git:20261012.bbafe9d to git:20261018.eaed515

20 added, 24 removed. Audit A to A.

---

name: reset-cursor-acp

- description: 'Reset a stuck Cursor ACP thread in a thread manager and reload its configuration. Use only when the user explicitly invokes /reset-cursor-acp.'

+ description: 'Reset a stuck Cursor ACP thread in your chat app and reload its configuration. Use only when the user explicitly invokes /reset-cursor-acp.'

disable-model-invocation: true

---

- # Reset Cursor ACP in a thread manager

+ # Reset Cursor ACP in your chat app

- For terminal sessions use `cursor-cli`; for general thread-manager control use `<thread-manager-cli>`.

+ For terminal sessions use `cursor-cli`; for general app control use `<app-cli>`.

## How resets work

- - An ACP integration may start one `cursor-agent acp` subprocess **per thread, on demand**, owned by a worker.

- - **There is no global Cursor ACP server.** Do not create one or keep it alive with a watchdog.

+ - The app's ACP integration starts one `cursor-agent acp` subprocess per thread, on demand.

+ - There is no global Cursor ACP server. Do not create one or keep it alive with a watchdog.

- Rules, skills, and `.cursor/mcp.json` reload when a new subprocess starts.

- - `<thread-manager-cli> thread stop <id>` releases the runtime and stops its subprocess, preserving history. The next message starts a fresh agent.

+ - Use the app's thread-stop command with the thread ID to release the runtime and stop its subprocess while preserving history. The next message starts a fresh agent.

## Quick reset

- Resolve `scripts/<reset-script>.sh` relative to this `SKILL.md`. Run:

+ Resolve `<reset-script>` relative to this `SKILL.md`. Run:

```bash

- scripts/<reset-script>.sh <thread-id> # reset one Cursor thread

- scripts/<reset-script>.sh --self # reset the current thread (<CURRENT_THREAD_ID_ENV>)

- scripts/<reset-script>.sh # no thread: orphan cleanup + health check only

- scripts/<reset-script>.sh --dry-run # show what would happen

- scripts/<reset-script>.sh <id> --kill-all # also kill live agents of OTHER Cursor threads

+ <reset-script> <thread-id> # reset one Cursor thread

+ <reset-script> --self # reset the current thread

+ <reset-script> # no thread: orphan cleanup + health check only

+ <reset-script> --dry-run # show what would happen

+ <reset-script> <id> --kill-all # also kill live agents of OTHER Cursor threads

```

- Find Cursor thread ids with `<thread-manager-cli> status` (current thread) or:

- ```bash

- <thread-manager-cli> thread list --json | python3 -c 'import json,sys; [print(t["id"], t["status"], t["title"]) for t in json.load(sys.stdin) if t.get("providerId")=="acp-cursor"]'

- ```

+ Find Cursor thread IDs using the app's status or thread-list command.

- The script stops the selected thread, removes orphaned `cursor-agent acp` processes (parent gone or not a bridge worker), reports CLI version/login/update status, and asks for the next message. Other threads stay running unless `--kill-all`.

+ The script stops the selected thread, removes orphaned `cursor-agent acp` processes, reports CLI version/login/update status, and asks for the next message. Other threads stay running unless `--kill-all`.

If sandboxing blocks `ps` with "operation not permitted", rerun the script outside the sandbox.

## Verify

Send one short message to verify a fresh agent responds. If it hangs again, check the causes below before another reset.

## Known causes

- - **Old Cursor CLI.** If the health check shows an update, run `<thread-manager-cli> updates apply`, then reset.

+ - **Old Cursor CLI.** If the health check shows an update, use the app's update command, then reset.

- **Expired login.** Symptom "Failed to initialize session services". Fix: `cursor-agent login`, then reset.

- **Pending permission.** Cursor waits for `session/request_permission`. Check and resolve the thread's pending approval before resetting.

- - **Resume failure.** For `session/load` → "Session not found", use a fresh session after `<thread-manager-cli> thread stop` rather than retrying the load.

+ - **Resume failure.** For `session/load` → "Session not found", use a fresh session after stopping the thread rather than retrying the load.

- **Team-level MCP servers** from the Cursor dashboard do not work in ACP mode. Only project or user `.cursor/mcp.json`.

- - **Rate limits.** Enable retries with `<plugin-manager> enable <retry-plugin>`.

+ - **Rate limits.** Enable retries using the app's plugin controls.

## Do not

- No launchd KeepAlive, cron, or watchdog for `cursor-agent acp`.

- - Never blindly run `pkill -f cursor-agent`; it kills all Cursor threads and the interactive TUI. Use the script, which targets orphans by default.

- - Do not restart the thread manager for one stuck thread; use `<thread-manager-cli> thread stop`.

- - Do not use `<thread-manager-cli> thread compact`; Cursor does not support it.

+ - Never blindly run `pkill -f cursor-agent`; it kills all Cursor threads and the interactive TUI. Use the reset script, which targets orphans by default.

+ - Do not restart the app for one stuck thread; stop only the affected thread.

+ - Do not use the app's thread-compaction feature; Cursor does not support it.