@check-spelling-bot Report

⚠️ Dictionary not found

Problems were encountered retrieving check dictionaries (cspell:cpp/src/stdlib-cpp.txt cspell:monkeyc/src/monkeyc_keywords.txt cspell:ada/dict/ada.txt cspell:typescript/dict/typescript.txt cspell:python/src/additional_words.txt cspell:clojure/src/clojure.txt cspell:gaming-terms/dict/gaming-terms.txt cspell:sql/src/sql.txt cspell:cpp/src/stdlib-cerrno.txt cspell:html/dict/html.txt cspell:cpp/src/compiler-msvc.txt cspell:node/dict/node.txt cspell:software-terms/dict/softwareTerms.txt cspell:npm/dict/npm.txt cspell:java/src/java.txt cspell:powershell/dict/powershell.txt cspell:python/src/python/python.txt cspell:css/dict/css.txt cspell:rust/dict/rust.txt cspell:cpp/src/lang-keywords.txt cspell:haskell/dict/haskell.txt cspell:scala/dict/scala.txt cspell:python/src/python/python-lib.txt cspell:fullstack/dict/fullstack.txt cspell:cpp/src/ecosystem.txt cspell:public-licenses/src/generated/public-licenses.txt cspell:cpp/src/lang-jargon.txt cspell:php/dict/php.txt cspell:python/src/common/extra.txt cspell:cpp/src/people.txt cspell:elixir/dict/elixir.txt cspell:dart/src/dart.txt cspell:k8s/dict/k8s.txt cspell:redis/dict/redis.txt cspell:cpp/src/compiler-clang-attributes.txt cspell:svelte/dict/svelte.txt cspell:latex/dict/latex.txt cspell:shell/dict/shell-all-words.txt cspell:dotnet/dict/dotnet.txt cspell:public-licenses/src/additional-licenses.txt cspell:swift/src/swift.txt cspell:ruby/dict/ruby.txt cspell:cpp/src/template-strings.txt cspell:cpp/src/stdlib-c.txt cspell:lua/dict/lua.txt cspell:cpp/src/compiler-gcc.txt cspell:django/dict/django.txt cspell:docker/src/docker-words.txt cspell:r/src/r.txt cspell:sql/src/tsql.txt cspell:golang/dict/go.txt cspell:software-terms/dict/webServices.txt cspell:java/src/java-terms.txt cspell:cpp/src/stdlib-cmath.txt).

⚠️ For more information, see check-dictionary-not-found.

🔴 Please review

See the 📂 files view, the 📜action log, 👼 SARIF report, or 📝 job summary for details.

Unrecognized words (4)

ollama
remotesigned
segfaults
Subscribe'd

curl -s -S -L 'https://raw.githubusercontent.com/check-spelling/check-spelling/cfb6f7e75bbfc89c71eaa30366d0c166f1bd9c8c/apply.pl' |
perl - 'https://github.com/microsoft/intelligent-terminal/actions/runs/27689675121/attempts/1' &&
git commit -m 'Update check-spelling metadata'

    "cspell:csharp/csharp.txt",
    "cspell:aws/aws.txt",
    "cspell:fonts/fonts.txt",

"check_extra_dictionaries": []

Forbidden patterns 🙅 (2)

In order to address this, you could change the content to not match the forbidden patterns (comments before forbidden patterns may help explain why they're forbidden), add patterns for acceptable instances, or adjust the forbidden patterns themselves.

These forbidden patterns matched content:

Should be preexisting

[Pp]re[- ]existing

Should be reentrancy

[Rr]e[- ]entrancy

@microsoft-github-policy-service agree

I tested this branch and wasn't able to get different agents per tab — switching the agent in one tab switched it everywhere. Here's why the current approach can't work:

1. agent_cmd is dropped before it reaches C++
In tools/wta/src/protocol/acp/client.rs (~line 2739), when the helper handles restart_rx.recv(), it receives the correct req.agent_cmd, but the JSON event it fires to WT is:

{"type": "event", "method": "restart_agent_stack", "params": {}}

The new agent command is silently discarded and never reaches the C++ side.

2. SharedWta is a process-wide singleton (architectural blocker)
Although TerminalPage._agentPane is independent per tab, every tab calls SharedWta::Instance() (SharedWta.cpp:15). There is exactly one wta-master process for the entire Windows Terminal window — there's no way to run a different agent per tab without changing this.

3. wta-master spawns a single agent CLI
wta-master spawns one agent subprocess in run_master_loop (tools/wta/src/master/mod.rs). All helpers from all tabs connect via named pipe to this same master and share the same agent subprocess.

4. Restarting master disrupts all tabs
Even if agent_cmd were forwarded correctly, calling SharedWta::Restart() kills the current wta-master and disconnects every helper. When the new master starts with the new agent, all tabs reconnect to it — so every tab switches to the new agent at once, breaking active sessions in other tabs. The code comment at TerminalPage.cpp:4449 describes exactly this behavior.

To support truly independent agents per tab simultaneously, this would require a significant architectural refactor: SharedWta would need to manage a map of wta-master processes keyed by agent type (one process per agent CLI, e.g. one for claude, one for copilot), or wta-master itself would need to internally multiplex multiple agent subprocesses and route each helper to the right one.

Thanks for this! I dug into the flow and have a few concerns — if I understand the code correctly:

1. In the helper+master path, the agent isn't actually switched. cmd_switch_agent sends RestartRequest{ agent_cmd: Some(new_cmd) }, but the helper's restart_rx handler (client.rs:2597) ignores agent_cmd and emits restart_agent_stack with "params": {}. The C++ handler OnRestartAgentStackRequested(hstring /*eventJson*/) ignores the payload too and respawns master "with the cached spawn args (same agent CLI)" (TerminalPage.cpp:4540). So /switch-agent claude seems to restart back into the same agent — the new command never crosses the bridge.

2. It wipes every tab. Before the restart, the handler clears clear_chat_history() / completed_turns / session_id across all tab_sessions plus the whole session_to_tab map, so other tabs lose history. This contradicts "Other tabs are completely unaffected."

3. The single-agent master is unchanged, so per-tab agents still can't coexist. If I understand correctly, this PR only touches the registry/command layer — wta-master still spawns one agent CLI (spawn_agent_process(&cli.agent, …)) and fans all panes onto it, so there's one agent process per window and any switch is global. The real fix for #295 would need a multi-agent master, which is already in flight in #296 (Per-tab AI agents via a multi-agent wta-master). It'd be worth reconciling this PR with that effort rather than layering a slash command on the single-agent model.

Minor: /switch-agent validates against the full KNOWN_AGENTS, including ollama (delegate-only, empty ACP command), so /switch-agent ollama is accepted but can't be hosted in a pane.

Thanks so much for putting this together, @ashishpatel26 — really appreciate the effort and the clear write-up, and the Ollama registry cleanup along the way. 🙏

After digging into the architecture, we're going to go with #296 (Per-tab AI agents via a multi-agent wta-master) for #295, since it matches the direction we want: it reworks wta-master into a multi-agent broker so each tab's helper declares its own agent (GPO-validated, id-based) and the master lazily spawns/reuses one CLI per agent. That gives true per-tab agents running simultaneously, with switching scoped to a single tab (no shared-master restart, sibling tabs untouched) — the per-tab independence this PR was aiming for, solved at the architecture layer.

Given the two PRs target the same issue and #296 is the route we're taking, I'll close this one to avoid overlap. Thank you again — your investigation comments here were spot on and genuinely helpful in confirming the approach. Hope to see more contributions from you! 🙌