Watch for connector requests with a one-shot background watcher, not Monitor

internal/commands/agent_hook_connector.go

@@ -58,8 +58,9 @@ func connectorReloadNotice(input agentHookInput) string {
 		"finished cut that skill to its first 5,000 tokens, which drops the handoff steps and the subagent's " +
 		"instructions. If this session is still the connector's orchestrator, read " + skill + " in full with " +
 		"the Read tool now, before you go on with anything, a request in hand included, then carry on from the " +
-		"step you were on. The connector, its monitor and any subagents in flight kept running through the " +
-		"compaction: don't start them again."
+		"step you were on. The connector and any subagents in flight kept running through the compaction: don't " +
+		"start them again. The request watcher exits on each request, so check whether it is still running; if it " +
+		"has exited, read its output and start it again as the skill says."
 }
 
 // lastConnectorSkillDir returns the directory of the last basecamp-connect

skills/basecamp-connect/SKILL.md

@@ -75,25 +75,22 @@ see `connector: running` in its log. If it exited instead, explain the error
 (see *When the connector stops*), and stop. If it says `could not get the
 agent's token yet`, Basecamp's token endpoint is rate-limiting it or having
 trouble: it waits as long as it says and tries again on its own. Tell the
-person, leave it running, and arm the monitor as usual. Restarting it doesn't
+person, leave it running, and start the watcher as usual. Restarting it doesn't
 help.
 
-Then arm a persistent monitor on that file with the Monitor tool
-(`persistent: true`), replaying from the start so nothing that arrived between
-the start and the monitor is missed:
+Then start the watcher, also with `run_in_background: true`:
 
 ```bash
-tail -f -n +1 '<connector-output-file>' | grep --line-buffered -F '"type":"request"'
+bash '${CLAUDE_SKILL_DIR}/watch-requests.sh' '<connector-output-file>' ~/.config/basecamp-connect/'handled-<profile>'
 ```
 
-Single-quote the path, as with every other value.
-
-Each notification is one request: handle it with *For each request*. A request
-that arrives while you're waiting on the person is not their reply.
-
-The connector starts from the present. Anything asked more than a minute
-before it started is not picked up. If someone says the agent ignored them, that's the first
-thing to check.
+It exits once request lines arrive whose `event_id` the handled file lacks.
+Read them in full from its output file, record each
+(`echo <event_id> >> ~/.config/basecamp-connect/'handled-<profile>'`), start
+the watcher again, then handle each with *For each request*. Keep one running.
+Not the Monitor tool: its watches expire, and a `grep -v` in one holds a line
+back until the next. A request that arrives while you're waiting on the person
+is not their reply.
 
 ### 2. The request line
 
@@ -549,12 +546,24 @@ output and tell the person what it means:
 
 Requests that arrive while it's stopped are not picked up later.
 
+Each start writes a new output file. Stop the watcher (TaskStop), then check
+the old file once, so a request printed just before the exit isn't left there:
+
+```bash
+bash '${CLAUDE_SKILL_DIR}/watch-requests.sh' --once '<old-output-file>' ~/.config/basecamp-connect/'handled-<profile>'
+```
+
+Handle any line it prints as usual. Start the watcher on the new file once the
+connector is running.
+
 ## Checking on it
 
 - `basecamp connect doctor -P '<profile>'`: the credential, the identity, the
   event feed and the ledger. Start here when something seems wrong. It posts
   nothing.
-- **"Did that mention reach it?"**: search the connector's output file for the
+- **"Did that mention reach it?"**: the connector starts from the present, so
+  anything asked more than a minute before it started is not picked up. Check
+  that first. Then search the connector's output file for the
   recording id. A `"type":"request"` line means it was handed to you. A
   `"type":"event"` line with a `reason` says why it was turned away (for
   example `untrusted_performer`). No line means the feed never delivered it.
@@ -564,7 +573,8 @@ Requests that arrive while it's stopped are not picked up later.
 
 ## Stopping
 
-When the person asks you to stop, or the session is ending, stop the connector
-task (TaskStop, or send it SIGINT). There's nothing else to clean up: no
-webhooks, nothing public. Subagents already running finish their requests and
-reply. Say that new mentions won't be picked up until it's started again.
+When the person asks you to stop, or the session is ending, stop the watcher
+and the connector task (TaskStop, or send it SIGINT). There's nothing else to
+clean up: no webhooks, nothing public. Subagents already running finish their
+requests and reply. Say that new mentions won't be picked up until it's
+started again.

skills/basecamp-connect/watch-requests.sh

@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+set -u
+
+once=false
+if [ "${1:-}" = --once ]; then
+  once=true
+  shift
+fi
+if [ $# -ne 2 ]; then
+  echo "usage: watch-requests.sh [--once] <connector-output-file> <handled-file>" >&2
+  exit 2
+fi
+
+out=$1
+handled=$2
+
+if [ ! -f "$out" ]; then
+  echo "watch-requests: no connector output at $out" >&2
+  exit 2
+fi
+if ! { mkdir -p "$(dirname "$handled")" && touch "$handled" && [ -f "$handled" ] && [ -r "$handled" ] && [ -w "$handled" ]; }; then
+  echo "watch-requests: can't use the handled file $handled" >&2
+  exit 2
+fi
+
+unhandled() {
+  local lines
+  if [ ! -r "$out" ]; then
+    echo "watch-requests: can't read the connector output at $out" >&2
+    exit 2
+  fi
+  lines=$(($(wc -l < "$out")))
+  [ "$lines" -gt 0 ] || return 1
+  head -n "$lines" "$out" | HANDLED=$handled awk '
+    BEGIN { while ((getline id < ENVIRON["HANDLED"]) > 0) seen[id] = 1 }
+    /"type":"request"/ && match($0, /"event_id":[0-9]+/) {
+      id = substr($0, RSTART + 11, RLENGTH - 11)
+      if (!(id in seen)) { print; seen[id] = 1; found = 1 }
+    }
+    END { exit !found }'
+}
+
+if [ "$once" = true ]; then
+  unhandled
+  exit 0
+fi
+
+until unhandled; do
+  sleep 2
+done