Add headless "ask" mode: ?-prefixed message runs claude -p, answer posted back

A message starting with `?` in a mapped room runs `claude -p` one-shot in that
repo on the Mac and posts the full answer back into the room — Matrix as a
request/response interface, not just a trigger. Non-`?` messages keep launching
interactive sessions as before.

New scripts/ask-claude.sh is a login-shell wrapper (so ~/.zprofile puts claude on
PATH) that exports CLAUDE_CODE_OAUTH_TOKEN from the Mac's .env and runs
`claude -p "$prompt" < /dev/null`, printing the answer to stdout. The bot adds a
`?`-dispatch with run_ask/ask: SSH stdout captured, 300s timeout, fail-loud, output
chunked under Matrix's event cap (no truncation).

Headless claude -p needs the long-lived token because a non-GUI SSH session can't
reach the login Keychain (reports "Not logged in") — the deliberate Approach A that
the interactive GUI-Terminal path (D11) avoided. Token is kept Mac-side only; the
Spark never runs claude. Sovereignty unchanged: claude -p uses the subscription, no
frontier API touches message payloads.

Proven live on the Spark; fresh-eyes reviewed before commit.

.env.example

@@ -8,3 +8,9 @@ MATRIX_ACCESS_TOKEN=
 # Optional — kept for recovery / re-minting a token. The bot authenticates with the access token,
 # not the password (logging in every start would spawn a new device each time).
 MATRIX_PASSWORD=
+
+# Headless "ask" mode (the `?`-prefix path). Used MAC-SIDE by scripts/ask-claude.sh, NOT by the
+# bot — a non-GUI SSH session can't reach the login Keychain, so `claude -p` needs this token to
+# authenticate. Mint once on the Mac: `claude setup-token` (requires a Claude subscription), then
+# paste the value here. Lives on the Mac; the Spark never runs claude, so it needs no copy.
+CLAUDE_CODE_OAUTH_TOKEN=

AGENTS.md

@@ -74,8 +74,12 @@ v1 decision surface.
 - `config.example.toml` — room→repo mapping template; the real `config.toml` is gitignored.
 - `scripts/gui-launch.sh` — opens the desktop Terminal via `osascript` (Approach B, D11); calls
   `launch-claude.sh` inside it. The bot invokes this over SSH.
-- `src/bot.py` — the matrix-nio bot (Phase 1): listens in mapped rooms; on a message runs
-  `ssh mac-bridge gui-launch.sh`; fans out for all-projects; reports failures back to the room.
+- `scripts/ask-claude.sh` — headless `?`-ask wrapper (`#!/bin/zsh -l`): runs `claude -p` in the repo
+  and prints the answer to stdout for the bot to capture and post back. Uses `CLAUDE_CODE_OAUTH_TOKEN`
+  (Mac-side `.env`) because a non-GUI SSH session can't reach the login Keychain (D12).
+- `src/bot.py` — the matrix-nio bot (Phase 1): listens in mapped rooms; a plain message runs
+  `ssh mac-bridge gui-launch.sh` (interactive, to the phone), a `?`-prefixed message runs
+  `ask-claude.sh` (headless, answer posted back); fans out for all-projects; reports failures back.
 - `requirements.txt` (matrix-nio) · `.env.example` (credential schema; real `.env` gitignored).
 - `.claude/` — Claude wiring (dir only for now).
 - `Dockerfile` · `docker-compose.yml` · `docker-entrypoint.sh` · `.dockerignore` — the Phase 1
@@ -125,6 +129,13 @@ Condensed from the scoping workshop. Each: the call, why, what it beat.
   and is fully unattended, but adds a credential to manage; kept as the documented fallback if the
   Mac is ever driven headless (logged out). *Cost:* requires the Mac logged in + a one-time
   Terminal Automation grant.
+- **D12 — Headless "ask" mode uses the long-lived token; interactive stays GUI-Terminal (2026-06-16).**
+  A `?`-prefixed message runs `claude -p` headlessly over plain SSH and posts the answer back, so its
+  stdout must be captured over the SSH pipe — which rules out the GUI-Terminal path (D11), and a
+  non-GUI session reports "Not logged in." Ask mode therefore deliberately adopts the long-lived
+  `claude setup-token` (`CLAUDE_CODE_OAUTH_TOKEN`) that D11 deferred — kept **Mac-side only** (in
+  `.env`; the Spark never runs claude). Interactive launches keep the token-free GUI-Terminal path.
+  *Sovereignty unchanged:* `claude -p` uses the subscription, no frontier API touches message payloads.
 
 ## Sovereignty constraint
 
@@ -223,9 +234,21 @@ once" is not done.
   (`modelo@10.59.211.6` — reachable over WireGuard but not authenticated; Phase 0 only set up the
   reverse, `mac-bridge`). So deploys/restarts on the Spark are run by the owner from the Spark, not
   driven from the Mac — until Phase 3 wires it behind Spark Control.
+- **Headless "ask" mode — SHIPPED + proven on the Spark (2026-06-16).** A `?`-prefixed message in a
+  mapped room runs `claude -p` one-shot in that repo on the Mac and posts the **full** answer back
+  into the room (Matrix as request/response, not just a trigger); non-`?` messages launch
+  interactively as before. New `scripts/ask-claude.sh` (login-shell wrapper: extracts
+  `CLAUDE_CODE_OAUTH_TOKEN` from the Mac's `.env`, runs `claude -p "$prompt" < /dev/null`); `bot.py`
+  gained the `?`-dispatch + `run_ask`/`ask` (SSH stdout captured, 300s timeout, fail-loud, output
+  chunked under Matrix's ~64KB cap). *Why a token (D12):* a non-GUI SSH session can't reach the login
+  Keychain, so headless `claude -p` reports "Not logged in" — Approach A, kept Mac-side only (the
+  Spark never runs claude). Fresh-eyes reviewed before commit; P1 nits fixed (reap killed ssh on
+  timeout; treat rc=0 + empty output as success, not failure). *Proven:* a real `?`-ask in an
+  already-trusted repo returned the answer into the room. *Open edge:* a `?`-ask in a repo `claude`
+  has **never** been opened in may stall on the first-run folder-trust gate (Phase 0 caveat) — add a
+  trust flag to the wrapper if/when hit, not preemptively.
 - **Next (open — discuss before building):** Phase 2 (multi-room routing) is effectively already
-  satisfied — the bot was built multi-room (11 rooms + all-projects) and routed correctly across 2
-  rooms in the Phase 1 proof; only a formal confirmation pass remains. Live candidates: **Phase 3**
-  (Spark Control: bot status + one-click update/restart on the dashboard, the SSH-behind-buttons
-  pattern — also closes the owner-run-ops gap above) or the **headless "ask" mode** from
-  `ROADMAP.md` (a message runs `claude -p` and posts the answer back into the room).
+  satisfied (built multi-room; routed correctly across rooms in the Phase 1 proof) — only a formal
+  confirmation pass remains. Main remaining candidate: **Phase 3** (Spark Control: bot status +
+  one-click update/restart on the dashboard, the SSH-behind-buttons pattern — also closes the
+  owner-run-ops gap above). Other backlog in `ROADMAP.md`.

ROADMAP.md

@@ -54,13 +54,13 @@ after it.
   is actually in use.
 - **E2EE (D9).** Add matrix-nio end-to-end encryption (libolm) if the bot ever handles
   sensitive content over untrusted transport. Low priority while everything is WireGuard-local.
-- **Headless "ask" mode — return output into the chat (no interactive session).** Today a message
-  opens an interactive session surfaced to the phone. Add a mode where a message instead runs
-  `claude -p "<prompt>"` headlessly in the repo (full Claude Code context, but one-shot), captures
-  stdout, and posts the result back into the Matrix room — Matrix as a request/response interface,
-  not just a trigger. *Design notes:* `claude -p` (print mode) is exactly this capability. Likely
-  uses the long-lived OAuth token (Approach A / D11) so it runs over plain SSH with no GUI Terminal
-  and stdout is captured directly. *Open Qs:* how to select interactive-vs-ask (per-room? a prefix
-  like `?` / `/ask`? a dedicated room?); output-length handling (truncate / thread / attach file);
-  same local-only sovereignty constraints apply (output is the user's own; `claude -p` uses the
-  subscription, no frontier API on message payloads).
+- **Headless "ask" mode — SHIPPED 2026-06-16.** A `?`-prefixed message runs `claude -p "<rest>"`
+  one-shot in the room's repo and posts the **full** answer back into the room — Matrix as a
+  request/response interface, not just a trigger. Built via `scripts/ask-claude.sh` (login-shell
+  wrapper) + the bot's `?`-dispatch (`run_ask`/`ask`). Resolved design choices: selector = `?` prefix
+  (per-message; the room still picks the repo); output posted in full, chunked under Matrix's event
+  cap (no truncation — chosen explicitly); auth = the long-lived `claude setup-token`
+  (`CLAUDE_CODE_OAUTH_TOKEN`, Approach A / D12) because a non-GUI SSH session can't reach the
+  Keychain; sovereignty unchanged (`claude -p` uses the subscription, no frontier API on payloads).
+  *Remaining open Qs:* very-long-output handling beyond chunking (thread / attach file); the
+  first-run folder-trust gate for a repo `claude` has never been opened in.

config.example.toml

@@ -15,6 +15,7 @@ user = "@matrix-bridge-bot:<your-domain>"  # a dedicated bot Matrix account (not
 [mac]
 ssh_alias = "mac-bridge"
 launcher = "/Users/macpro/Projects/<your-repo>/scripts/gui-launch.sh"
+ask_launcher = "/Users/macpro/Projects/<your-repo>/scripts/ask-claude.sh"   # headless `?`-prefix ask mode
 # Container only: docker-entrypoint.sh generates ~/.ssh/config for `ssh_alias` from these.
 # (On a host with `ssh_alias` already in ~/.ssh/config these are ignored.)
 hostname = "10.0.0.0"          # the Mac's address reachable from the Spark (e.g. WireGuard IP)

scripts/ask-claude.sh

@@ -0,0 +1,45 @@
+#!/bin/zsh -l
+# ask-claude.sh — matrix-bridge headless "ask" wrapper.
+#
+# Invoked over SSH by the bot:  ask-claude.sh <repo_dir> <prompt...>
+# Runs `claude -p` one-shot in the repo and prints the answer to STDOUT, which the bot
+# captures over the SSH pipe and posts back into the Matrix room. Unlike launch-claude.sh /
+# gui-launch.sh (interactive, surfaced to the phone), this NEVER opens a GUI Terminal.
+#
+# Two seams it owns, both proven the hard way in Phase 0:
+#  - LOGIN shell (-l): a non-login SSH shell loads neither ~/.zprofile nor ~/.zshrc, so
+#    ~/.local/bin isn't on PATH and `claude` isn't found. Same reason as launch-claude.sh.
+#  - Headless auth via CLAUDE_CODE_OAUTH_TOKEN (from `claude setup-token`, stored in ../.env):
+#    a non-GUI SSH session can't reach the login Keychain, so plain `claude -p` reports
+#    "Not logged in" (D11 / Approach A). We export the token to bypass the Keychain.
+
+set -e
+
+script_dir="${0:A:h}"
+
+# Pull just the token out of ../.env (don't `source` the whole file — other values, e.g. a
+# password, may not be shell-safe). Absent token => claude reports "Not logged in", reported
+# back to the room by the bot.
+env_file="$script_dir/../.env"
+if [[ -f "$env_file" ]]; then
+  token_line="$(grep -E '^CLAUDE_CODE_OAUTH_TOKEN=' "$env_file" | head -1)"
+  token="${token_line#*=}"
+  token="${token#\"}"           # strip one surrounding quote pair if present (KEY="value")
+  token="${token%\"}"
+  export CLAUDE_CODE_OAUTH_TOKEN="$token"
+fi
+
+repo_dir="$1"
+shift
+prompt="$*"
+
+if [[ -z "$repo_dir" || -z "$prompt" ]]; then
+  print -u2 "usage: ask-claude.sh <repo_dir> <prompt>"
+  exit 2
+fi
+
+# Fail loud on a bad directory — never run Claude in the wrong place.
+cd "$repo_dir" || { print -u2 "ask-claude: no such repo dir: $repo_dir"; exit 1; }
+
+# < /dev/null: print mode reads stdin by default and otherwise stalls ~3s waiting for it.
+exec claude -p "$prompt" < /dev/null

src/bot.py

@@ -22,6 +22,10 @@ from nio import AsyncClient, MatrixRoom, RoomMessageText
 
 REPO_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 
+# Headless "ask" mode tunables.
+ASK_TIMEOUT = 300        # seconds to wait for `claude -p` before giving up
+MAX_MSG_CHARS = 30000    # split answers into chunks well under Matrix's ~64KB event cap
+
 
 def load_env(path):
     env = {}
@@ -39,6 +43,27 @@ def load_config(path):
         return tomllib.load(f)
 
 
+def split_message(text, limit=MAX_MSG_CHARS):
+    """Split text into <=limit-char chunks on newline boundaries (no truncation)."""
+    if len(text) <= limit:
+        return [text]
+    chunks, buf = [], ""
+    for line in text.splitlines(keepends=True):
+        while len(line) > limit:            # one oversized line: hard-split it
+            if buf:
+                chunks.append(buf)
+                buf = ""
+            chunks.append(line[:limit])
+            line = line[limit:]
+        if len(buf) + len(line) > limit:
+            chunks.append(buf)
+            buf = ""
+        buf += line
+    if buf:
+        chunks.append(buf)
+    return chunks
+
+
 async def main():
     env = load_env(os.path.join(REPO_ROOT, ".env"))
     cfg = load_config(os.path.join(REPO_ROOT, "config.toml"))
@@ -52,6 +77,7 @@ async def main():
     all_projects_room = cfg.get("all_projects", {}).get("room_id")
     ssh_alias = os.environ.get("MB_SSH_ALIAS") or cfg["mac"]["ssh_alias"]
     launcher = cfg["mac"]["launcher"]
+    ask_launcher = cfg["mac"].get("ask_launcher")
 
     client = AsyncClient(homeserver, user_id)
     client.restore_login(user_id=user_id, device_id=device_id, access_token=token)
@@ -73,6 +99,28 @@ async def main():
         out, _ = await proc.communicate()
         return proc.returncode, out.decode(errors="replace").strip()
 
+    async def run_ask(repo_dir, prompt):
+        """Run ask-claude.sh on the Mac over SSH; return (rc, stdout, stderr).
+
+        Headless `claude -p`: its stdout is the answer (captured here), stderr is diagnostics.
+        This path never opens a GUI Terminal and is not surfaced to the phone.
+        """
+        remote = f"{shlex.quote(ask_launcher)} {shlex.quote(repo_dir)} {shlex.quote(prompt)}"
+        proc = await asyncio.create_subprocess_exec(
+            "ssh", ssh_alias, remote,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+        )
+        try:
+            out, err = await asyncio.wait_for(proc.communicate(), timeout=ASK_TIMEOUT)
+        except asyncio.TimeoutError:
+            proc.kill()
+            await proc.wait()           # reap the killed ssh client (no zombie)
+            return None, "", f"timed out after {ASK_TIMEOUT}s"
+        return (proc.returncode,
+                out.decode(errors="replace").strip(),
+                err.decode(errors="replace").strip())
+
     async def say(room_id, text):
         await client.room_send(
             room_id, "m.room.message", {"msgtype": "m.text", "body": text}
@@ -88,6 +136,24 @@ async def main():
                                f"(rc={rc}): {out[:300] or 'no output'}")
         return False
 
+    async def ask(report_room, repo, prompt):
+        """Headless ask: run `claude -p` in the repo and post the full answer back."""
+        if not ask_launcher:
+            await say(report_room,
+                      "⚠️ matrix-bridge: ask mode not configured ([mac].ask_launcher missing).")
+            return
+        await say(report_room, f"🤔 asking claude in {repo['label']}…")
+        rc, out, err = await run_ask(repo["repo_dir"], prompt)
+        if rc == 0:                     # success — even an empty answer is not a failure
+            print(f"ask {repo['label']}: {len(out)} chars", flush=True)
+            for chunk in split_message(out or "(claude returned no output)"):
+                await say(report_room, chunk)
+            return
+        detail = err or out or "no output"
+        print(f"ASK FAILED {repo['label']}: rc={rc} {detail[:300]}", flush=True)
+        await say(report_room, f"⚠️ matrix-bridge: ask failed in {repo['label']} "
+                               f"(rc={rc}): {detail[:500]}")
+
     async def on_message(room: MatrixRoom, event: RoomMessageText):
         if event.sender == user_id:
             return  # never react to our own messages
@@ -95,7 +161,7 @@ async def main():
         if not prompt:
             return
 
-        if room.room_id == all_projects_room:
+        if room.room_id == all_projects_room:   # fan-out room always launches, never asks
             date = datetime.date.today().isoformat()
             print(f"[all-projects] fan-out to {len(rooms)} repos: {prompt!r}", flush=True)
             results = await asyncio.gather(*[
@@ -106,7 +172,11 @@ async def main():
                       f"matrix-bridge: launched {sum(results)}/{len(rooms)} sessions ({date}).")
         elif room.room_id in rooms:
             r = rooms[room.room_id]
-            if await launch_one(room.room_id, r, prompt):
+            if prompt.startswith("?"):                      # headless ask mode
+                ask_prompt = prompt[1:].strip()
+                if ask_prompt:
+                    await ask(room.room_id, r, ask_prompt)
+            elif await launch_one(room.room_id, r, prompt):
                 await say(room.room_id,
                           f"matrix-bridge: launched {r['label']} — drive it on your phone.")