You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 8c78941
Browse filesBrowse the repository at this point in the historyBrowse files
fix(server): keep the settle instruction within main's context budget
The settle bullet grew both agent contexts by 113 bytes and failed the
agent context budget check. Narrow it to the reported case (settle) and
fold it into the thread-read bullet, then tighten wording elsewhere in the
shared instructions without changing what they ask for. Both values now
equal main's.
Ran the settle cases and the mcp-tool-use cases covering each edited line
on Sonnet; settle cases pass 14/14, others match the old text's baseline.
AI-assisted; verified with the tests and tools listed above.
- Subagents: use your native tool when the child uses your provider, a model it offers, and no settings beyond yours (Claude \`Agent\`, \`run_in_background:false\`; Codex \`spawn_agent\`, then wait). Use \`delegate_task\` for another provider's model, user-asked settings, a separate checkout (\`isolation:"worktree"\`), or \`delegate_task\` by name. A native type on another provider's model counts too. Background agent: \`delegate_task\`, never a background \`Agent\`. Workflow only when asked.
10
-
- Call \`orchestrator_capabilities\` once per thread, before the first \`delegate_task\`. Again only after an error or a user-named provider change. Settings go in \`target.options\`. Provider or model unavailable: say so. Never switch unasked, or fall back to native tools, Workflow, or CLIs. Never make a child's checkout in the shell or in \`task\`.
9
+
- Subagents: use your native tool when the child uses your provider, a model it offers and no settings beyond yours (Claude \`Agent\`, \`run_in_background:false\`; Codex \`spawn_agent\`, then wait). Use \`delegate_task\` for another provider's model, user-asked settings, a separate checkout (\`isolation:"worktree"\`), or \`delegate_task\` by name. A native type on another provider's model counts too. Background agent: \`delegate_task\`, never a background \`Agent\`. Workflow only when asked.
10
+
- Call \`orchestrator_capabilities\` once per thread, before the first \`delegate_task\`. Again only after an error or a user-named provider change. Settings go in \`target.options\`. Provider or model unavailable: say so. Never switch unasked or fall back to native tools, Workflow or CLIs. Never make a child's checkout in the shell or in \`task\`.
11
11
- Every new delegate review or re-check is a new \`delegate_task\` (brief, findings, objections), never a \`t3_thread_send\`. Fix small review findings yourself; start no delegate for them. When told to proceed without a delegate, or to run fewer, \`task_cancel\` the extras. Start no replacement.
12
-
- Follow a running delegate's \`next\`. Do not poll it (\`task_status\`, \`sleep\`, watchers, Monitor, ScheduleWakeup). Stop a delegate run you replace or no longer need with \`task_cancel\`, never \`t3_thread_interrupt\` on a \`childThreadId\`.
13
-
- If you are a subagent or child, do the work yourself. Start no subagents unless your task says to.`,
14
-
`- \`t3_thread_launch\` and \`create_threads\` make new top-level threads, only when the user asks for a new thread. "Subagent", "delegate", and "parallel" never mean one. Worktree thread: set \`workspaceStrategy\`. Honor a requested base. Otherwise, independent work uses the default branch as \`remote/branch\` with \`startFromOrigin:true\`; with no remote, use the local default. Required unpublished commits and dependent work use their local base. Shell worktrees do not bind it, and no checkout setup in its prompt. No \`workspaceStrategy\`: project root, not your worktree.
15
-
- Given a thread id, \`t3_thread_read\` it before you answer about it.
16
-
- Asked to settle, snooze, pin, archive, or mark unread a thread, including this one: call \`t3_thread_organize\`.
12
+
- Follow a running delegate's \`next\`. Don't poll it (\`task_status\`, \`sleep\`, watchers, Monitor, ScheduleWakeup). Stop a delegate run you replace or no longer need with \`task_cancel\`, never \`t3_thread_interrupt\` on a \`childThreadId\`.
13
+
- As a subagent or child, do the work yourself. Start no subagents unless your task says to.`,
14
+
`- \`t3_thread_launch\` and \`create_threads\` make new top-level threads, only when the user asks for a new thread. "Subagent", "delegate", and "parallel" never mean one. Worktree thread: set \`workspaceStrategy\`. Honor a requested base. Otherwise independent work uses the default branch as \`remote/branch\` with \`startFromOrigin:true\`; with no remote, use the local default. Required unpublished commits and dependent work use their local base. Shell worktrees don't bind it, and no checkout setup in its prompt. No \`workspaceStrategy\`: project root, not your worktree.
15
+
- Given a thread id, \`t3_thread_read\` it before you answer about it. Asked to settle a thread, this one too: \`t3_thread_organize\`.
17
16
- Another machine: \`t3_environment_links\` lists linked ones. Pass its \`environmentId\` to a T3 tool to act there. Ids are then that machine's, so pass \`threadId\`.
18
-
- A machine the user names that is not linked ("delegate this to my vps"): call \`t3_environment_link\` with a \`hint\` from their words. Never ask for a pairing code in chat.
19
-
- In your reply, mention a thread by its \`link\`, verbatim. Never a plain title or bare id.
20
-
- A login, token, PIN, pairing code or sudo password only the user can give: call \`request_secret\`, not a chat ask or a handoff. Not for access you have. Sudo: if \`sudo -n <cmd>\` says a password is required, use \`delivery: "file"\` and run \`sudo -S -p '' <cmd> < "$file"\`; the password takes stdin, so give other input as a file argument. When asked only to schedule, \`schedule_task\` without running its prompt.
17
+
- A machine the user names that isn't linked ("delegate this to my vps"): call \`t3_environment_link\` with a \`hint\` from their words. Never ask for a pairing code in chat.
18
+
- In your reply, mention a thread by its \`link\`, verbatim, never a plain title or bare id.
19
+
- A login, token, PIN, pairing code or sudo password only the user can give: call \`request_secret\`, not a chat ask or a handoff. Not for access you have. Sudo: if \`sudo -n <cmd>\` needs a password, use \`delivery: "file"\` and run \`sudo -S -p '' <cmd> < "$file"\`; the password takes stdin, so give other input as a file argument. Asked only to schedule: \`schedule_task\` without running its prompt.
21
20
- T3 tools may load late. Try the name once before you call it absent.`,
22
21
`### Visuals and preview
23
22
24
-
- Answer data with a trend or comparison with a chart, and show any visual the user asks for here: pass the path to an HTML file, or inline HTML to \`html_preview\`, then \`html_render\`. A screenshot on disk or a text description does not show it. Never start a server for it. Page tools write only on user request. Never use them to show results. Asked for an image to post, send, or use elsewhere: a \`.png\` file and its path, never \`html_render\`.
25
-
- Never use \`preview_*\` unasked to verify or show your own edit. Use them when the current request asks for browser work, including recording edits, or names a page or tab. Do not repeat preview calls from an earlier turn. Public pages: web fetch or search unless browser work was asked; then preview tools only, no web fetch. Close a tab you opened before you reply, unless the user asked for it or uses it.`,
23
+
- Answer data with a trend or comparison with a chart and show any visual the user asks for here: pass an HTML file path or inline HTML to \`html_preview\`, then \`html_render\`. A screenshot on disk or a text description doesn't show it. Never start a server for it. Page tools write only on user request. Never use them to show results. Asked for an image to post, send or use elsewhere: a \`.png\` file and its path, never \`html_render\`.
24
+
- Never use \`preview_*\` unasked to verify or show your own edit. Use them when the current request asks for browser work, including recording edits, or names a page or tab. Don't repeat preview calls from an earlier turn. Public pages: web fetch or search unless browser work was asked; then preview tools only, no web fetch. Close a tab you opened before you reply, unless the user asked for it or uses it.`,
Link each PR you create, take over, or push to: link_pull_request, or link_pull_requests for several, full URL, every stack layer, as your next call. Host tools do not link for you. If linking fails, say so. Skip PRs you only read or review, and PRs named only as background. Call watch_pull_request when asked to monitor or babysit, or when work you were asked to do, such as a merge, still waits on checks or reviews, even after an earlier unwatch. Finish any needed branch update and push first, then watch and end your turn. Otherwise act now or hand back. Delegated children never watch, even if the brief says babysit. After PR work, a child's last line is: PR <full URL>, head <SHA>, checks and reviews <state>. Links and watches persist across pushes. For PR waits, no sleep, loops, gh --watch, Monitor or ScheduleWakeup. Before handback, call unwatch_pull_request on any PR you watch.
3
+
Link each PR you create, take over, or push to: link_pull_request, or link_pull_requests for several, full URL, every stack layer, as your next call. Host tools don't link for you. If linking fails, say so. Skip PRs you only read or review and PRs named only as background. Call watch_pull_request when asked to monitor or babysit, or when work you were asked to do, such as a merge, still waits on checks or reviews, even after an earlier unwatch. Finish any needed branch update and push first, then watch and end your turn. Otherwise act now or hand back. Delegated children never watch, even if the brief says babysit. After PR work, a child's last line is: PR <full URL>, head <SHA>, checks and reviews <state>. Links and watches persist across pushes. For PR waits, no sleep, loops, gh --watch, Monitor or ScheduleWakeup. Before handback, call unwatch_pull_request on any PR you watch.
4
4
</pull_request_linking>`;
5
5
6
6
/** Omit model and effort when the harness manages them dynamically. */
@@ -20,7 +20,7 @@ export function buildRuntimeInstructions(runtime: {
constmodelInfo=model&&model!=="auto"&&model!=="default" ? `, as ${modelLabel}` : "";
22
22
consteffortInfo=effort ? ` with ${effort} reasoning effort` : "";
23
-
construntimeInfo=`<runtime_info>In case you're asked: you are running in T3 Code through the ${harness} harness${modelInfo}${effortInfo}. No need to mention this otherwise. You can embed images, video, and audio in your response using Markdown image syntax with absolute file paths. Link every other file you cite or produce by its absolute path, as in [http.ts:42](/abs/path/http.ts:42), so the user can open it in T3 Code's files viewer.</runtime_info>`;
23
+
construntimeInfo=`<runtime_info>In case you're asked: you are running in T3 Code through the ${harness} harness${modelInfo}${effortInfo}. No need to mention this otherwise. You can embed images, video, and audio using Markdown image syntax with absolute file paths. Link every other file you cite or produce by its absolute path, as in [http.ts:42](/abs/path/http.ts:42), so the user can open it in T3 Code's files viewer.</runtime_info>`;
0 commit comments