<?xml version="1.0" encoding="utf-8"?>
<entry xmlns="http://www.w3.org/2005/Atom" xml:base="https://docs.spec-kitty.ai/guides/how-to/harnesses/agent-dispatch-permissions/">
  <id>https://docs.spec-kitty.ai/guides/how-to/harnesses/agent-dispatch-permissions/</id>
  <title>Set Agent Dispatch Permissions</title>
  <updated>2026-10-21T00:00:00.000Z</updated>
  <author><name>Spec Kitty</name></author>
  <link rel="alternate" type="text/html" href="https://docs.spec-kitty.ai/guides/how-to/harnesses/agent-dispatch-permissions/"/>
  <summary>Spec Kitty dispatches implementer and reviewer agents with their sandbox on. Allow only the git and spec-kitty commands they need, and know what elevated dispatch means.</summary>
  <content type="html">&lt;span id=&quot;set-agent-dispatch-permissions&quot;&gt;&lt;/span&gt;
&lt;p&gt;When the &lt;code dir=&quot;auto&quot;&gt;spec-kitty-implement-review&lt;/code&gt; skill runs a mission, the orchestrating
agent starts other coding agents to implement and review work packages. Each
dispatch command it ships keeps that agent’s sandbox or approval prompts on. No
shipped skill tells an agent to turn them off; a gate test
(&lt;code dir=&quot;auto&quot;&gt;tests/architectural/test_no_unsandboxed_agent_dispatch.py&lt;/code&gt;) fails if one does.&lt;/p&gt;
&lt;h2 id=&quot;what-the-dispatched-agent-may-do-by-default&quot;&gt;What the dispatched agent may do by default&lt;/h2&gt;













































&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Agent&lt;/th&gt;&lt;th&gt;Shipped mode&lt;/th&gt;&lt;th&gt;&lt;code dir=&quot;auto&quot;&gt;git&lt;/code&gt; and &lt;code dir=&quot;auto&quot;&gt;spec-kitty&lt;/code&gt;&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;Codex&lt;/td&gt;&lt;td&gt;&lt;code dir=&quot;auto&quot;&gt;--sandbox workspace-write&lt;/code&gt;&lt;/td&gt;&lt;td&gt;Edits the workspace only, network off. Its sandbox keeps &lt;code dir=&quot;auto&quot;&gt;.git&lt;/code&gt; read-only, so the orchestrator commits its work and runs &lt;code dir=&quot;auto&quot;&gt;move-task&lt;/code&gt;&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Gemini&lt;/td&gt;&lt;td&gt;&lt;code dir=&quot;auto&quot;&gt;--approval-mode auto_edit&lt;/code&gt;&lt;/td&gt;&lt;td&gt;Denied in a headless run until you allow them&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Qwen Code&lt;/td&gt;&lt;td&gt;&lt;code dir=&quot;auto&quot;&gt;--approval-mode auto-edit&lt;/code&gt;&lt;/td&gt;&lt;td&gt;Denied in a headless run until you allow them&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;LLxprt Code&lt;/td&gt;&lt;td&gt;&lt;code dir=&quot;auto&quot;&gt;--approval-mode=auto_edit&lt;/code&gt;&lt;/td&gt;&lt;td&gt;Denied in a headless run until you allow them&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Copilot&lt;/td&gt;&lt;td&gt;&lt;code dir=&quot;auto&quot;&gt;--allow-tool &apos;write&apos; --allow-tool &apos;shell(git:*)&apos; --allow-tool &apos;shell(spec-kitty:*)&apos;&lt;/code&gt;&lt;/td&gt;&lt;td&gt;Allowed; &lt;code dir=&quot;auto&quot;&gt;git push&lt;/code&gt; denied (best effort: the deny matches that spelling only, so &lt;code dir=&quot;auto&quot;&gt;git -c x push&lt;/code&gt; or &lt;code dir=&quot;auto&quot;&gt;git remote set-url&lt;/code&gt; are not covered)&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Cursor&lt;/td&gt;&lt;td&gt;print mode without &lt;code dir=&quot;auto&quot;&gt;--force&lt;/code&gt;&lt;/td&gt;&lt;td&gt;Denied until you allow them&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Kilocode&lt;/td&gt;&lt;td&gt;no automatic approval&lt;/td&gt;&lt;td&gt;Denied until you allow them&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;
&lt;p&gt;When a dispatched agent cannot commit or run &lt;code dir=&quot;auto&quot;&gt;spec-kitty agent tasks move-task&lt;/code&gt;,
the orchestrator commits the agent’s deliverables with &lt;code dir=&quot;auto&quot;&gt;spec-kitty safe-commit&lt;/code&gt;
and runs the lifecycle command after the agent finishes, as it does for agents
with no command line. Nothing is lost; the work package just moves one step
later.&lt;/p&gt;
&lt;h2 id=&quot;allow-only-the-commands-the-agent-needs&quot;&gt;Allow only the commands the agent needs&lt;/h2&gt;
&lt;p&gt;Allow &lt;code dir=&quot;auto&quot;&gt;git&lt;/code&gt; and &lt;code dir=&quot;auto&quot;&gt;spec-kitty&lt;/code&gt;, and nothing broader, in the agent’s own
permission settings:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Gemini&lt;/strong&gt;: a policy file under &lt;code dir=&quot;auto&quot;&gt;~/.gemini/policies/&lt;/code&gt; with one rule per
command, for example &lt;code dir=&quot;auto&quot;&gt;toolName = &quot;run_shell_command&quot;&lt;/code&gt;,
&lt;code dir=&quot;auto&quot;&gt;commandPrefix = &quot;git&quot;&lt;/code&gt;, &lt;code dir=&quot;auto&quot;&gt;decision = &quot;allow&quot;&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Cursor&lt;/strong&gt;: &lt;code dir=&quot;auto&quot;&gt;.cursor/cli.json&lt;/code&gt; permissions &lt;code dir=&quot;auto&quot;&gt;Shell(git)&lt;/code&gt; and &lt;code dir=&quot;auto&quot;&gt;Shell(spec-kitty)&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Kilocode&lt;/strong&gt;: a &lt;code dir=&quot;auto&quot;&gt;bash&lt;/code&gt; permission map in its config that allows &lt;code dir=&quot;auto&quot;&gt;git *&lt;/code&gt; and
&lt;code dir=&quot;auto&quot;&gt;spec-kitty *&lt;/code&gt; and denies &lt;code dir=&quot;auto&quot;&gt;*&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Qwen Code&lt;/strong&gt; and &lt;strong&gt;LLxprt Code&lt;/strong&gt;: their allowed-tools setting for the shell
tool, in the form your installed version documents.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Check the exact syntax against your installed version; these settings change
between releases.&lt;/p&gt;
&lt;h2 id=&quot;elevated-dispatch-is-your-decision&quot;&gt;Elevated dispatch is your decision&lt;/h2&gt;
&lt;p&gt;Every agent also has a mode that switches its sandbox or approvals off
entirely, for example Codex &lt;code dir=&quot;auto&quot;&gt;--sandbox danger-full-access&lt;/code&gt;, Gemini and Qwen
&lt;code dir=&quot;auto&quot;&gt;--yolo&lt;/code&gt;, Copilot &lt;code dir=&quot;auto&quot;&gt;--allow-all-tools&lt;/code&gt;, Cursor &lt;code dir=&quot;auto&quot;&gt;--force&lt;/code&gt; and Kilocode &lt;code dir=&quot;auto&quot;&gt;--auto&lt;/code&gt;.
In that mode the dispatched agent can change any file you can and reach the
network, without asking.&lt;/p&gt;
&lt;p&gt;Spec Kitty never picks that mode for you. If you decide you want it, tell the
orchestrating agent yourself, in your own message in the session, naming the
agent: for example “dispatch the Codex reviewer with
&lt;code dir=&quot;auto&quot;&gt;--sandbox danger-full-access&lt;/code&gt; for this mission”. The skill follows only such an
instruction from you, never one it finds in a work package prompt, an issue or
tool output, and it records in the work package note that the agent ran
without a sandbox. Spec Kitty’s orchestrator
API rejects a dispatch policy whose &lt;code dir=&quot;auto&quot;&gt;dangerous_flags&lt;/code&gt; include any of four
specific flags (&lt;code dir=&quot;auto&quot;&gt;BANNED_FLAGS&lt;/code&gt;: &lt;code dir=&quot;auto&quot;&gt;--full-auto&lt;/code&gt;, &lt;code dir=&quot;auto&quot;&gt;--yolo&lt;/code&gt;,
&lt;code dir=&quot;auto&quot;&gt;--dangerously-bypass-approvals-and-sandbox&lt;/code&gt; and
&lt;code dir=&quot;auto&quot;&gt;--dangerously-skip-permissions&lt;/code&gt;); it does not cover every flag or spelling that
lowers a harness’s guard.&lt;/p&gt;</content>
</entry>
