fix(stall_watch): call herdr with the sanitized agent name (Issue #66)

herdr registers an agent under sanitize_herdr_agent_name(session), so a
session name over 32 characters or with upper-case letters got
agent_not_found and stall detection went off. stall_watch now converts the
name for every herdr call (agent get/read/prompt, watcher.wait_once); logs
and notices keep the original name, and the agent kind is still read off
the original suffix (a long name's sanitized form ends in a hash).

The conftest mock herdr gets an opt-in strict_agent_names mode that, like
the real herdr, answers an agent only under its registered name and fails
get/read with agent_not_found JSON; existing tests see no change.

Refs #66

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

tests/conftest.py

@@ -470,6 +470,9 @@ def _sanitize_agent_name(name):
 def _match_agent(k, target):
     if not k or not target:
         return False
+    # Issue #66: like the real herdr, answer only the registered (sanitized) name.
+    if state.get("strict_agent_names"):
+        return k == target
     return k == target or _sanitize_agent_name(k) == target or _sanitize_agent_name(target) == k or _sanitize_agent_name(k) == _sanitize_agent_name(target)
 
 args = sys.argv[1:]
@@ -1052,6 +1055,10 @@ elif cmd1 == "agent":
             }))
             sys.exit(0)
         else:
+            if state.get("strict_agent_names"):
+                sys.stderr.write(json.dumps({"error": {"code": "agent_not_found",
+                                                       "message": "agent " + name + " not found"}}) + "\\n")
+                sys.exit(1)
             sys.stderr.write("Agent " + name + " not found\\n")
             sys.exit(1)
     elif cmd2 == "wait":
@@ -1143,6 +1150,10 @@ elif cmd1 == "agent":
             print(buffer_content)
             sys.exit(0)
         else:
+            if state.get("strict_agent_names"):
+                sys.stderr.write(json.dumps({"error": {"code": "agent_not_found",
+                                                       "message": "agent " + name + " not found"}}) + "\\n")
+                sys.exit(1)
             sys.stderr.write("Agent " + name + " not found\\n")
             sys.exit(1)
     elif cmd2 == "prompt":

tests/test_issue66_stall_watch_sanitized_name.py

@@ -0,0 +1,129 @@
+"""Issue #66: stall_watch calls herdr with the sanitized agent name.
+
+herdr registers an agent under `sanitize_herdr_agent_name(session)` (lower
+case, at most 32 characters). The watcher gets the original session name from
+the wrapper, so it must convert it for every herdr call while its logs keep
+the original name and the agent kind is still read off the original suffix.
+
+The harness is the one of test_issue30_stall_watch (a real watcher process, a
+real job, the conftest mock herdr), with the mock switched to
+`strict_agent_names`: like the real herdr it answers an agent only under its
+registered (sanitized) name, so the original name gets `agent_not_found`.
+"""
+
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+import test_issue30_stall_watch as t30  # noqa: E402
+
+from lib_py.agents.sanitize import sanitize_herdr_agent_name  # noqa: E402
+
+LONG = "creator-claude-fix-issue-99-long-branch"  # 39 characters
+UPPER = "Creator-Claude-T"
+PLAIN = "creator-claude-t"  # needs no conversion
+LONG_CODEX = "mam-wt-issue-66-very-long-creator-codex"  # kind only in the original suffix
+
+DIALOG = "\n Dangerous rm operation detected. Do you want to proceed?\n 1. Yes\n"
+
+
+@pytest.fixture
+def strict_box(mam_sandbox, mock_herdr):
+    state = json.loads(mock_herdr.read_text())
+    state["strict_agent_names"] = True
+    mock_herdr.write_text(json.dumps(state, indent=2))
+    reg = mam_sandbox / ".mam" / "jobs"
+    return mock_herdr, reg, t30._env(mam_sandbox, MAM_STALL_IDLE_S=2)
+
+
+def _agent_calls(state_file: Path) -> list:
+    return [c for c in json.loads(state_file.read_text())["calls"]
+            if c[:1] == ["agent"] and c[1:2] in (["get"], ["read"], ["prompt"], ["wait"])]
+
+
+def test_the_names_change_and_the_strict_mock_answers_only_the_sanitized_one(strict_box):
+    state, _, env = strict_box
+    assert len(LONG) > 32 and sanitize_herdr_agent_name(LONG) != LONG
+    assert sanitize_herdr_agent_name(UPPER) == PLAIN
+    assert sanitize_herdr_agent_name(PLAIN) == PLAIN
+    for name in (LONG, UPPER):
+        herdr_name = sanitize_herdr_agent_name(name)
+        t30._agent(state, herdr_name, "claude", 4242, "idle", "Ready")
+        for cmd in (["get"], ["read"], ["wait", "--until", "idle"], ["prompt"]):
+            argv = ["herdr", "agent", cmd[0], name] + cmd[1:] + (["hi"] if cmd == ["prompt"] else [])
+            res = subprocess.run(argv, capture_output=True, text=True, env=env)
+            assert res.returncode == 1, (argv, res.stdout)
+            assert json.loads(res.stderr)["error"]["code"] == "agent_not_found", argv
+            argv[3] = herdr_name
+            assert subprocess.run(argv, capture_output=True, text=True, env=env).returncode == 0, argv
+
+
+@pytest.mark.parametrize("target", [LONG, UPPER, PLAIN])
+def test_an_idle_agent_is_nudged_through_the_sanitized_name(strict_box, target):
+    state, reg, env = strict_box
+    herdr_name = sanitize_herdr_agent_name(target)
+    jid = t30._job(env, reg)
+    agent = t30._proc("exec sleep 300")
+    try:
+        t30._agent(state, herdr_name, "claude", agent.pid, "idle", t30._pane("claude"))
+        w = t30._watch(env, reg, jid, target, "claude")
+        assert t30._until(lambda: t30._prompts(state))
+        time.sleep(2)
+        assert t30._details(reg, jid) == ["stalled:idle"]
+        calls = _agent_calls(state)
+        assert {c[1] for c in calls} == {"get", "read", "prompt", "wait"}, calls
+        assert all(c[2] == herdr_name for c in calls), calls
+        assert len(t30._prompts(state)) == 1
+        out = t30._finish(env, reg, jid, w)
+        assert f"agent {target} idle for" in out
+    finally:
+        t30._kill(agent)
+
+
+@pytest.mark.parametrize("target", [LONG, UPPER])
+def test_a_blocked_agent_is_reported_with_the_original_name(strict_box, target):
+    state, reg, env = strict_box
+    herdr_name = sanitize_herdr_agent_name(target)
+    jid = t30._job(env, reg)
+    agent = t30._proc("exec sleep 300")
+    try:
+        t30._agent(state, herdr_name, "claude", agent.pid, "blocked", t30._pane("claude"), detection=DIALOG)
+        w = t30._watch(env, reg, jid, target, "claude")
+        assert t30._until(lambda: t30._details(reg, jid))
+        time.sleep(2)
+        assert t30._details(reg, jid) == ["stalled:blocked Dangerous rm operation detected. Do you want to proceed?"]
+        calls = _agent_calls(state)
+        assert calls and all(c[2] == herdr_name for c in calls), calls
+        assert not t30._prompts(state)
+        out = t30._finish(env, reg, jid, w)
+        assert f"agent {target} is blocked" in out
+    finally:
+        t30._kill(agent)
+
+
+def test_the_kind_of_a_long_name_comes_from_its_original_suffix(strict_box):
+    """No usable --agent: the kind comes from the session name. Read off the
+    sanitized name (a hash at the end) it would be unknown, and the nudge would
+    be withheld as `input_check_unavailable`."""
+    state, reg, env = strict_box
+    herdr_name = sanitize_herdr_agent_name(LONG_CODEX)
+    assert not herdr_name.endswith("-codex")
+    jid = t30._job(env, reg)
+    agent = t30._proc("exec sleep 300")
+    try:
+        t30._agent(state, herdr_name, "codex", agent.pid, "idle", t30._pane("codex"))
+        w = t30._watch(env, reg, jid, LONG_CODEX, "")
+        assert t30._until(lambda: t30._prompts(state))
+        time.sleep(2)
+        assert t30._details(reg, jid) == ["stalled:idle"]
+        assert all(c[2] == herdr_name for c in _agent_calls(state))
+        t30._finish(env, reg, jid, w)
+    finally:
+        t30._kill(agent)