assignment source of truth. See <a href="./ab.html">Live A/B metrics</a>.</p><h2 id="policy" tabindex="-1">Policy <a class="header-anchor" href="#policy" aria-label="Permalink to &quot;Policy&quot;">​</a></h2><p>Two knobs change behavior:</p><div class="language-ts vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">ts</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">export</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> default</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;"> defineStorage</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">({</span></span>
+<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><div class="note custom-block github-alert"><p class="custom-block-title">NOTE</p><p>Import paths here use <code>@cursor/july/storage</code>. On projects still using <code>@anysphere/agent-serve</code>, swap the import. See <a href="/docs/#run-the-cli">Run the CLI</a> for the full rename table.</p></div><h2 id="which-fields-to-provide" tabindex="-1">Which fields to provide <a class="header-anchor" href="#which-fields-to-provide" aria-label="Permalink to &quot;Which fields to provide&quot;">​</a></h2><p>Implement the small KV core — <code>put</code>/<code>get</code>/<code>delete</code>/<code>list</code> plus the <code>cas</code> group — and you get <strong>full functionality</strong>: eval-run and A/B history are derived over the core automatically. The dedicated <code>evals</code> / <code>abs</code> groups are backend-native optimizations, not required-or-lose-history hooks.</p><table tabindex="0"><thead><tr><th>Field</th><th>Required</th><th>Role</th></tr></thead><tbody><tr><td><code>put</code></td><td>Yes</td><td>Write or update a value</td></tr><tr><td><code>cas</code></td><td>For coordination</td><td>Conditional writes; see <a href="#conditional-writes-the-cas-group">Conditional writes</a></td></tr><tr><td><code>get</code></td><td>For restore</td><td>Look up one key (also: derived A/B snapshot backfill)</td></tr><tr><td><code>list</code></td><td>For restore</td><td>Return entries under a prefix, in key order (also: derived eval-runs hydrate)</td></tr><tr><td><code>delete</code></td><td>For cleanup</td><td>Remove a key (also: derived eval-runs pruning)</td></tr><tr><td><code>name</code></td><td>No</td><td>Label surfaced on <code>GET /v1/info</code> diagnostics</td></tr><tr><td><code>policy</code></td><td>No</td><td>Timing knobs; see <a href="#policy">Policy</a></td></tr><tr><td><code>evals</code></td><td>No</td><td>Backend-native eval-runs table; derived over the core when omitted — see <a href="#eval-and-a-b-tables">Eval and A/B tables</a></td></tr><tr><td><code>abs</code></td><td>No</td><td>Backend-native A/B metrics table; derived over the core when omitted — see <a href="#eval-and-a-b-tables">Eval and A/B tables</a></td></tr></tbody></table><p>A throwing <code>put</code> is logged and dropped. It never fails a turn. When resolving a missing continuation token, a throwing <code>get</code> fails the follow-up so a store outage does not open a new session. Return <code>undefined</code> only for a real miss.</p><h2 id="eval-and-a-b-tables" tabindex="-1">Eval and A/B tables <a class="header-anchor" href="#eval-and-a-b-tables" aria-label="Permalink to &quot;Eval and A/B tables&quot;">​</a></h2><p>Two dedicated table groups carry structured rows instead of opaque KV values. Both are <strong>optional optimizations</strong>: when a group is not authored, <code>defineStorage</code> derives it over the KV core, so a backend that implements only the core loses nothing. Author a group only when the backend has a better native shape (a real database table, an analytics pipeline) — the built-in <code>fileKv</code> and <code>cursorHostedStorage</code> both do.</p><p><code>evals</code> keeps playground eval batches across restarts (<code>put</code>, <code>delete</code>, <code>list</code> over run snapshots keyed by <code>runId</code>). The Agent SDK upserts a snapshot as a batch starts, progresses, and finishes, prunes runs past the playground history window, and lists everything back at serve start. <strong>Derived form</strong>: one key per run under <code>agentkit/v1/{agent}/eval-runs/{runId}</code> — needs core <code>put</code> + <code>delete</code> + <code>list</code>. Only a core missing <code>delete</code> or <code>list</code> leaves eval history in process memory (cleared on restart). See <a href="./evals.html#configure-eval-runs">Evals</a>.</p><p><code>abs</code> exports live A/B metrics: <code>putSample</code> appends one cumulative metric sample per enrolled experiment on each completed or failed turn; optional <code>putSnapshot</code> / <code>getSnapshot</code> store and serve back the latest aggregate so a replacement host with no local sessions can still serve the A/Bs surface. <strong>Derived form</strong>: each sample lands as its own key (<code>agentkit/v1/{agent}/ab-samples/{experiment}/{sessionId}/{at}</code> — a blind append-only put, never a read-modify-write of one growing array) and the snapshot lives at the fixed <code>agentkit/v1/{agent}/ab-snapshot</code> key (last-write-wins is correct for &quot;latest aggregate&quot;). <code>putSample</code> and <code>putSnapshot</code> need only core <code>put</code>; <code>getSnapshot</code> needs core <code>get</code>. Session event logs remain the assignment source of truth either way. See <a href="./ab.html">Live A/B metrics</a>.</p><h2 id="policy" tabindex="-1">Policy <a class="header-anchor" href="#policy" aria-label="Permalink to &quot;Policy&quot;">​</a></h2><p>Two knobs change behavior:</p><div class="language-ts vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">ts</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">export</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> default</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;"> defineStorage</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">({</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">  policy: {</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">    // Batch event writes while the session is busy (default: once per turn)</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">    debounceMs: </span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">30_000</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">,</span></span>
@@ -27,4 +27,4 @@ import{_ as e,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const c
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">export</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> default</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;"> defineStorage</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">({</span></span>
 <span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">  ...</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">cursorHostedStorage</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(),</span></span>
-<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><p>Built-in helpers:</p><table tabindex="0"><thead><tr><th>Import</th><th>Backend</th></tr></thead><tbody><tr><td><code>@cursor/july/storage/file-kv</code></td><td>File-per-key under <code>.agent-serve/kv</code></td></tr><tr><td><code>@cursor/july/storage/cursor-hosted</code></td><td>Platform Bugbot <code>agent_serve_*</code> via control-plane proxy</td></tr></tbody></table><h2 id="bring-your-own-backend" tabindex="-1">Bring your own backend <a class="header-anchor" href="#bring-your-own-backend" aria-label="Permalink to &quot;Bring your own backend&quot;">​</a></h2><p>There is no built-in Postgres backend on purpose. Cursor&#39;s internal <code>agent_serve_*</code> tables are owned by the backend and reachable only through the hosted proxy, and this doc does not prescribe a schema — what you back the KV with is your call. Any durable store works:</p><ul><li><strong>Local disk</strong> — the built-in <code>fileKv</code> (single process only).</li><li><strong>Object storage (S3-class)</strong> — one object per key; conditional writes map directly onto the contract (<code>putIfAbsent</code> = put with <code>If-None-Match: *</code>, <code>putIfVersion</code> = put with <code>If-Match: &lt;etag&gt;</code>, the ETag is the version token, <code>listKeys</code> is a prefix listing).</li><li><strong>Redis, DynamoDB, a SQL table, …</strong> — anything that can do an atomic compare-and-set and a prefix listing.</li></ul><p>Implement the <code>StorageConfig</code> methods (and the <code>cas</code> group when the deployment uses coordination features) against that store. The contract, defined at <code>@cursor/july/kv</code>: version tokens are opaque strings that change on every successful write — including plain <code>put</code>, so a stale token fences instead of clobbering; conditional writes are atomic; <code>listKeys</code> returns every key under the prefix. <code>@cursor/july/kv/memory</code> is a complete reference implementation to compare behavior against.</p><h2 id="conditional-writes-the-cas-group" tabindex="-1">Conditional writes (the <code>cas</code> group) <a class="header-anchor" href="#conditional-writes-the-cas-group" aria-label="Permalink to &quot;Conditional writes (the \`cas\` group)&quot;">​</a></h2><p>The <code>cas</code> group is compare-and-swap over the same keyspace: <code>getWithVersion</code> / <code>putIfAbsent</code> / <code>putIfVersion</code> / <code>listKeys</code>, defined backend-agnostically at <code>@cursor/july/kv</code>. Plain storage works without it, so an existing backend keeps working across a platform upgrade. Coordination features require it: they fail at startup, with a message naming this group, when the backend lacks it. <code>defineStorage</code> rejects a partial group — implement all four methods or none. Version tokens are opaque strings that must change on every write (a counter column, a row version, a content hash). The built-in backends both include it: <code>fileKv</code> uses content-hash tokens (single-process correctness) and <code>cursorHostedStorage</code> the control-plane proxy (a <code>version</code> counter on the server).</p><h2 id="durable-sessions-the-session-ledger" tabindex="-1">Durable sessions (the session ledger) <a class="header-anchor" href="#durable-sessions-the-session-ledger" aria-label="Permalink to &quot;Durable sessions (the session ledger)&quot;">​</a></h2><p>Conditional writes are the substrate for the session ledger, which ships behind <code>defineAgent({ serving: &quot;ledger&quot; })</code>: session identity, one-writer-per-session leases with generation fencing, a fenced commit sequence, and a <strong>durable inbox</strong>. A ledger-mode delivery parks its event in the session mailbox first — the source ack point moves after that write — then tries the lease: if another holder is serving the session, the delivery walks away and the holder&#39;s commit drains the parked event in a follow-up turn under the same lease. Sessions whose holder died with events pending are recovered by a janitor loop that scans ledgers by prefix, which is why ledger mode requires the full <code>cas</code> group <strong>including <code>listKeys</code></strong>: the janitor scan and the orphan sweep are prefix scans, and mailbox payload blobs go through the backend&#39;s plain <code>get</code>/<code>put</code>/<code>delete</code>. Coordination runs through the agent&#39;s storage backend by default or an explicit <code>durableSessionStore</code> override. Either way you supply the <strong>raw</strong> store: the engine namespaces every ledger key under <code>agentkit/v1/{agent}/durable-session/</code> itself, at the same boundary where <code>host.kv</code> gets its agent scoping — a self-hoster never writes prefixes. A backend that implements <code>cas</code> needs no storage migration to adopt this.</p><p>One sharp edge: the <code>cas</code> group and the plain methods must be backed by the <strong>same durable store</strong>. Pairing a real plain backend with <code>memoryCasTable()</code> (or any process-local <code>cas</code> stand-in) and <code>serving: &quot;ledger&quot;</code> silently voids ack-after-durable — the mailbox blobs persist but the leases, heads, and pending indexes evaporate on restart, so acked events are never redelivered. <code>memoryCasTable()</code> is for test fixtures and inert sinks only.</p>`,40)])])}const g=e(n,[["render",o]]);export{c as __pageData,g as default};
+<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><p>Built-in helpers:</p><table tabindex="0"><thead><tr><th>Import</th><th>Backend</th></tr></thead><tbody><tr><td><code>@cursor/july/storage/file-kv</code></td><td>File-per-key under <code>.agent-serve/kv</code></td></tr><tr><td><code>@cursor/july/storage/cursor-hosted</code></td><td>Platform Bugbot <code>agent_serve_*</code> via control-plane proxy</td></tr></tbody></table><h2 id="bring-your-own-backend" tabindex="-1">Bring your own backend <a class="header-anchor" href="#bring-your-own-backend" aria-label="Permalink to &quot;Bring your own backend&quot;">​</a></h2><p>There is no built-in Postgres backend on purpose. Cursor&#39;s internal <code>agent_serve_*</code> tables are owned by the backend and reachable only through the hosted proxy, and this doc does not prescribe a schema — what you back the KV with is your call. Any durable store works:</p><ul><li><strong>Local disk</strong> — the built-in <code>fileKv</code> (single process only).</li><li><strong>Object storage (S3-class)</strong> — one object per key; conditional writes map directly onto the contract (<code>putIfAbsent</code> = put with <code>If-None-Match: *</code>, <code>putIfVersion</code> = put with <code>If-Match: &lt;etag&gt;</code>, the ETag is the version token, <code>listKeys</code> is a prefix listing).</li><li><strong>Redis, DynamoDB, a SQL table, …</strong> — anything that can do an atomic compare-and-set and a prefix listing.</li></ul><p>Implement the <code>StorageConfig</code> methods (and the <code>cas</code> group when the deployment uses coordination features) against that store. The contract, defined at <code>@cursor/july/kv</code>: version tokens are opaque strings that change on every successful write — including plain <code>put</code>, so a stale token fences instead of clobbering; conditional writes are atomic; <code>listKeys</code> returns every key under the prefix. <code>@cursor/july/kv/memory</code> is a complete reference implementation to compare behavior against.</p><h2 id="conditional-writes-the-cas-group" tabindex="-1">Conditional writes (the <code>cas</code> group) <a class="header-anchor" href="#conditional-writes-the-cas-group" aria-label="Permalink to &quot;Conditional writes (the \`cas\` group)&quot;">​</a></h2><p>The <code>cas</code> group is compare-and-swap over the same keyspace: <code>getWithVersion</code> / <code>putIfAbsent</code> / <code>putIfVersion</code> / <code>listKeys</code>, defined backend-agnostically at <code>@cursor/july/kv</code>. Plain storage works without it, so an existing backend keeps working across a platform upgrade. Coordination features require it: they fail at startup, with a message naming this group, when the backend lacks it. <code>defineStorage</code> rejects a partial group — implement all four methods or none. Version tokens are opaque strings that must change on every write (a counter column, a row version, a content hash). The built-in backends both include it: <code>fileKv</code> uses content-hash tokens (single-process correctness) and <code>cursorHostedStorage</code> the control-plane proxy (a <code>version</code> counter on the server).</p><h2 id="durable-sessions-the-session-ledger" tabindex="-1">Durable sessions (the session ledger) <a class="header-anchor" href="#durable-sessions-the-session-ledger" aria-label="Permalink to &quot;Durable sessions (the session ledger)&quot;">​</a></h2><p>Conditional writes are the substrate for the session ledger, which ships behind <code>defineAgent({ serving: &quot;ledger&quot; })</code>: session identity, one-writer-per-session leases with generation fencing, a fenced commit sequence, and a <strong>durable inbox</strong>. A ledger-mode delivery parks its event in the session mailbox first — the source ack point moves after that write — then tries the lease: if another holder is serving the session, the delivery walks away and the holder&#39;s commit drains the parked event in a follow-up turn under the same lease. Sessions whose holder died with events pending are recovered by a janitor loop that scans ledgers by prefix, which is why ledger mode requires the full <code>cas</code> group <strong>including <code>listKeys</code></strong>: the janitor scan and the orphan sweep are prefix scans, and mailbox payload blobs go through the backend&#39;s plain <code>get</code>/<code>put</code>/<code>delete</code>. Coordination runs through the agent&#39;s storage backend by default or an explicit <code>durableSessionStore</code> override. Either way you supply the <strong>raw</strong> store: the engine namespaces every ledger key under <code>agentkit/v1/{agent}/durable-session/</code> itself, at the same boundary where <code>host.kv</code> gets its agent scoping — a self-hoster never writes prefixes. A backend that implements <code>cas</code> needs no storage migration to adopt this.</p><p>One sharp edge: the <code>cas</code> group and the plain methods must be backed by the <strong>same durable store</strong>. Pairing a real plain backend with <code>memoryCasTable()</code> (or any process-local <code>cas</code> stand-in) and <code>serving: &quot;ledger&quot;</code> silently voids ack-after-durable — the mailbox blobs persist but the leases, heads, and pending indexes evaporate on restart, so acked events are never redelivered. <code>memoryCasTable()</code> is for test fixtures and inert sinks only.</p>`,41)])])}const g=s(n,[["render",o]]);export{k as __pageData,g as default};

package/dist/docs/assets/storage.md.D8NQ389l.lean.js

@@ -0,0 +1 @@
+import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const k=JSON.parse(`{"title":"Storage","description":"Point the Agent SDK's durable storage at a backend you own with defineStorage.","frontmatter":{"title":"Storage","description":"Point the Agent SDK's durable storage at a backend you own with defineStorage."},"headers":[],"relativePath":"storage.md","filePath":"storage.md"}`),n={name:"storage.md"};function o(r,e,d,l,h,p){return a(),t("div",null,[...e[0]||(e[0]=[i("",41)])])}const g=s(n,[["render",o]]);export{k as __pageData,g as default};

package/dist/docs/building-with-agents.html

@@ -9,9 +9,9 @@
     <link rel="preload stylesheet" href="/docs/assets/style.CK3hwt6q.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.IEXEim5m.js"></script>
+    <script type="module" src="/docs/assets/app.BpWAgjaR.js"></script>
     <link rel="preload" href="/docs/assets/inter-roman-latin.Di8DUHzh.woff2" as="font" type="font/woff2" crossorigin="">
-    <link rel="modulepreload" href="/docs/assets/chunks/theme.DroGdz_Q.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.B5IvUxOB.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
     <link rel="modulepreload" href="/docs/assets/building-with-agents.md.PeZaZA1P.lean.js">
     <script id="check-dark-mode">(()=>{const e=localStorage.getItem("vitepress-theme-appearance")||"auto",a=window.matchMedia("(prefers-color-scheme: dark)").matches;(!e||e==="auto"?a:e==="dark")&&document.documentElement.classList.add("dark")})();</script>
@@ -31,7 +31,7 @@
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> eval</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> .</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --list</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> eval</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> .</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --json</span></span></code></pre></div><p><code>serve</code> boots without an API key, which is enough to check channel mounts and the playground shell. Model turns need a credential. When the environment lacks one, finish every key-free check, then hand the <code>run</code> and <code>eval</code> commands to the operator.</p><p>Test server tools with <code>call</code> before tuning the prompt. It runs a tool in-process with schema validation and no model turn. If the tool returns the wrong data, a prompt change won&#39;t fix it.</p><p><code>validate</code> and <code>run</code> don&#39;t type-check the project because tsx strips types. Run the project&#39;s TypeScript check before shipping. Tool results must also be JSON-shaped. Use object literals or <code>type</code> aliases for <code>execute</code> return types instead of <code>interface</code> types.</p><h2 id="how-do-i-improve-the-agent-after-its-first-run" tabindex="-1">How do I improve the agent after its first run? <a class="header-anchor" href="#how-do-i-improve-the-agent-after-its-first-run" aria-label="Permalink to &quot;How do I improve the agent after its first run?&quot;">​</a></h2><p>Scaffolding proves the agent runs. It says nothing about quality. Once a smoke turn passes, give the hillclimb skill:</p><ol><li><strong>Fixtures</strong>: one to three fixed inputs, such as a PR URL, a canonical question, or a saved webhook payload</li><li><strong>Success criteria</strong>: correct tool choice, fewer tool calls, lower wall time, or better output</li><li><strong>The freeze line</strong>: the API contract, output shape, and existing evals that must stay unchanged</li></ol><p>Have the coding agent read <a href="./../skills/hillclimb/SKILL.html"><code>skills/hillclimb/SKILL.md</code></a>. It measures the current run, proposes one change, remeasures the same fixtures, and adds an eval for each kept improvement.</p><h2 id="related" tabindex="-1">Related <a class="header-anchor" href="#related" aria-label="Permalink to &quot;Related&quot;">​</a></h2><ul><li><a href="./quickstart.html">Create your first agent</a></li><li><a href="./scaffolding-agents.html">Scaffold an agent with Cursor</a></li><li><a href="./evals.html">Evals</a></li><li><a href="./hillclimbing.html">Hillclimbing</a></li><li><a href="./reference/cli.html">CLI reference</a></li></ul></div></div></main><footer class="VPDocFooter" data-v-d5d7ccd1 data-v-a5908b0f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-a5908b0f><span class="visually-hidden" id="doc-footer-aria-label" data-v-a5908b0f>Pager</span><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link prev" href="/docs/templates/triage.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Previous page</span><span class="title" data-v-a5908b0f>Triage agent</span><!--]--></a></div><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link next" href="/docs/evals.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Next page</span><span class="title" data-v-a5908b0f>Evals</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"hdxsvX5M\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"qeoGgPw0\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DGrw_wOu\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"CJYd5YmN\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
+    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"DYjwREAP\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"CVe_O75-\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DMbdFGVQ\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"D8NQ389l\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
     
   </body>
 </html>

package/dist/docs/concepts.html

@@ -9,9 +9,9 @@
     <link rel="preload stylesheet" href="/docs/assets/style.CK3hwt6q.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.IEXEim5m.js"></script>
+    <script type="module" src="/docs/assets/app.BpWAgjaR.js"></script>
     <link rel="preload" href="/docs/assets/inter-roman-latin.Di8DUHzh.woff2" as="font" type="font/woff2" crossorigin="">
-    <link rel="modulepreload" href="/docs/assets/chunks/theme.DroGdz_Q.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.B5IvUxOB.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
     <link rel="modulepreload" href="/docs/assets/concepts.md.CqOsxbMU.lean.js">
     <script id="check-dark-mode">(()=>{const e=localStorage.getItem("vitepress-theme-appearance")||"auto",a=window.matchMedia("(prefers-color-scheme: dark)").matches;(!e||e==="auto"?a:e==="dark")&&document.documentElement.classList.add("dark")})();</script>
@@ -22,7 +22,7 @@
 <span class="line"><span>  sessions/&lt;id&gt;/events.ndjson</span></span>
 <span class="line"><span>  sessions/&lt;id&gt;/workspace/</span></span>
 <span class="line"><span>  traces/&lt;sessionId&gt;.ndjson</span></span></code></pre></div><h2 id="how-can-one-agent-call-another" tabindex="-1">How can one agent call another? <a class="header-anchor" href="#how-can-one-agent-call-another" aria-label="Permalink to &quot;How can one agent call another?&quot;">​</a></h2><p>Every mounted agent also serves MCP at <code>/&lt;slug&gt;/v1/mcp</code>. Another agent or MCP client can use <code>ask</code>, <code>check</code>, and <code>call_tool</code> to delegate work. A peer MCP connection such as <code>defineConnection({ agent: &quot;weather-agent&quot; })</code> adds those tools to the calling agent.</p><p>See <a href="./guides/agent-to-agent.html">Agent-to-agent</a> for a complete example.</p><h2 id="which-rules-prevent-common-setup-problems" tabindex="-1">Which rules prevent common setup problems? <a class="header-anchor" href="#which-rules-prevent-common-setup-problems" aria-label="Permalink to &quot;Which rules prevent common setup problems?&quot;">​</a></h2><ul><li>Use Node 22.13 or newer. Bun isn&#39;t supported.</li><li>Put evals under the project-root <code>evals/</code> directory, not <code>agent/evals/</code>.</li><li>Run a TypeScript check before shipping. <code>validate</code> and <code>run</code> execute TypeScript but don&#39;t type-check it.</li><li>Return JSON-shaped values from tool <code>execute</code> functions.</li><li>Keep local session workspaces away from parent rules you don&#39;t want the agent to inherit.</li><li>Sign in or set <code>CURSOR_API_KEY</code> before starting a model turn. Discovery, validation, direct tool calls, and server startup work without a credential.</li></ul><h2 id="related" tabindex="-1">Related <a class="header-anchor" href="#related" aria-label="Permalink to &quot;Related&quot;">​</a></h2><ul><li><a href="./quickstart.html">Quickstart</a></li><li><a href="./reference/project-layout.html">Project layout</a></li><li><a href="./reference/sessions.html">Sessions and streaming</a></li><li><a href="./reference/channels.html">Channels</a></li><li><a href="./ab.html">Live A/B metrics</a></li><li><a href="./guides/cloud-runtime.html">Cloud runtime</a></li></ul></div></div></main><footer class="VPDocFooter" data-v-d5d7ccd1 data-v-a5908b0f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-a5908b0f><span class="visually-hidden" id="doc-footer-aria-label" data-v-a5908b0f>Pager</span><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link prev" href="/docs/scaffolding-agents.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Previous page</span><span class="title" data-v-a5908b0f>Scaffold an agent with Cursor</span><!--]--></a></div><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link next" href="/docs/templates/triage.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Next page</span><span class="title" data-v-a5908b0f>Triage agent</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"hdxsvX5M\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"qeoGgPw0\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DGrw_wOu\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"CJYd5YmN\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
+    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"DYjwREAP\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"CVe_O75-\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DMbdFGVQ\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"D8NQ389l\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
     
   </body>
 </html>

package/dist/docs/deployment.html

@@ -9,9 +9,9 @@
     <link rel="preload stylesheet" href="/docs/assets/style.CK3hwt6q.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.IEXEim5m.js"></script>
+    <script type="module" src="/docs/assets/app.BpWAgjaR.js"></script>
     <link rel="preload" href="/docs/assets/inter-roman-latin.Di8DUHzh.woff2" as="font" type="font/woff2" crossorigin="">
-    <link rel="modulepreload" href="/docs/assets/chunks/theme.DroGdz_Q.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.B5IvUxOB.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
     <link rel="modulepreload" href="/docs/assets/deployment.md.Cdp4rTE4.lean.js">
     <script id="check-dark-mode">(()=>{const e=localStorage.getItem("vitepress-theme-appearance")||"auto",a=window.matchMedia("(prefers-color-scheme: dark)").matches;(!e||e==="auto"?a:e==="dark")&&document.documentElement.classList.add("dark")})();</script>
@@ -73,7 +73,7 @@
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --state-root</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> /var/lib/agent-serve</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --bearer-token</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">$AGENT_SDK_BEARER_TOKEN</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">&quot;</span></span></code></pre></div><p>Mount the state root as a persistent volume and inject secrets at startup. Install <code>git</code> and <code>gh</code> when channels need host-side GitHub work. Don&#39;t put secrets in the image.</p><h3 id="serve-many-agents-from-one-process" tabindex="-1">Serve many agents from one process <a class="header-anchor" href="#serve-many-agents-from-one-process" aria-label="Permalink to &quot;Serve many agents from one process&quot;">​</a></h3><p>Point <code>serve</code> at a folder of agent projects and every child mounts under its directory name on one port. One process, one state root, one credential:</p><div class="language-bash vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">bash</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> serve</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> /srv/agents</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"># index at /, each agent at /&lt;slug&gt;/v1/*, /&lt;slug&gt;/playground</span></span></code></pre></div><p>Only mount what you mean to run. Every mounted agent&#39;s channels are live, and webhook-driven agents spend model budget on every wake. <code>--mode single</code> serves exactly one agent at the unslugged <code>/v1/*</code> when the agent is the whole host. See the <a href="./reference/http-api.html">HTTP API</a> for route layout and the <a href="./guides/slack.html">Slack guide</a> for multi-agent token setup.</p><h3 id="the-production-flags" tabindex="-1">The production flags <a class="header-anchor" href="#the-production-flags" aria-label="Permalink to &quot;The production flags&quot;">​</a></h3><p>Use these settings in production:</p><table tabindex="0"><thead><tr><th>Flag</th><th>In production</th></tr></thead><tbody><tr><td><code>--dev</code></td><td>Leave off. Dev mode admits unsigned loopback GitHub deliveries, widens playground session listing on loopback, and never auto-fires schedules.</td></tr><tr><td><code>--bearer-token</code></td><td>Set on shared hosts unless an authenticating proxy is the trust boundary and you use <code>--allow-anonymous</code> instead.</td></tr><tr><td><code>--allow-anonymous</code></td><td>Use only behind an authenticating network boundary. It also widens playground session access so Slack and webhook sessions appear.</td></tr><tr><td><code>--state-root</code></td><td>Place on a persistent volume outside any repo.</td></tr><tr><td><code>--public-url</code></td><td>Set when cloud-runtime turns must call back into peers on this host.</td></tr><tr><td><code>--no-playground</code></td><td>Set when no human needs the UI.</td></tr><tr><td><code>--no-docs</code></td><td>Set to remove the documentation site at <code>/docs</code>.</td></tr><tr><td><code>--no-schedules</code></td><td>Set on secondary hosts so schedules run exactly once.</td></tr></tbody></table><p>Schedules fire on their cron cadence (UTC) in production mode. They have no cross-host coordination, so enable them on exactly one serving process per project.</p><h3 id="restarts-and-upgrades" tabindex="-1">Restarts and upgrades <a class="header-anchor" href="#restarts-and-upgrades" aria-label="Permalink to &quot;Restarts and upgrades&quot;">​</a></h3><p>Restarts preserve sessions, event streams, and SDK conversation state under the state root. Parked approvals and in-memory reminders don&#39;t survive a restart; re-run or recreate them afterward.</p><h3 id="observability" tabindex="-1">Observability <a class="header-anchor" href="#observability" aria-label="Permalink to &quot;Observability&quot;">​</a></h3><p>Use <a href="./reference/cli.html#logs"><code>agent-sdk logs</code></a> for runtime output, <a href="./guides/opentelemetry.html">OpenTelemetry</a> for OTLP traces and metrics, <a href="./reference/hooks.html">hooks</a> for in-process subscribers, and <a href="./reference/sessions.html#how-do-i-inspect-a-saved-event-stream">session traces</a> for incident review.</p><h2 id="what-s-next" tabindex="-1">What&#39;s next <a class="header-anchor" href="#what-s-next" aria-label="Permalink to &quot;What&#39;s next&quot;">​</a></h2><p>Continue with these pages:</p><ul><li><a href="./reference/cli.html#deploy">CLI reference</a>: deploy, inspect, stop, and rotate hosted agents</li><li><a href="./storage.html">Storage</a>: preserve supported records across engine replacements</li><li><a href="./reference/channels.html#auth-policies">Channels</a>: the auth policies in detail</li><li><a href="./guides/github.html">GitHub guide</a>: delivery paths without a public URL</li><li><a href="./troubleshooting.html">Troubleshooting</a>: the symptom table for when a deploy misbehaves</li></ul></div></div></main><footer class="VPDocFooter" data-v-d5d7ccd1 data-v-a5908b0f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-a5908b0f><span class="visually-hidden" id="doc-footer-aria-label" data-v-a5908b0f>Pager</span><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link prev" href="/docs/example-agents/codeowners-review.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Previous page</span><span class="title" data-v-a5908b0f>Codeowners review</span><!--]--></a></div><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link next" href="/docs/troubleshooting.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Next page</span><span class="title" data-v-a5908b0f>Fix common problems</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"hdxsvX5M\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"qeoGgPw0\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DGrw_wOu\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"CJYd5YmN\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
+    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"DYjwREAP\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"CVe_O75-\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DMbdFGVQ\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"D8NQ389l\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
     
   </body>
 </html>

package/dist/docs/evals.html

@@ -9,11 +9,11 @@
     <link rel="preload stylesheet" href="/docs/assets/style.CK3hwt6q.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.IEXEim5m.js"></script>
+    <script type="module" src="/docs/assets/app.BpWAgjaR.js"></script>
     <link rel="preload" href="/docs/assets/inter-roman-latin.Di8DUHzh.woff2" as="font" type="font/woff2" crossorigin="">
-    <link rel="modulepreload" href="/docs/assets/chunks/theme.DroGdz_Q.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.B5IvUxOB.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
-    <link rel="modulepreload" href="/docs/assets/evals.md.qeoGgPw0.lean.js">
+    <link rel="modulepreload" href="/docs/assets/evals.md.CVe_O75-.lean.js">
     <script id="check-dark-mode">(()=>{const e=localStorage.getItem("vitepress-theme-appearance")||"auto",a=window.matchMedia("(prefers-color-scheme: dark)").matches;(!e||e==="auto"?a:e==="dark")&&document.documentElement.classList.add("dark")})();</script>
     <script id="check-mac-os">document.documentElement.classList.toggle("mac",/Mac|iPhone|iPod|iPad/i.test(navigator.platform));</script>
   </head>
@@ -68,7 +68,7 @@
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">  // judge: { model: &quot;...&quot; },</span><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"> // default judge model for t.judge.*</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">  // reporters: [],</span><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"> // destinations that observe every case</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">  // maxPlaygroundRuns: 50,</span><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"> // playground /v1/dev/evals history only (default 20)</span></span>
-<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><p>The timeout order is case or file <code>timeoutMs</code>, CLI <code>--timeout-ms</code>, project config <code>timeoutMs</code>, then the 180-second runner default.</p><p>The optional fields:</p><table tabindex="0"><thead><tr><th>Option</th><th>Default</th><th>Meaning</th></tr></thead><tbody><tr><td><code>timeoutMs</code></td><td><code>180_000</code></td><td>Project-wide per-case timeout</td></tr><tr><td><code>judge</code></td><td>unset</td><td>Default judge model for <code>t.judge.*</code>; see <a href="#judge-free-form-output">Judge free-form output</a></td></tr><tr><td><code>reporters</code></td><td>unset</td><td>Destinations that observe every case; <code>--skip-report</code> suppresses them</td></tr><tr><td><code>maxPlaygroundRuns</code></td><td><code>20</code></td><td>Max batches in the playground / <code>/v1/dev/evals*</code> history (not CLI <code>eval</code>)</td></tr></tbody></table><p>Reporters come from <code>@cursor/july/evals/reporters</code>: <code>JUnit</code> writes a JUnit XML file for CI, <code>Artifacts</code> writes per-case files, and <code>combineReporters</code> merges several into one (<code>renderJUnitXml</code> renders the XML for a custom destination). A file or case can add its own <code>reporters</code> on top of the config list.</p><p>Playground batches live in process memory and disappear when <code>serve</code> exits. Navigating away and back still works while the process is up. To keep batches across restarts, declare an <code>evals</code> table in <code>agent/storage.ts</code>; see <a href="./storage.html#eval-and-a-b-tables">Storage</a>.</p><h2 id="drive-and-assert-with-t" tabindex="-1">Drive and assert with <code>t</code> <a class="header-anchor" href="#drive-and-assert-with-t" aria-label="Permalink to &quot;Drive and assert with `t`&quot;">​</a></h2><p><code>t</code> is both the driver and the assertion surface. You write ordinary control flow, sending turns and asserting inline.</p><p>Drive the agent with <code>t.send(message, options?)</code>. It runs one turn and waits for the session to park or fail. Multiple sends in one case share the session, which is how you write multi-turn evals.</p><p>Each <code>t.send</code> resolves to a turn result with <code>message</code>, <code>sessionId</code>, <code>events</code>, <code>toolCalls</code>, <code>ok</code>, and <code>index</code>. The turn carries the same assertion vocabulary as <code>t</code>, scoped to that turn, so you can grade an intermediate turn before the next send overwrites <code>t.reply</code>. <code>turn.expectOk()</code> throws when the turn failed, for later steps that depend on it.</p><p>Read the full case state with <code>t.reply</code> (the last assistant text), <code>t.events</code> (every captured session event across turns), <code>t.turns</code> (settled turns, oldest first), and <code>t.sessionId</code>. <code>t.signal</code> aborts when the case hits its timeout; pass it to your own async work.</p><p>Assert with the gates:</p><table tabindex="0"><thead><tr><th>Gate</th><th>Checks</th></tr></thead><tbody><tr><td><code>t.succeeded()</code></td><td>the run did not fail and is not parked on an unanswered approval</td></tr><tr><td><code>t.parked()</code></td><td>the run cleanly parked on an unanswered approval request</td></tr><tr><td><code>t.messageIncludes(token)</code></td><td>the joined assistant text matches a string or <code>RegExp</code></td></tr><tr><td><code>t.calledTool(name, matcher?)</code></td><td>a matching call to <code>name</code> happened</td></tr><tr><td><code>t.notCalledTool(name)</code></td><td>no request for <code>name</code>, in any lifecycle state</td></tr><tr><td><code>t.loadedSkill(name)</code></td><td>the agent opened the skill&#39;s <code>SKILL.md</code> (read, grep, or shell <code>cat</code>)</td></tr><tr><td><code>t.toolOrder(names)</code></td><td>tool requests appear in this relative order (extra calls allowed)</td></tr><tr><td><code>t.usedNoTools()</code></td><td>no tool calls at all</td></tr><tr><td><code>t.maxToolCalls(max)</code></td><td>at most <code>max</code> tool calls</td></tr><tr><td><code>t.noFailedActions()</code></td><td>no tool call reported an error</td></tr><tr><td><code>t.calledSubagent(name, matcher?)</code></td><td>a matching subagent delegation happened</td></tr><tr><td><code>t.taggedArtifact(kind?, predicate?)</code></td><td>at least one <a href="./reference/artifacts.html">artifact</a> was tagged</td></tr><tr><td><code>t.event(type, matcher?)</code></td><td>at least one matching event of <code>type</code> occurred</td></tr><tr><td><code>t.notEvent(type, matcher?)</code></td><td>no matching event of <code>type</code> occurred</td></tr><tr><td><code>t.eventOrder(matchers)</code></td><td>matching event groups occur in this relative order</td></tr><tr><td><code>t.eventsSatisfy(label, predicate)</code></td><td>your predicate over the typed event stream</td></tr><tr><td><code>t.check(value, expectation)</code></td><td>any value, against a builder</td></tr><tr><td><code>t.score(name, value)</code></td><td>records a 0–1 score you computed; soft until you add a bar</td></tr><tr><td><code>t.requireToolCall(name, matcher?)</code></td><td>gates on a matching call and returns it, so later code can read its input and output</td></tr><tr><td><code>t.requireInputRequest(filter?)</code></td><td>gates on exactly one pending approval request and returns it</td></tr></tbody></table><p>Every gate returns a handle: <code>.soft()</code> demotes it to tracked-only, <code>.atLeast(0.7)</code> adds a soft score bar, and <code>.gate(0.8)</code> promotes a scored assertion into a hard gate.</p><p>With no matcher, <code>calledTool</code> is request-based: a requested call counts even when its result has not arrived. Pass <code>t.calledTool(&quot;inspect_pr&quot;, { status: &quot;completed&quot; })</code> to require the call to return. <code>input</code>, <code>output</code>, and <code>count</code> matcher fields accept a literal, a <code>RegExp</code>, or a predicate.</p><p>The expectation builders are <code>includes(string | RegExp)</code>, <code>equals(value)</code>, <code>matches(schema)</code>, <code>similarity(expected)</code>, and <code>satisfies(predicate, label)</code>. <code>includes</code> stringifies its input, <code>equals</code> compares values deeply, <code>matches</code> validates against a Standard Schema (or anything with <code>safeParse</code>, like Zod), <code>similarity</code> scores normalized text similarity, and <code>satisfies</code> runs your predicate. The plain function <code>normalizedSimilarity(actual, expected)</code> returns the same 0–1 score for use with <code>t.score</code>.</p><p>A few more context members shape a case: <code>t.require(value, expectation)</code> records a gate and stops the test body when it fails, without a duplicate execution error. <code>t.skip(reason)</code> ends the case as skipped (reported separately, never changes the exit code; call it before sending messages). <code>t.metric(name, value)</code> records a structured score for the playground case card. <code>t.log(message)</code> records a debug line for the CLI and playground result.</p><p>Three <code>t.send</code> options apply on session create (first <code>t.send</code> only):</p><ul><li><code>workspaceFiles</code> — <code>{ path: contents }</code>, seeded into the local session workspace. Prefer this over machine-local paths.</li><li><code>workspaceDir</code> — absolute harness cwd (local runtime).</li><li><code>cloud</code> — per-session cloud options merged over the agent&#39;s static <code>cloud</code> config (repos / env / …). Use a pinned <code>repos</code> override to attach a fixture repo for cloud evals without putting it on the agent&#39;s default <code>cloud.repos</code>. Cloud ignores <code>workspaceFiles</code> seeds.</li></ul><div class="language-ts vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">ts</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">const</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> toolResults</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> =</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> t.events.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">filter</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">((</span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">e</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">) </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">=&gt;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> e.type </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">===</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &quot;action.result&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">);</span></span>
+<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><p>The timeout order is case or file <code>timeoutMs</code>, CLI <code>--timeout-ms</code>, project config <code>timeoutMs</code>, then the 180-second runner default.</p><p>The optional fields:</p><table tabindex="0"><thead><tr><th>Option</th><th>Default</th><th>Meaning</th></tr></thead><tbody><tr><td><code>timeoutMs</code></td><td><code>180_000</code></td><td>Project-wide per-case timeout</td></tr><tr><td><code>judge</code></td><td>unset</td><td>Default judge model for <code>t.judge.*</code>; see <a href="#judge-free-form-output">Judge free-form output</a></td></tr><tr><td><code>reporters</code></td><td>unset</td><td>Destinations that observe every case; <code>--skip-report</code> suppresses them</td></tr><tr><td><code>maxPlaygroundRuns</code></td><td><code>20</code></td><td>Max batches in the playground / <code>/v1/dev/evals*</code> history (not CLI <code>eval</code>)</td></tr></tbody></table><p>Reporters come from <code>@cursor/july/evals/reporters</code>: <code>JUnit</code> writes a JUnit XML file for CI, <code>Artifacts</code> writes per-case files, and <code>combineReporters</code> merges several into one (<code>renderJUnitXml</code> renders the XML for a custom destination). A file or case can add its own <code>reporters</code> on top of the config list.</p><p>Playground batches survive restarts whenever <code>agent/storage.ts</code> exists with an <code>evals</code> table or a KV core providing <code>delete</code> and <code>list</code> (the table is derived over the core); see <a href="./storage.html#eval-and-a-b-tables">Storage</a>. Without storage they live in process memory and disappear when <code>serve</code> exits — navigating away and back still works while the process is up.</p><h2 id="drive-and-assert-with-t" tabindex="-1">Drive and assert with <code>t</code> <a class="header-anchor" href="#drive-and-assert-with-t" aria-label="Permalink to &quot;Drive and assert with `t`&quot;">​</a></h2><p><code>t</code> is both the driver and the assertion surface. You write ordinary control flow, sending turns and asserting inline.</p><p>Drive the agent with <code>t.send(message, options?)</code>. It runs one turn and waits for the session to park or fail. Multiple sends in one case share the session, which is how you write multi-turn evals.</p><p>Each <code>t.send</code> resolves to a turn result with <code>message</code>, <code>sessionId</code>, <code>events</code>, <code>toolCalls</code>, <code>ok</code>, and <code>index</code>. The turn carries the same assertion vocabulary as <code>t</code>, scoped to that turn, so you can grade an intermediate turn before the next send overwrites <code>t.reply</code>. <code>turn.expectOk()</code> throws when the turn failed, for later steps that depend on it.</p><p>Read the full case state with <code>t.reply</code> (the last assistant text), <code>t.events</code> (every captured session event across turns), <code>t.turns</code> (settled turns, oldest first), and <code>t.sessionId</code>. <code>t.signal</code> aborts when the case hits its timeout; pass it to your own async work.</p><p>Assert with the gates:</p><table tabindex="0"><thead><tr><th>Gate</th><th>Checks</th></tr></thead><tbody><tr><td><code>t.succeeded()</code></td><td>the run did not fail and is not parked on an unanswered approval</td></tr><tr><td><code>t.parked()</code></td><td>the run cleanly parked on an unanswered approval request</td></tr><tr><td><code>t.messageIncludes(token)</code></td><td>the joined assistant text matches a string or <code>RegExp</code></td></tr><tr><td><code>t.calledTool(name, matcher?)</code></td><td>a matching call to <code>name</code> happened</td></tr><tr><td><code>t.notCalledTool(name)</code></td><td>no request for <code>name</code>, in any lifecycle state</td></tr><tr><td><code>t.loadedSkill(name)</code></td><td>the agent opened the skill&#39;s <code>SKILL.md</code> (read, grep, or shell <code>cat</code>)</td></tr><tr><td><code>t.toolOrder(names)</code></td><td>tool requests appear in this relative order (extra calls allowed)</td></tr><tr><td><code>t.usedNoTools()</code></td><td>no tool calls at all</td></tr><tr><td><code>t.maxToolCalls(max)</code></td><td>at most <code>max</code> tool calls</td></tr><tr><td><code>t.noFailedActions()</code></td><td>no tool call reported an error</td></tr><tr><td><code>t.calledSubagent(name, matcher?)</code></td><td>a matching subagent delegation happened</td></tr><tr><td><code>t.taggedArtifact(kind?, predicate?)</code></td><td>at least one <a href="./reference/artifacts.html">artifact</a> was tagged</td></tr><tr><td><code>t.event(type, matcher?)</code></td><td>at least one matching event of <code>type</code> occurred</td></tr><tr><td><code>t.notEvent(type, matcher?)</code></td><td>no matching event of <code>type</code> occurred</td></tr><tr><td><code>t.eventOrder(matchers)</code></td><td>matching event groups occur in this relative order</td></tr><tr><td><code>t.eventsSatisfy(label, predicate)</code></td><td>your predicate over the typed event stream</td></tr><tr><td><code>t.check(value, expectation)</code></td><td>any value, against a builder</td></tr><tr><td><code>t.score(name, value)</code></td><td>records a 0–1 score you computed; soft until you add a bar</td></tr><tr><td><code>t.requireToolCall(name, matcher?)</code></td><td>gates on a matching call and returns it, so later code can read its input and output</td></tr><tr><td><code>t.requireInputRequest(filter?)</code></td><td>gates on exactly one pending approval request and returns it</td></tr></tbody></table><p>Every gate returns a handle: <code>.soft()</code> demotes it to tracked-only, <code>.atLeast(0.7)</code> adds a soft score bar, and <code>.gate(0.8)</code> promotes a scored assertion into a hard gate.</p><p>With no matcher, <code>calledTool</code> is request-based: a requested call counts even when its result has not arrived. Pass <code>t.calledTool(&quot;inspect_pr&quot;, { status: &quot;completed&quot; })</code> to require the call to return. <code>input</code>, <code>output</code>, and <code>count</code> matcher fields accept a literal, a <code>RegExp</code>, or a predicate.</p><p>The expectation builders are <code>includes(string | RegExp)</code>, <code>equals(value)</code>, <code>matches(schema)</code>, <code>similarity(expected)</code>, and <code>satisfies(predicate, label)</code>. <code>includes</code> stringifies its input, <code>equals</code> compares values deeply, <code>matches</code> validates against a Standard Schema (or anything with <code>safeParse</code>, like Zod), <code>similarity</code> scores normalized text similarity, and <code>satisfies</code> runs your predicate. The plain function <code>normalizedSimilarity(actual, expected)</code> returns the same 0–1 score for use with <code>t.score</code>.</p><p>A few more context members shape a case: <code>t.require(value, expectation)</code> records a gate and stops the test body when it fails, without a duplicate execution error. <code>t.skip(reason)</code> ends the case as skipped (reported separately, never changes the exit code; call it before sending messages). <code>t.metric(name, value)</code> records a structured score for the playground case card. <code>t.log(message)</code> records a debug line for the CLI and playground result.</p><p>Three <code>t.send</code> options apply on session create (first <code>t.send</code> only):</p><ul><li><code>workspaceFiles</code> — <code>{ path: contents }</code>, seeded into the local session workspace. Prefer this over machine-local paths.</li><li><code>workspaceDir</code> — absolute harness cwd (local runtime).</li><li><code>cloud</code> — per-session cloud options merged over the agent&#39;s static <code>cloud</code> config (repos / env / …). Use a pinned <code>repos</code> override to attach a fixture repo for cloud evals without putting it on the agent&#39;s default <code>cloud.repos</code>. Cloud ignores <code>workspaceFiles</code> seeds.</li></ul><div class="language-ts vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">ts</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">const</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> toolResults</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> =</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> t.events.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">filter</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">((</span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">e</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">) </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">=&gt;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> e.type </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">===</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &quot;action.result&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">);</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">t.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">check</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">  toolResults.</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">length</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">,</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">  satisfies</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">((</span><span style="--shiki-light:#E36209;--shiki-dark:#FFAB70;">n</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">) </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">=&gt;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> (n </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">as</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> number</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">) </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">&lt;=</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> 4</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">, </span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">&quot;at most 4 tool calls&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">)</span></span>
@@ -96,14 +96,14 @@
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">      &quot;durationMs&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">: </span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">12340</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">    }</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">  ]</span></span>
-<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">}</span></span></code></pre></div><p>Each case result can also include <code>description</code>, <code>finalText</code>, <code>tools</code>, <code>error</code>, and tool arguments or output. This shape lets CI report the failed assertion without parsing terminal text.</p><h2 id="run-evals-in-the-playground" tabindex="-1">Run evals in the playground <a class="header-anchor" href="#run-evals-in-the-playground" aria-label="Permalink to &quot;Run evals in the playground&quot;">​</a></h2><p>Start the server with <code>--dev</code>, open the playground, and choose <strong>Evals</strong>. You can run every case or one case, watch progress, and open the resulting session trace.</p><div class="language-bash vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">bash</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> serve</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> .</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dev</span></span></code></pre></div><p>Playground runs target the live server instead of an ephemeral one. Their sessions appear in the session list. One eval batch can run at a time. By default those batches are <strong>in-memory only</strong> (capped by <code>maxPlaygroundRuns</code>); declare an <code>evals</code> table in <code>agent/storage.ts</code> if you need them after a serve restart — see <a href="./storage.html#eval-and-a-b-tables">Storage</a>.</p><p>The UI uses the playground eval routes (available without <code>--dev</code>): <code>GET /v1/dev/evals</code> lists datapoints and config (includes <code>maxPlaygroundRuns</code> / <code>durableRuns</code>), <code>GET /v1/dev/evals/runs</code> rehydrates recent batches after navigation, <code>POST /v1/dev/evals/runs</code> starts a batch (returns an <strong>Eval ID</strong> / <code>runId</code>), <code>GET /v1/dev/evals/runs/:runId</code> polls it, and <code>POST /v1/dev/evals/runs/:runId/cancel</code> cancels a running batch. See <a href="./reference/http-api.html#playground-eval-routes">Playground eval routes</a>. The start request returns <code>202</code> while cases run in the background. Poll until the snapshot status becomes <code>completed</code>, <code>failed</code>, or <code>cancelled</code>. Configuration errors appear on a failed snapshot.</p><p>On <code>--prod</code> / <code>--url</code>, the CLI prints the Eval ID as soon as the batch is accepted (and a Playground deep link with <code>?view=evals&amp;evalRunId=…</code>):</p><div class="language-bash vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">bash</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> eval</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --prod</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --slug</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> vulnerability-scanner</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --tag</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> deepsec</span></span>
+<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">}</span></span></code></pre></div><p>Each case result can also include <code>description</code>, <code>finalText</code>, <code>tools</code>, <code>error</code>, and tool arguments or output. This shape lets CI report the failed assertion without parsing terminal text.</p><h2 id="run-evals-in-the-playground" tabindex="-1">Run evals in the playground <a class="header-anchor" href="#run-evals-in-the-playground" aria-label="Permalink to &quot;Run evals in the playground&quot;">​</a></h2><p>Start the server with <code>--dev</code>, open the playground, and choose <strong>Evals</strong>. You can run every case or one case, watch progress, and open the resulting session trace.</p><div class="language-bash vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">bash</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> serve</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> .</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dev</span></span></code></pre></div><p>Playground runs target the live server instead of an ephemeral one. Their sessions appear in the session list. One eval batch can run at a time. Batches persist across restarts whenever <code>agent/storage.ts</code> provides an <code>evals</code> table or a KV core with <code>delete</code> and <code>list</code> (the table is derived over the core); without storage they are <strong>in-memory only</strong> (capped by <code>maxPlaygroundRuns</code>) — see <a href="./storage.html#eval-and-a-b-tables">Storage</a>.</p><p>The UI uses the playground eval routes (available without <code>--dev</code>): <code>GET /v1/dev/evals</code> lists datapoints and config (includes <code>maxPlaygroundRuns</code> / <code>durableRuns</code>), <code>GET /v1/dev/evals/runs</code> rehydrates recent batches after navigation, <code>POST /v1/dev/evals/runs</code> starts a batch (returns an <strong>Eval ID</strong> / <code>runId</code>), <code>GET /v1/dev/evals/runs/:runId</code> polls it, and <code>POST /v1/dev/evals/runs/:runId/cancel</code> cancels a running batch. See <a href="./reference/http-api.html#playground-eval-routes">Playground eval routes</a>. The start request returns <code>202</code> while cases run in the background. Poll until the snapshot status becomes <code>completed</code>, <code>failed</code>, or <code>cancelled</code>. Configuration errors appear on a failed snapshot.</p><p>On <code>--prod</code> / <code>--url</code>, the CLI prints the Eval ID as soon as the batch is accepted (and a Playground deep link with <code>?view=evals&amp;evalRunId=…</code>):</p><div class="language-bash vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">bash</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> eval</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --prod</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --slug</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> vulnerability-scanner</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --tag</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> deepsec</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"># Eval ID: evalrun_…</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"># Cancel:  agent-sdk eval cancel evalrun_… --prod --slug vulnerability-scanner</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;"># Playground: https://…/playground?view=evals&amp;evalRunId=evalrun_…</span></span>
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> eval</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> cancel</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> evalrun_…</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --prod</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --slug</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> vulnerability-scanner</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agent-sdk</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> eval</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> status</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> evalrun_…</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --prod</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --slug</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> vulnerability-scanner</span></span></code></pre></div><p>The Evals tab prefers the server’s in-flight batch (<code>activeRunId</code>) over a stale tab-local remembered id, so CLI / Slack kicks show up without an incognito window.</p><h2 id="what-good-cases-assert" tabindex="-1">What good cases assert <a class="header-anchor" href="#what-good-cases-assert" aria-label="Permalink to &quot;What good cases assert&quot;">​</a></h2><p>Gate decisions and shape, not prose. Model wording varies run to run. Tool choice, tool avoidance, and output structure are the stable contract.</p><ol><li><code>t.succeeded()</code>: always, first.</li><li>The tool decision: <code>calledTool</code> for the intended path, <code>notCalledTool</code> for the likely wrong alternative. The pair is stronger than either alone.</li><li>Output shape: a regex for the contract (<code>/ready|blocked/i</code>, a JSON marker, a findings-block fence), never exact sentences.</li><li>For structured output, parse <code>t.reply</code> and check fields with <code>satisfies</code> instead of substring-matching JSON.</li></ol><p>The common failure modes: asserting exact phrasing, packing more than about five gates into one case (split it), and cases that depend on live external state that drifts (pin the input; see fixtures).</p><h2 id="pick-fixtures-by-agent-type" tabindex="-1">Pick fixtures by agent type <a class="header-anchor" href="#pick-fixtures-by-agent-type" aria-label="Permalink to &quot;Pick fixtures by agent type&quot;">​</a></h2><p>The right fixture depends on the surface under test.</p><table tabindex="0"><thead><tr><th>Agent surface</th><th>Fixture</th></tr></thead><tbody><tr><td>Chat / domain assistant</td><td>A canonical prompt string, chosen once and frozen</td></tr><tr><td>Tool-heavy</td><td>Run <code>agent-sdk call &lt;tool&gt;</code> first to pin what the tool returns, then freeze the prompt that triggers it</td></tr><tr><td>GitHub webhook</td><td><code>agent-sdk github replay &lt;pr&gt; --events &#39;*&#39; --dry-run --out fixtures/github</code> snapshots real payloads for offline replay (<a href="./guides/github.html">GitHub guide</a>)</td></tr><tr><td>PR reviewer with host preparation</td><td>Diff, metadata, and gold labels pinned to commit SHAs; keep any live PR matrix small</td></tr><tr><td>Workspace-dependent</td><td><code>workspaceFiles</code> in <code>t.send</code> options, never developer-machine paths</td></tr></tbody></table><p>Tag the fast, reliably passing core <code>smoke</code> and run <code>--tag smoke</code> in the inner loop. Leave slow or flaky-prone cases untagged for explicit runs.</p><h3 id="materialize-api-backed-fixtures" tabindex="-1">Materialize API-backed fixtures <a class="header-anchor" href="#materialize-api-backed-fixtures" aria-label="Permalink to &quot;Materialize API-backed fixtures&quot;">​</a></h3><p>An input that only points at external data, such as a pull request URL, snapshot id, or pair of commit SHAs, is not self-contained. Fetch it once and commit the rendered fixture before you expand the suite.</p><ol><li>Save the diff, metadata, and labels under <code>fixtures/</code> at pinned revisions.</li><li>Seed those files with <code>workspaceFiles</code>, or read them from the fixture directory.</li><li>Assert decisions and output shape against the saved evidence.</li><li>Keep a small <code>smoke</code> subset for any remaining live pipeline checks.</li></ol><p>Read committed fixtures with <code>@cursor/july/evals/loaders</code>: <code>loadJson</code>, <code>loadJsonl</code>, and <code>loadYaml</code> resolve relative paths against the project root the runner discovered, not the cwd the CLI was invoked from (<code>resolveFixturePath</code> and <code>evalFixtureRoot</code> expose the same resolution for other file formats).</p><p><code>maxConcurrency</code> limits parallel datapoints. It does not limit model or API fan-out inside one datapoint. Materialized fixtures prevent a large suite from exhausting provider and GitHub rate limits. The <a href="./../skills/evals/SKILL.html">evals skill</a> has the full fixture workflow.</p><h2 id="keep-improvements-with-regression-evals" tabindex="-1">Keep improvements with regression evals <a class="header-anchor" href="#keep-improvements-with-regression-evals" aria-label="Permalink to &quot;Keep improvements with regression evals&quot;">​</a></h2><p>Every <a href="./hillclimbing.html">hillclimb</a> round that keeps a change must land an eval that would have failed before the change. If you can&#39;t express the improvement as a gate (a <code>calledTool</code> shift, a bounded <code>action.result</code> count, an output-shape regex), the improvement is unverified, and it&#39;ll regress silently.</p><p>The rule cuts the other way too: never weaken an existing gate to make a round pass. That&#39;s the freeze line moving, and it turns your regression suite into a list of checks that no longer protect anything.</p><h2 id="compare-variants-on-live-traffic" tabindex="-1">Compare variants on live traffic <a class="header-anchor" href="#compare-variants-on-live-traffic" aria-label="Permalink to &quot;Compare variants on live traffic&quot;">​</a></h2><p>Use <code>defineAB</code> to compare variant metrics on live sessions. It is not a test runner and has no <code>agent-sdk ab</code> command. Keep <code>defineEval</code> as the regression ratchet. Eval sessions do not enroll or change live metrics. See <a href="./ab.html">Live A/B metrics</a> for assignment, behavior, collection, and inspection.</p><h2 id="what-s-next" tabindex="-1">What&#39;s next <a class="header-anchor" href="#what-s-next" aria-label="Permalink to &quot;What&#39;s next&quot;">​</a></h2><p>Continue with these pages:</p><ul><li><a href="./ab.html">Live A/B metrics</a>: sticky variants and cumulative metrics on live sessions</li><li><a href="./hillclimbing.html">Hillclimbing</a>: the loop evals make trustworthy</li><li><a href="./building-with-agents.html">Building agents with agents</a>: have a coding agent write the first suite</li><li><a href="./guides/github.html">GitHub guide</a>: deterministic webhook fixtures with <code>github replay</code></li><li><a href="./reference/sessions.html">Sessions and streaming</a>: the events <code>t.events</code> contains</li></ul></div></div></main><footer class="VPDocFooter" data-v-d5d7ccd1 data-v-a5908b0f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-a5908b0f><span class="visually-hidden" id="doc-footer-aria-label" data-v-a5908b0f>Pager</span><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link prev" href="/docs/building-with-agents.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Previous page</span><span class="title" data-v-a5908b0f>Building agents with agents</span><!--]--></a></div><div class="pager" data-v-a5908b0f><a class="VPLink link pager-link next" href="/docs/ab.html" data-v-a5908b0f><!--[--><span class="desc" data-v-a5908b0f>Next page</span><span class="title" data-v-a5908b0f>Live A/B metrics</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"hdxsvX5M\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"qeoGgPw0\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DGrw_wOu\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"CJYd5YmN\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
+    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"DYjwREAP\",\"building-with-agents.md\":\"PeZaZA1P\",\"concepts.md\":\"CqOsxbMU\",\"deployment.md\":\"Cdp4rTE4\",\"evals.md\":\"CVe_O75-\",\"example-agents_approval-buddy.md\":\"CIiZ9coo\",\"example-agents_benny.md\":\"l7JTmm8X\",\"example-agents_bugbot.md\":\"Dp5JqHSQ\",\"example-agents_codebase-wiki.md\":\"D-lteFf0\",\"example-agents_codeowners-review.md\":\"BU2ZXLf-\",\"example-agents_concierge.md\":\"DA2al_NK\",\"example-agents_fsd.md\":\"DPz9ezO4\",\"example-agents_index.md\":\"xmJ87d_3\",\"example-agents_knowledge-base.md\":\"IneynQSR\",\"example-agents_oncall.md\":\"ZE0n6ZFN\",\"example-agents_security-reviewer.md\":\"Dkf1gyo6\",\"example-agents_slack-agent.md\":\"06jQXTAI\",\"example-agents_weather-agent.md\":\"DC3lINjo\",\"guides_agent-to-agent.md\":\"Bmbxy-FA\",\"guides_cloud-runtime.md\":\"V5igN4Sq\",\"guides_github.md\":\"B8QKtIpw\",\"guides_human-in-the-loop.md\":\"BWvT7UqY\",\"guides_mcp-oauth.md\":\"C7G7IykG\",\"guides_opentelemetry.md\":\"Csn7ZI25\",\"guides_slack.md\":\"zriQpU_9\",\"guides_webhooks.md\":\"DB-r_er9\",\"hillclimbing.md\":\"yXqdlv2R\",\"index.md\":\"COeoSkcF\",\"quickstart.md\":\"DZxBu44y\",\"reference_agent-config.md\":\"DHkkvcEZ\",\"reference_artifacts.md\":\"BGG4bZo-\",\"reference_channels.md\":\"MN8iDy8V\",\"reference_cli.md\":\"BDmjPPAm\",\"reference_connections.md\":\"CSQiNzre\",\"reference_hooks.md\":\"DJE5DXcT\",\"reference_http-api.md\":\"DMbdFGVQ\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"D3MdHM2z\",\"reference_prompt.md\":\"BaiweQxE\",\"reference_schedules.md\":\"gmfYzf_I\",\"reference_sessions.md\":\"DBVFi2Sx\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"DSrGLIuB\",\"reference_tools.md\":\"kJfJAP2m\",\"scaffolding-agents.md\":\"mkc3B_ZW\",\"storage.md\":\"D8NQ389l\",\"templates_triage.md\":\"CConZt1z\",\"troubleshooting.md\":\"CkgwmVI0\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"Agent SDK\",\"description\":\"Filesystem-first framework for building and serving Cursor agents.\",\"base\":\"/docs/\",\"head\":[],\"router\":{\"prefetchLinks\":true},\"appearance\":true,\"themeConfig\":{\"nav\":[{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Guides\",\"link\":\"/guides/webhooks\"},{\"text\":\"Examples\",\"link\":\"/example-agents/\"},{\"text\":\"Reference\",\"link\":\"/reference/project-layout\"}],\"search\":{\"provider\":\"local\"},\"outline\":{\"level\":[2,3]},\"sidebar\":[{\"text\":\"Getting started\",\"items\":[{\"text\":\"Overview\",\"link\":\"/\"},{\"text\":\"Quickstart\",\"link\":\"/quickstart\"},{\"text\":\"Scaffold an agent with Cursor\",\"link\":\"/scaffolding-agents\"},{\"text\":\"Concepts\",\"link\":\"/concepts\"}]},{\"text\":\"Templates\",\"items\":[{\"text\":\"Triage agent\",\"link\":\"/templates/triage\"}]},{\"text\":\"Improving agents\",\"items\":[{\"text\":\"Building agents with agents\",\"link\":\"/building-with-agents\"},{\"text\":\"Evals\",\"link\":\"/evals\"},{\"text\":\"Live A/B metrics\",\"link\":\"/ab\"},{\"text\":\"Storage\",\"link\":\"/storage\"},{\"text\":\"Hillclimbing\",\"link\":\"/hillclimbing\"}]},{\"text\":\"Guides\",\"items\":[{\"text\":\"Webhooks & custom channels\",\"link\":\"/guides/webhooks\"},{\"text\":\"GitHub\",\"link\":\"/guides/github\"},{\"text\":\"Slack\",\"link\":\"/guides/slack\"},{\"text\":\"Human-in-the-loop approvals\",\"link\":\"/guides/human-in-the-loop\"},{\"text\":\"Agent-to-agent\",\"link\":\"/guides/agent-to-agent\"},{\"text\":\"Host MCP OAuth\",\"link\":\"/guides/mcp-oauth\"},{\"text\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"},{\"text\":\"OpenTelemetry\",\"link\":\"/guides/opentelemetry\"}]},{\"text\":\"Example agents\",\"items\":[{\"text\":\"Choose an example\",\"link\":\"/example-agents/\"},{\"text\":\"Weather agent\",\"link\":\"/example-agents/weather-agent\"},{\"text\":\"Slack agent\",\"link\":\"/example-agents/slack-agent\"},{\"text\":\"Concierge\",\"link\":\"/example-agents/concierge\"},{\"text\":\"Playbook router\",\"link\":\"/example-agents/benny\"},{\"text\":\"Alert investigator\",\"link\":\"/example-agents/oncall\"},{\"text\":\"PR evidence reviewer\",\"link\":\"/example-agents/bugbot\"},{\"text\":\"Approval Buddy\",\"link\":\"/example-agents/approval-buddy\"},{\"text\":\"Security Reviewer\",\"link\":\"/example-agents/security-reviewer\"},{\"text\":\"Remote PR coordinator\",\"link\":\"/example-agents/fsd\"},{\"text\":\"Knowledge base\",\"link\":\"/example-agents/knowledge-base\"},{\"text\":\"Codebase wiki\",\"link\":\"/example-agents/codebase-wiki\"},{\"text\":\"Codeowners review\",\"link\":\"/example-agents/codeowners-review\"}]},{\"text\":\"Operating\",\"items\":[{\"text\":\"Deployment\",\"link\":\"/deployment\"},{\"text\":\"Fix common problems\",\"link\":\"/troubleshooting\"}]},{\"text\":\"Reference\",\"collapsed\":false,\"items\":[{\"text\":\"Project layout\",\"link\":\"/reference/project-layout\"},{\"text\":\"Agent config\",\"link\":\"/reference/agent-config\"},{\"text\":\"Instructions\",\"link\":\"/reference/instructions\"},{\"text\":\"Tools\",\"link\":\"/reference/tools\"},{\"text\":\"prompt\",\"link\":\"/reference/prompt\"},{\"text\":\"Skills\",\"link\":\"/reference/skills\"},{\"text\":\"MCP Connections\",\"link\":\"/reference/connections\"},{\"text\":\"Subagents\",\"link\":\"/reference/subagents\"},{\"text\":\"Channels\",\"link\":\"/reference/channels\"},{\"text\":\"Schedules & reminders\",\"link\":\"/reference/schedules\"},{\"text\":\"Hooks\",\"link\":\"/reference/hooks\"},{\"text\":\"Artifacts\",\"link\":\"/reference/artifacts\"},{\"text\":\"Sessions & streaming\",\"link\":\"/reference/sessions\"},{\"text\":\"Playground\",\"link\":\"/reference/playground\"},{\"text\":\"CLI\",\"link\":\"/reference/cli\"},{\"text\":\"HTTP API\",\"link\":\"/reference/http-api\"}]}]},\"locales\":{},\"scrollOffset\":134,\"cleanUrls\":false}");</script>
     
   </body>
 </html>