@cursor/july 0.1.8 → 0.1.9

package/dist/docs/assets/{example-agents_security-reviewer.md.D2rtwDTO.js → example-agents_security-reviewer.md.CE9v-aKm.js}

@@ -1,4 +1,4 @@
-import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const u=JSON.parse('{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics.","frontmatter":{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics."},"headers":[],"relativePath":"example-agents/security-reviewer.md","filePath":"example-agents/security-reviewer.md"}'),r={name:"example-agents/security-reviewer.md"};function n(l,e,o,h,d,p){return a(),t("div",null,[...e[0]||(e[0]=[i(`<h1 id="run-staged-security-reviews-from-github-events" tabindex="-1">Run staged security reviews from GitHub events <a class="header-anchor" href="#run-staged-security-reviews-from-github-events" aria-label="Permalink to &quot;Run staged security reviews from GitHub events&quot;">​</a></h1><p>Security Reviewer turns a pull request into a staged host-side review. One tool prepares the diff and selects modules. A second fans out specialized reviewers and triages candidates as they arrive. A third deduplicates the confirmed findings, writes artifacts, and may publish a GitHub review.</p><p>Use this example when the workflow needs several model workers, but the host must own orchestration, progress, artifacts, and the final write.</p><p><a href="./../../examples/security-reviewer/">Browse the Security Reviewer source.</a></p><h2 id="run-a-three-stage-host-pipeline" tabindex="-1">Run a three-stage host pipeline <a class="header-anchor" href="#run-a-three-stage-host-pipeline" aria-label="Permalink to &quot;Run a three-stage host pipeline&quot;">​</a></h2><p>Security Reviewer is a pipeline, not one long agent turn:</p><table tabindex="0"><thead><tr><th>Stage</th><th>Tool</th><th>Result</th></tr></thead><tbody><tr><td>Prepare</td><td><code>prepare_review</code></td><td>Fetch metadata and diff, create a <code>runId</code>, and select security modules.</td></tr><tr><td>Review and triage</td><td><code>run_reviewers</code></td><td>Run module reviewers in parallel and start triage as each candidate arrives.</td></tr><tr><td>Finalize</td><td><code>finalize_review</code></td><td>Apply thresholds, deduplicate findings, write artifacts, and optionally post a review.</td></tr></tbody></table><p><code>run_triage</code> remains available as a compatibility stage. In the normal flow, triage has already completed inside <code>run_reviewers</code>, so it reports existing results. If candidates exist without triage output, it starts triage workers and writes their state.</p><p>The configured root agent chooses and sequences tools in chat. The review workers use a model selected by the host pipeline. They are created programmatically with the agent SDK, not discovered from <code>agent/subagents/</code>.</p><h2 id="follow-a-github-wake" tabindex="-1">Follow a GitHub wake <a class="header-anchor" href="#follow-a-github-wake" aria-label="Permalink to &quot;Follow a GitHub wake&quot;">​</a></h2><ol><li>A non-draft <code>pull_request.opened</code> or <code>pull_request.synchronize</code> event arrives for an allowlisted repository.</li><li>The GitHub channel returns a host <code>{ task }</code>, so the webhook gets a 202 response before the long review starts.</li><li>Host code tries to post a pending commit status and creates a playground session.</li><li>The root model sends one acknowledgement. It doesn&#39;t run review tools on this path.</li><li>The task calls <code>prepare_review</code>, <code>run_reviewers</code>, and <code>finalize_review</code> deterministically inside that session.</li><li>Reviewer candidates stream into duplicate gating and triage.</li><li>Finalization writes artifacts and tries to post the GitHub review.</li><li>The host tries to set a success commit status when no findings remain, a failure status when findings remain, or an error status when the pipeline throws.</li><li>The final response is appended to the session as an assistant message.</li></ol><p>The session records each stage as a normal tool event, even though host code selected the tools.</p><p>Review and status posting are best-effort. The channel chooses its final status from the finding count even when the review posting result says <code>posted: false</code>.</p><h2 id="map-the-framework-features" tabindex="-1">Map the framework features <a class="header-anchor" href="#map-the-framework-features" aria-label="Permalink to &quot;Map the framework features&quot;">​</a></h2><table tabindex="0"><thead><tr><th>Capability</th><th>Source</th><th>Role</th></tr></thead><tbody><tr><td>Root agent</td><td><a href="../../examples/security-reviewer/agent/agent.ts"><code>agent/agent.ts</code></a>, <a href="./../../examples/security-reviewer/agent/instructions.html"><code>agent/instructions.md</code></a></td><td>Configure local chat and explain the three-stage contract.</td></tr><tr><td>Server tools</td><td><a href="./../../examples/security-reviewer/agent/tools/"><code>agent/tools/</code></a></td><td>Expose each review stage to chat and host orchestration.</td></tr><tr><td>GitHub channel</td><td><a href="../../examples/security-reviewer/agent/channels/github.ts"><code>agent/channels/github.ts</code></a></td><td>Filter wakes, run background tasks, and publish status.</td></tr><tr><td>Progress channel</td><td><a href="../../examples/security-reviewer/agent/channels/asr-progress.ts"><code>agent/channels/asr-progress.ts</code></a></td><td>Serve live reviewer and triage state by <code>runId</code>.</td></tr><tr><td>Playground renderer</td><td><a href="./../../examples/security-reviewer/agent/playground/tools/run_reviewers.tsx.html"><code>agent/playground/tools/run_reviewers.tsx</code></a></td><td>Replace the generic tool chip with live module rows.</td></tr><tr><td>SDK review pipeline</td><td><a href="./../../examples/security-reviewer/agent/lib/asr/"><code>agent/lib/asr/</code></a>, <a href="../../examples/security-reviewer/agent/lib/review-stages.ts"><code>review-stages.ts</code></a></td><td>Select modules, call model workers, triage, deduplicate, and write artifacts.</td></tr><tr><td>Storage</td><td><a href="../../examples/security-reviewer/agent/storage.ts"><code>agent/storage.ts</code></a></td><td>Persist framework sessions through a custom adapter with lazy restore.</td></tr><tr><td>A/B</td><td><a href="../../examples/security-reviewer/agent/ab.ts"><code>agent/ab.ts</code></a>, <a href="../../examples/security-reviewer/agent/ab.config.ts"><code>agent/ab.config.ts</code></a></td><td>Compare all-severity versus high-only GitHub comments.</td></tr><tr><td>Eval</td><td><a href="./../../examples/security-reviewer/evals/"><code>evals/</code></a></td><td>Check stage-tool presence against a pinned sample.</td></tr></tbody></table><p>There is no Slack channel, authored skill, discovered subagent, MCP connection, schedule, reminder, hook, tool approval, or cloud runtime.</p><h2 id="prepare-the-host" tabindex="-1">Prepare the host <a class="header-anchor" href="#prepare-the-host" aria-label="Permalink to &quot;Prepare the host&quot;">​</a></h2><p>You need:</p><ul><li>Node 22.13 or newer.</li><li>An agent-runtime credential for the root turn and review workers.</li><li>GitHub read access for preparation.</li><li>GitHub write access for webhook-driven reviews and commit statuses.</li></ul><p>The pipeline exposes settings for:</p><ul><li>the worker model,</li><li>reviewer and triage parallelism,</li><li>reviewer, triage, duplicate-gate, and final-dedupe timeouts, and</li><li>prior-comment loading.</li></ul><p>The active names live beside the orchestration in <a href="../../examples/security-reviewer/agent/lib/review-stages.ts"><code>review-stages.ts</code></a>.</p><h2 id="validate-the-discovered-agent" tabindex="-1">Validate the discovered agent <a class="header-anchor" href="#validate-the-discovered-agent" aria-label="Permalink to &quot;Validate the discovered agent&quot;">​</a></h2><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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> validate</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span></span>
+import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const u=JSON.parse('{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics.","frontmatter":{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics."},"headers":[],"relativePath":"example-agents/security-reviewer.md","filePath":"example-agents/security-reviewer.md"}'),r={name:"example-agents/security-reviewer.md"};function n(l,e,o,d,h,p){return a(),t("div",null,[...e[0]||(e[0]=[i(`<h1 id="run-staged-security-reviews-from-github-events" tabindex="-1">Run staged security reviews from GitHub events <a class="header-anchor" href="#run-staged-security-reviews-from-github-events" aria-label="Permalink to &quot;Run staged security reviews from GitHub events&quot;">​</a></h1><p>Security Reviewer turns a pull request into a staged host-side review. One tool prepares the diff and selects modules. A second fans out specialized reviewers and triages candidates as they arrive. A third deduplicates the confirmed findings, writes artifacts, and may publish a GitHub review.</p><p>Use this example when the workflow needs several model workers, but the host must own orchestration, progress, artifacts, and the final write.</p><p><a href="./../../examples/security-reviewer/">Browse the Security Reviewer source.</a></p><h2 id="run-a-three-stage-host-pipeline" tabindex="-1">Run a three-stage host pipeline <a class="header-anchor" href="#run-a-three-stage-host-pipeline" aria-label="Permalink to &quot;Run a three-stage host pipeline&quot;">​</a></h2><p>Security Reviewer is a pipeline, not one long agent turn:</p><table tabindex="0"><thead><tr><th>Stage</th><th>Tool</th><th>Result</th></tr></thead><tbody><tr><td>Prepare</td><td><code>prepare_review</code></td><td>Fetch metadata and diff, create a <code>runId</code>, and select security modules.</td></tr><tr><td>Review and triage</td><td><code>run_reviewers</code></td><td>Run module reviewers in parallel and start triage as each candidate arrives.</td></tr><tr><td>Finalize</td><td><code>finalize_review</code></td><td>Apply thresholds, deduplicate findings, write artifacts, and optionally post a review.</td></tr></tbody></table><p><code>run_triage</code> remains available as a compatibility stage. In the normal flow, triage has already completed inside <code>run_reviewers</code>, so it reports existing results. If candidates exist without triage output, it starts triage workers and writes their state.</p><p>The configured root agent chooses and sequences tools in chat. The review workers use a model selected by the host pipeline. They are created programmatically with the agent SDK, not discovered from <code>agent/subagents/</code>.</p><h2 id="follow-a-github-wake" tabindex="-1">Follow a GitHub wake <a class="header-anchor" href="#follow-a-github-wake" aria-label="Permalink to &quot;Follow a GitHub wake&quot;">​</a></h2><ol><li>A non-draft <code>pull_request.opened</code> or <code>pull_request.synchronize</code> event arrives for an allowlisted repository.</li><li>The GitHub channel returns a host <code>{ task }</code>, so the webhook gets a 202 response before the long review starts.</li><li>Host code tries to post a pending commit status and creates a playground session.</li><li>The root model sends one acknowledgement. It doesn&#39;t run review tools on this path.</li><li>The task calls <code>prepare_review</code>, <code>run_reviewers</code>, and <code>finalize_review</code> deterministically inside that session.</li><li>Reviewer candidates stream into duplicate gating and triage.</li><li>Finalization writes artifacts and tries to post the GitHub review.</li><li>The host tries to set a success commit status when no findings remain, a failure status when findings remain, or an error status when the pipeline throws.</li><li>The final response is appended to the session as an assistant message.</li></ol><p>The session records each stage as a normal tool event, even though host code selected the tools.</p><p>Review and status posting are best-effort. The channel chooses its final status from the finding count even when the review posting result says <code>posted: false</code>.</p><h2 id="map-the-framework-features" tabindex="-1">Map the framework features <a class="header-anchor" href="#map-the-framework-features" aria-label="Permalink to &quot;Map the framework features&quot;">​</a></h2><table tabindex="0"><thead><tr><th>Capability</th><th>Source</th><th>Role</th></tr></thead><tbody><tr><td>Root agent</td><td><a href="../../examples/security-reviewer/agent/agent.ts"><code>agent/agent.ts</code></a>, <a href="./../../examples/security-reviewer/agent/instructions.html"><code>agent/instructions.md</code></a></td><td>Configure local chat and explain the three-stage contract.</td></tr><tr><td>Server tools</td><td><a href="./../../examples/security-reviewer/agent/tools/"><code>agent/tools/</code></a></td><td>Expose each review stage to chat and host orchestration.</td></tr><tr><td>GitHub channel</td><td><a href="../../examples/security-reviewer/agent/channels/github.ts"><code>agent/channels/github.ts</code></a></td><td>Filter wakes, run background tasks, and publish status.</td></tr><tr><td>Progress channel</td><td><a href="../../examples/security-reviewer/agent/channels/asr-progress.ts"><code>agent/channels/asr-progress.ts</code></a></td><td>Serve live reviewer and triage state by <code>runId</code>.</td></tr><tr><td>Playground renderer</td><td><a href="./../../examples/security-reviewer/agent/playground/tools/run_reviewers.tsx.html"><code>agent/playground/tools/run_reviewers.tsx</code></a></td><td>Replace the generic tool chip with live module rows.</td></tr><tr><td>SDK review pipeline</td><td><a href="./../../examples/security-reviewer/agent/lib/asr/"><code>agent/lib/asr/</code></a>, <a href="../../examples/security-reviewer/agent/lib/review-stages.ts"><code>review-stages.ts</code></a></td><td>Select modules, call model workers, triage, deduplicate, and write artifacts.</td></tr><tr><td>Storage</td><td><a href="../../examples/security-reviewer/agent/storage.ts"><code>agent/storage.ts</code></a></td><td>Persist framework sessions with <code>cursorHostedStorage</code> (lazy restore).</td></tr><tr><td>A/B</td><td><a href="../../examples/security-reviewer/agent/ab.ts"><code>agent/ab.ts</code></a>, <a href="../../examples/security-reviewer/agent/ab.config.ts"><code>agent/ab.config.ts</code></a></td><td>Compare all-severity versus high-only GitHub comments.</td></tr><tr><td>Eval</td><td><a href="./../../examples/security-reviewer/evals/"><code>evals/</code></a></td><td>Check stage-tool presence against a pinned sample.</td></tr></tbody></table><p>There is no Slack channel, authored skill, discovered subagent, MCP connection, schedule, reminder, hook, tool approval, or cloud runtime.</p><h2 id="prepare-the-host" tabindex="-1">Prepare the host <a class="header-anchor" href="#prepare-the-host" aria-label="Permalink to &quot;Prepare the host&quot;">​</a></h2><p>You need:</p><ul><li>Node 22.13 or newer.</li><li>An agent-runtime credential for the root turn and review workers.</li><li>GitHub read access for preparation.</li><li>GitHub write access for webhook-driven reviews and commit statuses.</li></ul><p>The pipeline exposes settings for:</p><ul><li>the worker model,</li><li>reviewer and triage parallelism,</li><li>reviewer, triage, duplicate-gate, and final-dedupe timeouts, and</li><li>prior-comment loading.</li></ul><p>The active names live beside the orchestration in <a href="../../examples/security-reviewer/agent/lib/review-stages.ts"><code>review-stages.ts</code></a>.</p><h2 id="validate-the-discovered-agent" tabindex="-1">Validate the discovered agent <a class="header-anchor" href="#validate-the-discovered-agent" aria-label="Permalink to &quot;Validate the discovered agent&quot;">​</a></h2><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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> validate</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> info</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --json</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</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;"> examples/security-reviewer</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --list</span></span></code></pre></div><p>The manifest should show four server tools, two authored channels, one storage definition, and one A/B experiment. The eval listing should show one case.</p><h2 id="know-the-chat-path-s-write-boundary" tabindex="-1">Know the chat path&#39;s write boundary <a class="header-anchor" href="#know-the-chat-path-s-write-boundary" aria-label="Permalink to &quot;Know the chat path&#39;s write boundary&quot;">​</a></h2><p>In chat, the root instructions ask the model to use this order:</p><div class="language-text vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">text</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span>prepare_review -&gt; run_reviewers -&gt; finalize_review</span></span></code></pre></div><p>They also ask the model to set <code>postComment: true</code> only on request. This is prompt policy, not a deterministic safety gate. The model chooses tool arguments, and <code>finalize_review</code> has no human approval. Use the direct stage calls below when a no-post proof must be enforced.</p><h2 id="call-stages-directly-without-publishing" tabindex="-1">Call stages directly without publishing <a class="header-anchor" href="#call-stages-directly-without-publishing" aria-label="Permalink to &quot;Call stages directly without publishing&quot;">​</a></h2><p>Call each stage and pass <code>postComment: false</code> yourself:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> call</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> prepare_review</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
@@ -10,7 +10,7 @@ import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const u
 <span class="line"></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> call</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> finalize_review</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
-<span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --input</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &#39;{&quot;runId&quot;:&quot;&lt;run-id&gt;&quot;,&quot;postComment&quot;:false}&#39;</span></span></code></pre></div><p>Review state lives under the project&#39;s run-artifact directory, so later stages can open the prepared <code>runId</code>.</p><div class="caution custom-block github-alert"><p class="custom-block-title">CAUTION</p><p><code>finalize_review</code> with <code>postComment: true</code> writes to GitHub. The webhook path always requests that write. Chat instructions alone don&#39;t prevent it.</p></div><h2 id="watch-parallel-work-in-the-playground" tabindex="-1">Watch parallel work in the playground <a class="header-anchor" href="#watch-parallel-work-in-the-playground" aria-label="Permalink to &quot;Watch parallel work in the playground&quot;">​</a></h2><p>Run the dev server:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> dev</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span></span></code></pre></div><p>Open the printed playground and start a review. The custom <code>run_reviewers</code> renderer polls the progress channel&#39;s <code>GET /:runId</code> route.</p><p>It refreshes every 500 ms while the stage runs. Each row shows a reviewer module&#39;s state, candidates, reviewed areas, and failure. A second section shows triage jobs and confirmed or rejected counts.</p><p>This is an authored playground extension. Agentkit discovers it by the tool name, so the generic <code>run_reviewers</code> chip becomes a domain-specific view without changing the framework playground.</p><h2 id="fan-out-reviewers-while-triage-starts" tabindex="-1">Fan out reviewers while triage starts <a class="header-anchor" href="#fan-out-reviewers-while-triage-starts" aria-label="Permalink to &quot;Fan out reviewers while triage starts&quot;">​</a></h2><p>Module selection uses repository and path rules. The current module set covers:</p><ul><li>agent tooling trust boundaries,</li><li>privileged service RPCs,</li><li>product-specific security risks,</li><li>dependency and supply-chain changes,</li><li>deployment and infrastructure code,</li><li>filesystem and workspace boundaries,</li><li>privacy, and</li><li>general security review.</li></ul><p>Selected modules may run more than once. Candidates pass through a duplicate gate, then bounded triage. Reviewer or triage failures can produce partial results. A final dedupe failure stops finalization.</p><p>The pipeline writes JSONL journals as work completes. Final artifacts include the review bundle, patch, reviewer outputs, candidates, triage decisions, findings, accounting, and audit events.</p><h2 id="separate-session-storage-from-review-artifacts" tabindex="-1">Separate session storage from review artifacts <a class="header-anchor" href="#separate-session-storage-from-review-artifacts" aria-label="Permalink to &quot;Separate session storage from review artifacts&quot;">​</a></h2><p><code>defineStorage</code> sends agentkit session and event records to a custom durable store. Security Reviewer sets <code>restore: &quot;off&quot;</code> so startup doesn&#39;t load old review sessions in bulk. A continuation lookup can still fetch a needed session.</p><p>The staged review files are separate from session storage. Session-store durability doesn&#39;t preserve those files. All stages for one <code>runId</code> must see the same filesystem.</p><p>This split is useful when conversation history needs shared durability but large review artifacts belong on attached storage or an object store.</p><h2 id="compare-live-comment-variants" tabindex="-1">Compare live comment variants <a class="header-anchor" href="#compare-live-comment-variants" aria-label="Permalink to &quot;Compare live comment variants&quot;">​</a></h2><p>The comment-severity experiment uses sticky session assignment with a 5% holdout:</p><ul><li><code>control</code> posts every finding.</li><li><code>treatment</code> posts only high and critical findings.</li></ul><p>Finalization enforces the comment filter. The treatment also adds an instruction overlay asking chat and playground summaries to lead with high and critical findings. Full artifacts, <code>finalResponse</code>, finding counts, and status still include every finding. Stage-tool counters appear in the playground A/B view. Local sample and snapshot files persist under <code>.agent-serve/</code>.</p><p>When a treatment session has only low or medium findings, the filtered review body currently says no vulnerabilities were found even though artifacts and status retain findings. Account for that mismatch before using this experiment as a publishing policy.</p><p>Eval sessions skip A/B enrollment.</p><h2 id="test-the-github-channel-carefully" tabindex="-1">Test the GitHub channel carefully <a class="header-anchor" href="#test-the-github-channel-carefully" aria-label="Permalink to &quot;Test the GitHub channel carefully&quot;">​</a></h2><p>The channel currently accepts two configured repositories. It wakes on <code>opened</code> and <code>synchronize</code>, skips drafts, and requests <code>pr-write</code> access.</p><p>Inspect its event surface:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> github</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> events</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
+<span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --input</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &#39;{&quot;runId&quot;:&quot;&lt;run-id&gt;&quot;,&quot;postComment&quot;:false}&#39;</span></span></code></pre></div><p>Review state lives under the project&#39;s run-artifact directory, so later stages can open the prepared <code>runId</code>.</p><div class="caution custom-block github-alert"><p class="custom-block-title">CAUTION</p><p><code>finalize_review</code> with <code>postComment: true</code> writes to GitHub. The webhook path always requests that write. Chat instructions alone don&#39;t prevent it.</p></div><h2 id="watch-parallel-work-in-the-playground" tabindex="-1">Watch parallel work in the playground <a class="header-anchor" href="#watch-parallel-work-in-the-playground" aria-label="Permalink to &quot;Watch parallel work in the playground&quot;">​</a></h2><p>Run the dev server:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> dev</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span></span></code></pre></div><p>Open the printed playground and start a review. The custom <code>run_reviewers</code> renderer polls the progress channel&#39;s <code>GET /:runId</code> route.</p><p>It refreshes every 500 ms while the stage runs. Each row shows a reviewer module&#39;s state, candidates, reviewed areas, and failure. A second section shows triage jobs and confirmed or rejected counts.</p><p>This is an authored playground extension. Agentkit discovers it by the tool name, so the generic <code>run_reviewers</code> chip becomes a domain-specific view without changing the framework playground.</p><h2 id="fan-out-reviewers-while-triage-starts" tabindex="-1">Fan out reviewers while triage starts <a class="header-anchor" href="#fan-out-reviewers-while-triage-starts" aria-label="Permalink to &quot;Fan out reviewers while triage starts&quot;">​</a></h2><p>Module selection uses repository and path rules. The current module set covers:</p><ul><li>agent tooling trust boundaries,</li><li>privileged service RPCs,</li><li>product-specific security risks,</li><li>dependency and supply-chain changes,</li><li>deployment and infrastructure code,</li><li>filesystem and workspace boundaries,</li><li>privacy, and</li><li>general security review.</li></ul><p>Selected modules may run more than once. Candidates pass through a duplicate gate, then bounded triage. Reviewer or triage failures can produce partial results. A final dedupe failure stops finalization.</p><p>The pipeline writes JSONL journals as work completes. Final artifacts include the review bundle, patch, reviewer outputs, candidates, triage decisions, findings, accounting, and audit events.</p><h2 id="separate-session-storage-from-review-artifacts" tabindex="-1">Separate session storage from review artifacts <a class="header-anchor" href="#separate-session-storage-from-review-artifacts" aria-label="Permalink to &quot;Separate session storage from review artifacts&quot;">​</a></h2><p><code>defineStorage</code> + <code>cursorHostedStorage</code> sends agentkit session and event records to Cursor-hosted Bugbot storage through the control-plane proxy. Security Reviewer sets <code>restore: &quot;off&quot;</code> so startup doesn&#39;t load old review sessions in bulk. A continuation lookup can still fetch a needed session.</p><p>The staged review files are separate from session storage. Session-store durability doesn&#39;t preserve those files. All stages for one <code>runId</code> must see the same filesystem.</p><p>This split is useful when conversation history needs shared durability but large review artifacts belong on attached storage or an object store.</p><h2 id="compare-live-comment-variants" tabindex="-1">Compare live comment variants <a class="header-anchor" href="#compare-live-comment-variants" aria-label="Permalink to &quot;Compare live comment variants&quot;">​</a></h2><p>The comment-severity experiment uses sticky session assignment with a 5% holdout:</p><ul><li><code>control</code> posts every finding.</li><li><code>treatment</code> posts only high and critical findings.</li></ul><p>Finalization enforces the comment filter. The treatment also adds an instruction overlay asking chat and playground summaries to lead with high and critical findings. Full artifacts, <code>finalResponse</code>, finding counts, and status still include every finding. Stage-tool counters appear in the playground A/B view. Local sample and snapshot files persist under <code>.agent-serve/</code>.</p><p>When a treatment session has only low or medium findings, the filtered review body currently says no vulnerabilities were found even though artifacts and status retain findings. Account for that mismatch before using this experiment as a publishing policy.</p><p>Eval sessions skip A/B enrollment.</p><h2 id="test-the-github-channel-carefully" tabindex="-1">Test the GitHub channel carefully <a class="header-anchor" href="#test-the-github-channel-carefully" aria-label="Permalink to &quot;Test the GitHub channel carefully&quot;">​</a></h2><p>The channel currently accepts two configured repositories. It wakes on <code>opened</code> and <code>synchronize</code>, skips drafts, and requests <code>pr-write</code> access.</p><p>Inspect its event surface:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> github</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> events</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --dir</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> examples/security-reviewer</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>
 <span class="line"><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;">  --json</span></span></code></pre></div><p>Replay reaches the full publishing path:</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:#24292E;--shiki-dark:#E1E4E8;">TEST_PR_URL</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">=</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">https://github.com/your-org/allowlisted-test-repo/pull/123</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> github</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> replay</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> \\</span></span>

package/dist/docs/assets/{example-agents_security-reviewer.md.D2rtwDTO.lean.js → example-agents_security-reviewer.md.CE9v-aKm.lean.js}

@@ -1 +1 @@
-import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const u=JSON.parse('{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics.","frontmatter":{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics."},"headers":[],"relativePath":"example-agents/security-reviewer.md","filePath":"example-agents/security-reviewer.md"}'),r={name:"example-agents/security-reviewer.md"};function n(l,e,o,h,d,p){return a(),t("div",null,[...e[0]||(e[0]=[i("",73)])])}const g=s(r,[["render",n]]);export{u as __pageData,g as default};
+import{_ as s,c as t,o as a,ag as i}from"./chunks/framework.CAZyNGu9.js";const u=JSON.parse('{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics.","frontmatter":{"title":"Run staged security reviews from GitHub events","description":"Orchestrate parallel reviewer agents, streamed triage, deterministic stages, live playground progress, durable sessions, and severity A/B metrics."},"headers":[],"relativePath":"example-agents/security-reviewer.md","filePath":"example-agents/security-reviewer.md"}'),r={name:"example-agents/security-reviewer.md"};function n(l,e,o,d,h,p){return a(),t("div",null,[...e[0]||(e[0]=[i("",73)])])}const g=s(r,[["render",n]]);export{u as __pageData,g as default};

package/dist/docs/assets/{storage.md.Cu3MwM2a.js → storage.md.DRTdnFvd.js}

@@ -15,4 +15,10 @@ import{_ as t,c as e,o as i,ag as a}from"./chunks/framework.CAZyNGu9.js";const c
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">  },</span></span>
 <span class="line"><span style="--shiki-light:#6A737D;--shiki-dark:#6A737D;">  // put / get / delete / list …</span></span>
 <span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">});</span></span></code></pre></div><table tabindex="0"><thead><tr><th>Knob</th><th>Default</th><th>Meaning</th></tr></thead><tbody><tr><td><code>debounceMs</code></td><td>unset (once per turn)</td><td>Wait this long after activity before writing event batches</td></tr><tr><td><code>restore</code></td><td>caps below</td><td>How much to load at serve start</td></tr><tr><td><code>restore.maxSessions</code></td><td><code>1000</code></td><td>Max sessions loaded at serve start</td></tr><tr><td><code>restore.maxAgeMs</code></td><td>30 days</td><td>Skip older sessions at serve start</td></tr><tr><td><code>restore.maxTotalBytes</code></td><td>1 GiB</td><td>Stop loading once this budget is reached</td></tr></tbody></table><p>Set <code>restore: &quot;off&quot;</code> on high-traffic hosts. Sessions then load when a follow-up arrives instead of at startup.</p><h2 id="restore-after-restart" tabindex="-1">Restore after restart <a class="header-anchor" href="#restore-after-restart" aria-label="Permalink to &quot;Restore after restart&quot;">​</a></h2><p>With <code>get</code> and <code>list</code>, serve can rebuild local state from your store:</p><ul><li>At startup, agentkit loads recent sessions up to the restore caps. Local disk wins when both sides have the same session. Reminders hydrate the same way into <code>--state-root/reminders</code>.</li><li>On demand, a missing continuation token resolves through the store and resumes that session.</li><li>Playground eval history and A/B aggregates can load from the same sink.</li></ul><p>A turn in flight at crash time is not replayed. The next follow-up resumes from the last flushed state.</p><h2 id="author-kv-ctx-host-kv" tabindex="-1">Author KV (<code>ctx.host.kv</code>) <a class="header-anchor" href="#author-kv-ctx-host-kv" aria-label="Permalink to &quot;Author KV (\`ctx.host.kv\`)&quot;">​</a></h2><p>Handlers can store their own JSON under the same sink without minting framework keys:</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;">await</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> ctx.host.kv.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">put</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">&quot;alert-memory/abc&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">, { updated: </span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">&quot;…&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> });</span></span>
-<span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">const</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> prior</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> =</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> await</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> ctx.host.kv.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">get</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">&quot;alert-memory/abc&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">);</span></span></code></pre></div><p>agentkit prefixes author keys as <code>agentkit/v1/{agent}/kv/{key}</code> (same bounded encoding as continuation tokens). Writes <strong>await</strong> the sink and propagate errors — unlike session mirrors, which are at-most-once.</p><p>Without <code>agent/storage.ts</code>, <code>host.kv</code> falls back to files under <code>--state-root/kv</code>. That is fine for local dogfood; it does <strong>not</strong> survive replacing the host. For prod, author <code>defineStorage</code> (for example <code>@cursor/july/storage/postgres-kv</code>) so both sessions and author KV share the durable backend.</p><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/postgres-kv</code></td><td><code>agent_serve_*</code> tables (<code>BUGBOTDB_URL</code>)</td></tr></tbody></table>`,26)])])}const g=t(n,[["render",o]]);export{c as __pageData,g as default};
+<span class="line"><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">const</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> prior</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> =</span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;"> await</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> ctx.host.kv.</span><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">get</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">(</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;">&quot;alert-memory/abc&quot;</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">);</span></span></code></pre></div><p>agentkit prefixes author keys as <code>agentkit/v1/{agent}/kv/{key}</code> (same bounded encoding as continuation tokens). Writes <strong>await</strong> the sink and propagate errors — unlike session mirrors, which are at-most-once.</p><p>Without <code>agent/storage.ts</code>, <code>host.kv</code> falls back to files under <code>--state-root/kv</code>. That is fine for local dogfood; it does <strong>not</strong> survive replacing the host. For Cursor-managed hosting, prefer <code>@cursor/july/storage/cursor-hosted</code> so sessions and author KV share the platform Bugbot tables through a control-plane HTTP proxy (authenticated as the deployment pod credential — engines never receive a database URL). For self-hosted Postgres you own, use <code>@cursor/july/storage/postgres-kv</code> and supply <code>BUGBOTDB_URL</code> / <code>AGENT_SERVE_DEPLOYMENT_ID</code> yourself.</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 — Cursor-managed hosting</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 style="--shiki-light:#D73A49;--shiki-dark:#F97583;">import</span><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;"> { cursorHostedStorage } </span><span style="--shiki-light:#D73A49;--shiki-dark:#F97583;">from</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &quot;@cursor/july/storage/cursor-hosted&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>
+<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><tr><td><code>@cursor/july/storage/postgres-kv</code></td><td>Same tables; you supply <code>BUGBOTDB_URL</code> / <code>AGENT_SERVE_DEPLOYMENT_ID</code></td></tr></tbody></table>`,27)])])}const g=t(n,[["render",o]]);export{c as __pageData,g as default};

package/dist/docs/assets/{storage.md.Cu3MwM2a.lean.js → storage.md.DRTdnFvd.lean.js}

@@ -1 +1 @@
-import{_ as t,c as e,o as i,ag as a}from"./chunks/framework.CAZyNGu9.js";const c=JSON.parse(`{"title":"Storage","description":"Point agentkit's durable storage at a backend you own with defineStorage.","frontmatter":{"title":"Storage","description":"Point agentkit's durable storage at a backend you own with defineStorage."},"headers":[],"relativePath":"storage.md","filePath":"storage.md"}`),n={name:"storage.md"};function o(h,s,r,d,l,p){return i(),e("div",null,[...s[0]||(s[0]=[a("",26)])])}const g=t(n,[["render",o]]);export{c as __pageData,g as default};
+import{_ as t,c as e,o as i,ag as a}from"./chunks/framework.CAZyNGu9.js";const c=JSON.parse(`{"title":"Storage","description":"Point agentkit's durable storage at a backend you own with defineStorage.","frontmatter":{"title":"Storage","description":"Point agentkit's durable storage at a backend you own with defineStorage."},"headers":[],"relativePath":"storage.md","filePath":"storage.md"}`),n={name:"storage.md"};function o(h,s,r,d,l,p){return i(),e("div",null,[...s[0]||(s[0]=[a("",27)])])}const g=t(n,[["render",o]]);export{c as __pageData,g as default};

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

@@ -9,9 +9,9 @@
     <link rel="preload stylesheet" href="/docs/assets/style.Bj-IXa5W.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.BP8mAVMW.js"></script>
+    <script type="module" src="/docs/assets/app.Ciosf24M.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.Dm1R02v9.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.CMpbjdfx.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
     <link rel="modulepreload" href="/docs/assets/building-with-agents.md.CJCtZCyi.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;">agentkit</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;">agentkit</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-53b9fda6 data-v-3958cc3f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-3958cc3f><span class="visually-hidden" id="doc-footer-aria-label" data-v-3958cc3f>Pager</span><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link prev" href="/docs/concepts.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Previous page</span><span class="title" data-v-3958cc3f>Concepts</span><!--]--></a></div><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link next" href="/docs/evals.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Next page</span><span class="title" data-v-3958cc3f>Evals</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"BtfEsc9S\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"8R5phXb5\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"CwtWu5ix\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"D2rtwDTO\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"Cu3MwM2a\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"DC5-ImcM\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"DFGBYLcc\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"BPXjUtRB\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"CE9v-aKm\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"DRTdnFvd\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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.Bj-IXa5W.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.BP8mAVMW.js"></script>
+    <script type="module" src="/docs/assets/app.Ciosf24M.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.Dm1R02v9.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.CMpbjdfx.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
     <link rel="modulepreload" href="/docs/assets/concepts.md.Cfb9b-k1.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-53b9fda6 data-v-3958cc3f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-3958cc3f><span class="visually-hidden" id="doc-footer-aria-label" data-v-3958cc3f>Pager</span><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link prev" href="/docs/scaffolding-agents.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Previous page</span><span class="title" data-v-3958cc3f>Scaffold an agent with Cursor</span><!--]--></a></div><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link next" href="/docs/building-with-agents.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Next page</span><span class="title" data-v-3958cc3f>Building agents with agents</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"BtfEsc9S\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"8R5phXb5\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"CwtWu5ix\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"D2rtwDTO\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"Cu3MwM2a\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"DC5-ImcM\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"DFGBYLcc\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"BPXjUtRB\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"CE9v-aKm\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"DRTdnFvd\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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

@@ -9,11 +9,11 @@
     <link rel="preload stylesheet" href="/docs/assets/style.Bj-IXa5W.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.BP8mAVMW.js"></script>
+    <script type="module" src="/docs/assets/app.Ciosf24M.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.Dm1R02v9.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.CMpbjdfx.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
-    <link rel="modulepreload" href="/docs/assets/deployment.md.BtfEsc9S.lean.js">
+    <link rel="modulepreload" href="/docs/assets/deployment.md.DC5-ImcM.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>
@@ -37,7 +37,7 @@
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> deployment</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> weather-agent</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> logs</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;"> weather-agent</span></span></code></pre></div><h3 id="set-deployment-secrets" tabindex="-1">Set deployment secrets <a class="header-anchor" href="#set-deployment-secrets" aria-label="Permalink to &quot;Set deployment secrets&quot;">​</a></h3><p>A deployment must exist before you can set its secrets. Omit values from the command line to enter them through the hidden prompt:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> secrets</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> set</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> weather-agent</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> WEATHER_API_KEY</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> secrets</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> list</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> weather-agent</span></span>
-<span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> deploy</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;"> --slug</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> weather-agent</span></span></code></pre></div><p>The engine reads secret changes on its next deploy. <code>secrets list</code> returns names and creation times, never values. To set several values from automation, pipe one line per name instead of putting values in shell arguments.</p><h3 id="choose-durable-storage" tabindex="-1">Choose durable storage <a class="header-anchor" href="#choose-durable-storage" aria-label="Permalink to &quot;Choose durable storage&quot;">​</a></h3><p>Hosted filesystem state can reset during a deploy or runtime replacement. Use <a href="./storage.html"><code>defineStorage</code></a> to preserve supported records across deployments. Self-host with a persistent <code>--state-root</code> when the complete filesystem must survive.</p><h3 id="use-the-hosted-agent" tabindex="-1">Use the hosted agent <a class="header-anchor" href="#use-the-hosted-agent" aria-label="Permalink to &quot;Use the hosted agent&quot;">​</a></h3><p>The CLI handles authentication for <code>--prod</code> commands. External clients and managed HTTP channels send <code>X-Agent-Alias-Token</code>; authored channel auth still applies. Use a Cursor relay, Socket Mode, a signature-validating intermediary, or self-host when a webhook provider can&#39;t add this header.</p><p>Use <code>--prod</code> with the normal client commands:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> playground</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;"> weather-agent</span></span>
+<span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> deploy</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;"> --slug</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> weather-agent</span></span></code></pre></div><p>The engine reads secret changes on its next deploy. <code>secrets list</code> returns names and creation times, never values. To set several values from automation, pipe one line per name instead of putting values in shell arguments.</p><h3 id="choose-durable-storage" tabindex="-1">Choose durable storage <a class="header-anchor" href="#choose-durable-storage" aria-label="Permalink to &quot;Choose durable storage&quot;">​</a></h3><p>Hosted filesystem state can reset during a deploy or runtime replacement. Prefer <a href="./storage.html"><code>cursorHostedStorage</code></a> (<code>@cursor/july/storage/cursor-hosted</code>) so durable records land in Cursor&#39;s Bugbot <code>agent_serve_*</code> tables through a control-plane HTTP proxy (pod credential auth — no database URL in the engine). Do not put <code>BUGBOTDB_URL</code> or <code>AGENT_SERVE_DEPLOYMENT_ID</code> in <code>hosting.secretNames</code>. Self-host with your own <code>defineStorage</code> backend or a persistent <code>--state-root</code> when the complete filesystem must survive.</p><h3 id="use-the-hosted-agent" tabindex="-1">Use the hosted agent <a class="header-anchor" href="#use-the-hosted-agent" aria-label="Permalink to &quot;Use the hosted agent&quot;">​</a></h3><p>The CLI handles authentication for <code>--prod</code> commands. External clients and managed HTTP channels send <code>X-Agent-Alias-Token</code>; authored channel auth still applies. Use a Cursor relay, Socket Mode, a signature-validating intermediary, or self-host when a webhook provider can&#39;t add this header.</p><p>Use <code>--prod</code> with the normal client commands:</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;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> playground</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;"> weather-agent</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> chat</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;"> weather-agent</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> run</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;"> weather-agent</span><span style="--shiki-light:#005CC5;--shiki-dark:#79B8FF;"> --message</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> &quot;Forecast for Paris&quot;</span></span>
 <span class="line"><span style="--shiki-light:#6F42C1;--shiki-dark:#B392F0;">agentkit</span><span style="--shiki-light:#032F62;--shiki-dark:#9ECBFF;"> sessions</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;"> weather-agent</span></span>
@@ -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;">$AGENTKIT_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;">agentkit</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>agentkit logs</code></a> for runtime output, <a href="./reference/hooks.html">hooks</a> for metrics export, 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-53b9fda6 data-v-3958cc3f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-3958cc3f><span class="visually-hidden" id="doc-footer-aria-label" data-v-3958cc3f>Pager</span><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link prev" href="/docs/example-agents/codeowners-review.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Previous page</span><span class="title" data-v-3958cc3f>Codeowners review</span><!--]--></a></div><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link next" href="/docs/troubleshooting.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Next page</span><span class="title" data-v-3958cc3f>Fix common problems</span><!--]--></a></div></nav></footer><!--[--><!--]--></div></div></div><!--[--><!--]--></div></div><!----><!--[--><!--]--></div></div>
-    <script>window.__VP_HASH_MAP__=JSON.parse("{\"ab.md\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"BtfEsc9S\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"8R5phXb5\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"CwtWu5ix\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"D2rtwDTO\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"Cu3MwM2a\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"DC5-ImcM\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"DFGBYLcc\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"BPXjUtRB\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"CE9v-aKm\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"DRTdnFvd\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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

@@ -9,9 +9,9 @@
     <link rel="preload stylesheet" href="/docs/assets/style.Bj-IXa5W.css" as="style">
     <link rel="preload stylesheet" href="/docs/vp-icons.css" as="style">
     
-    <script type="module" src="/docs/assets/app.BP8mAVMW.js"></script>
+    <script type="module" src="/docs/assets/app.Ciosf24M.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.Dm1R02v9.js">
+    <link rel="modulepreload" href="/docs/assets/chunks/theme.CMpbjdfx.js">
     <link rel="modulepreload" href="/docs/assets/chunks/framework.CAZyNGu9.js">
     <link rel="modulepreload" href="/docs/assets/evals.md.DAgEc_hL.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>
@@ -103,7 +103,7 @@
 <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;">agentkit</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>); set <code>persistRuns</code> in <code>evals.config.ts</code> if you need them after a serve restart — see <a href="#configure-eval-runs">Configure eval runs</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, and <code>GET /v1/dev/evals/runs/:runId</code> polls it. 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> or <code>failed</code>. Configuration errors appear on a failed snapshot.</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>agentkit 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>agentkit 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><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>agentkit 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-53b9fda6 data-v-3958cc3f><!--[--><!--]--><!----><nav class="prev-next" aria-labelledby="doc-footer-aria-label" data-v-3958cc3f><span class="visually-hidden" id="doc-footer-aria-label" data-v-3958cc3f>Pager</span><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link prev" href="/docs/building-with-agents.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Previous page</span><span class="title" data-v-3958cc3f>Building agents with agents</span><!--]--></a></div><div class="pager" data-v-3958cc3f><a class="VPLink link pager-link next" href="/docs/ab.html" data-v-3958cc3f><!--[--><span class="desc" data-v-3958cc3f>Next page</span><span class="title" data-v-3958cc3f>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\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"BtfEsc9S\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"8R5phXb5\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"CwtWu5ix\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"D2rtwDTO\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"Cu3MwM2a\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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\":\"BMCZ6Hd7\",\"building-with-agents.md\":\"CJCtZCyi\",\"concepts.md\":\"Cfb9b-k1\",\"deployment.md\":\"DC5-ImcM\",\"evals.md\":\"DAgEc_hL\",\"example-agents_approval-buddy.md\":\"DFGBYLcc\",\"example-agents_benny.md\":\"B0gjhI-p\",\"example-agents_bugbot.md\":\"DelIdhxB\",\"example-agents_codebase-wiki.md\":\"DC6sgwn0\",\"example-agents_codeowners-review.md\":\"Ku_tG2RY\",\"example-agents_concierge.md\":\"4rQTSMXt\",\"example-agents_fsd.md\":\"CzgUrDfi\",\"example-agents_index.md\":\"BPXjUtRB\",\"example-agents_knowledge-base.md\":\"BPJiVueF\",\"example-agents_oncall.md\":\"BG_sUMly\",\"example-agents_security-reviewer.md\":\"CE9v-aKm\",\"example-agents_slack-agent.md\":\"buLbgvBf\",\"example-agents_weather-agent.md\":\"C9Qv-W0o\",\"guides_agent-to-agent.md\":\"Bpzgq2Pq\",\"guides_cloud-runtime.md\":\"gVzabdQL\",\"guides_github.md\":\"DOOCpqsW\",\"guides_human-in-the-loop.md\":\"DlUqsp1S\",\"guides_slack.md\":\"FWale6Ip\",\"guides_webhooks.md\":\"B1EswtUu\",\"hillclimbing.md\":\"DHNast08\",\"index.md\":\"COiu-1jL\",\"quickstart.md\":\"BU6Iwi_9\",\"reference_agent-config.md\":\"DrW2JUM8\",\"reference_channels.md\":\"DdmiKgqf\",\"reference_cli.md\":\"D189RBCH\",\"reference_connections.md\":\"zaEYCLHT\",\"reference_hooks.md\":\"DyLVfE1O\",\"reference_http-api.md\":\"Dx_nmDG6\",\"reference_instructions.md\":\"CgoV-YEb\",\"reference_playground.md\":\"CPZhfYaO\",\"reference_project-layout.md\":\"Bd_CKtNS\",\"reference_schedules.md\":\"w_F2mXB6\",\"reference_sessions.md\":\"DLd6mvbv\",\"reference_skills.md\":\"B_jHN7JL\",\"reference_subagents.md\":\"zWAMNfi1\",\"reference_tools.md\":\"CqgJroI0\",\"scaffolding-agents.md\":\"C3pTrmoE\",\"storage.md\":\"DRTdnFvd\",\"troubleshooting.md\":\"DkovjS7h\"}");window.__VP_SITE_DATA__=JSON.parse("{\"lang\":\"en-US\",\"dir\":\"ltr\",\"title\":\"agentkit\",\"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\":\"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\":\"Cloud runtime\",\"link\":\"/guides/cloud-runtime\"}]},{\"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\":\"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\":\"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>