fix(resume): allow fresh spawn fallback and refresh epoch on 0-turn Class A resume

- Branch on Class A agents (agy, hermes, opencode) in resume_session.sh to fall back to clean spawn (CMD_FULL) when stopped before first turn (empty UUID)
- Update update_yaml_resumed.sh to support empty UUID with --cmd-full and refresh herdr_session_epoch on fresh spawn resume to prevent stale transcript capture
- Strictly preserve Class B (claude, grok) assigned UUID escape hatch and test_t8 contract
- Add regression tests test_t14, test_t15, test_t16 (Closes #3 Item 2)
This commit is contained in:
2026-08-31 10:30:44 +09:00
parent c49ee3bdd8
commit 7e15081d87
3 changed files with 204 additions and 23 deletions
@@ -51,9 +51,23 @@ esac
UUID=$(bash "$(dirname "${BASH_SOURCE[0]}")/resolve_session_id.sh" \
--workspace "$WORKSPACE" --agent "$AGENT" --session "$SESSION_NAME")
# ISSUE-3: agy/hermes/opencode (Class A) are spawned without a UUID - their
# own conversation id is only assigned by the agent itself on its first turn.
# A session stopped before any message therefore has no UUID to resolve here,
# but IS still cleanly recoverable: relaunching the agent's plain first-run
# spawn command restores the exact state multi-agent-mux-create would have
# produced. claude/grok (Class B) always resolve a UUID via the
# verify_session.py:99-101 assigned-but-unverified escape hatch (see test_t8)
# and never reach this branch - do not extend the fallback to them.
FRESH_SPAWN=0
if [ -z "$UUID" ]; then
case "$AGENT" in
agy|hermes|opencode) FRESH_SPAWN=1 ;;
*)
echo "ERROR: No saved session for $WORKSPACE ($AGENT). Use multi-agent-mux-create first." >&2
exit 1
;;
esac
fi
if [ -n "$HERDR_SERVER_OPT" ]; then
@@ -96,8 +110,20 @@ if [ "$(uname)" = "Darwin" ] && [ -f "$RESOLVED_BIN" ]; then
fi
# Determine CMD_FULL via adapter
CMD_FULL="$("$(_delegate_py_bin)" -m lib_py.agents resume-spec "$AGENT" "$RESOLVED_BIN" "$UUID" "$WORKSPACE" 2>/dev/null || true)"
if [ -z "$CMD_FULL" ]; then
if [ "$FRESH_SPAWN" = "1" ]; then
# ISSUE-3: no UUID to resume with - spawn the same plain first-run command
# multi-agent-mux-create would use (spawn-spec, not resume-spec).
CMD_FULL="$("$(_delegate_py_bin)" -m lib_py.agents spawn-spec "$AGENT" "$RESOLVED_BIN" "" 0 2>/dev/null || true)"
if [ -z "$CMD_FULL" ]; then
case "$AGENT" in
agy) CMD_FULL="${RESOLVED_BIN} --dangerously-skip-permissions" ;;
hermes) CMD_FULL="${RESOLVED_BIN} --yolo --accept-hooks" ;;
opencode) CMD_FULL="${RESOLVED_BIN} --auto --agent build" ;;
esac
fi
else
CMD_FULL="$("$(_delegate_py_bin)" -m lib_py.agents resume-spec "$AGENT" "$RESOLVED_BIN" "$UUID" "$WORKSPACE" 2>/dev/null || true)"
if [ -z "$CMD_FULL" ]; then
case "$AGENT" in
claude) CMD_FULL="${RESOLVED_BIN} --dangerously-skip-permissions -r $UUID" ;;
agy) CMD_FULL="${RESOLVED_BIN} --dangerously-skip-permissions --conversation $UUID" ;;
@@ -106,6 +132,7 @@ if [ -z "$CMD_FULL" ]; then
opencode) CMD_FULL="${RESOLVED_BIN} --session $UUID --auto --agent build" ;;
*) echo "ERROR: unsupported agent: $AGENT" >&2; exit 2 ;;
esac
fi
fi
# Validate binary exists and is executable
@@ -145,7 +172,7 @@ sleep 2
# 5. Update agent-sessions.yaml: status running, last_visible_status
bash "$(dirname "${BASH_SOURCE[0]}")/update_yaml_resumed.sh" \
--session "$SESSION_NAME" --uuid "$UUID" --agent "$AGENT" --workspace "$WORKSPACE" \
--herdr-session "$HERDR_SESSION_NAME" \
--herdr-session "$HERDR_SESSION_NAME" --cmd-full "$CMD_FULL" \
${HERDR_WORKSPACE_OPT:+--herdr-workspace "$HERDR_WORKSPACE_OPT"}
echo "Successfully resumed $SESSION_NAME ($AGENT)"
@@ -11,7 +11,11 @@ source "$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)/lib.sh"
usage() {
cat <<EOF
Usage: $0 --session <name> --uuid <id> [--agent claude|agy|hermes|grok|opencode] [--herdr-session <name>]
Usage: $0 --session <name> --uuid <id> [--agent claude|agy|hermes|grok|opencode] [--herdr-session <name>] [--cmd-full <cmd>]
--uuid may be empty for an Issue-3 Class A fresh-spawn resume (agy/hermes/
opencode 0-turn stopped session, no conversation id ever assigned) - pass
--cmd-full so the display command can be recorded without a uuid to embed.
EOF
}
@@ -22,6 +26,7 @@ WORKSPACE=""
ROLE=""
HERDR_SERVER_OPT=""
HERDR_WORKSPACE_OPT=""
CMD_FULL_OPT=""
while [ $# -gt 0 ]; do
case "$1" in
@@ -32,13 +37,13 @@ while [ $# -gt 0 ]; do
--role) ROLE="$2"; shift 2 ;;
--herdr-session|--herdr-server) HERDR_SERVER_OPT="$2"; shift 2 ;;
--herdr-workspace) HERDR_WORKSPACE_OPT="$2"; shift 2 ;;
--cmd-full) CMD_FULL_OPT="$2"; shift 2 ;;
-h|--help) usage; exit 0 ;;
*) echo "ERROR: unknown arg: $1" >&2; exit 2 ;;
esac
done
[ -n "$SESSION_NAME" ] || { echo "ERROR: --session required" >&2; exit 2; }
[ -n "$UUID" ] || { echo "ERROR: --uuid required" >&2; exit 2; }
[ -f "$AGENT_SESSIONS_YAML" ] || { echo "ERROR: $AGENT_SESSIONS_YAML not found" >&2; exit 1; }
if [ -n "$HERDR_SERVER_OPT" ]; then
@@ -104,7 +109,8 @@ atomic_dump_yaml "$AGENT_SESSIONS_YAML" \
SESSION_NAME="$SESSION_NAME" UUID="$UUID" AGENT="$AGENT" NOW_ISO="$NOW_ISO" \
NOW_EPOCH="$NOW_EPOCH" TARGET_WORKSPACE="${WORKSPACE:-$WORKSPACE_ROOT}" ROLE="$ROLE" \
PANE_PID="$PANE_PID" CHILD_PID="$CHILD_PID" HERDR_SERVER_OPT_EXPLICIT="${HERDR_SERVER_OPT_EXPLICIT:-0}" \
MAM_WS_LABEL="$MAM_WS_LABEL" MAM_WS_LABEL_EXPLICIT="${MAM_WS_LABEL_EXPLICIT:-0}" <<'PYEOF'
MAM_WS_LABEL="$MAM_WS_LABEL" MAM_WS_LABEL_EXPLICIT="${MAM_WS_LABEL_EXPLICIT:-0}" \
CMD_FULL_OPT="$CMD_FULL_OPT" <<'PYEOF'
name = os.environ['SESSION_NAME']
uuid = os.environ['UUID']
agent = os.environ['AGENT']
@@ -157,6 +163,18 @@ else:
target['herdr_workspace'] = wsl
target['status'] = 'running'
if not uuid:
# ISSUE-3 (challenge c4b0a075): a Class A fresh-spawn resume re-arms
# discovery from scratch - the resumed agent has not spoken yet, so any
# on-disk conversation older than THIS resume must not be auto-pinned to
# it by reconcile.sh's mode="discover" mtime >= epoch check. Without this,
# the row's watermark stays at the original create_session.sh timestamp,
# letting reconcile.sh capture a stale/unrelated transcript before the
# resumed agent's first real turn.
target['herdr_session_epoch'] = epoch
target['herdr_session_created_at'] = now
target['session_id_source'] = 'pending-discovery'
target['session_id_verified'] = False
target.pop('terminated_at', None)
target.pop('terminated_at_epoch', None)
target.pop('termination_mode', None)
@@ -166,7 +184,11 @@ target.pop('stopped_at', None)
target.pop('stopped_at_epoch', None)
target.pop('stop_reason', None)
target.pop('resumable', None)
target['last_visible_status'] = f'resumed conversation {uuid} at {now}'
if uuid:
target['last_visible_status'] = f'resumed conversation {uuid} at {now}'
else:
# ISSUE-3 Class A fresh-spawn resume: no conversation id existed yet.
target['last_visible_status'] = f'resumed (fresh spawn, no prior turn) at {now}'
target.setdefault('pane', {})
if pane_pid.isdigit():
@@ -178,15 +200,21 @@ if agent == 'claude':
target['claude_session_id_own'] = uuid
elif agent == 'agy':
target['pane']['cmd'] = 'agy'
if uuid:
target['pane']['cmd_full'] = f'agy --dangerously-skip-permissions --conversation {uuid}'
target['agy_conversation_id_own'] = uuid
else:
target['pane']['cmd_full'] = os.environ.get('CMD_FULL_OPT', '') or 'agy --dangerously-skip-permissions'
cp = os.environ.get('CHILD_PID', '0')
if cp.isdigit() and int(cp) > 0:
target['child_pid'] = int(cp)
elif agent == 'hermes':
target['pane']['cmd'] = 'hermes'
if uuid:
target['pane']['cmd_full'] = f'hermes --resume {uuid}'
target['hermes_conversation_id_own'] = uuid
else:
target['pane']['cmd_full'] = os.environ.get('CMD_FULL_OPT', '') or 'hermes --yolo --accept-hooks'
cp = os.environ.get('CHILD_PID', '0')
if cp.isdigit() and int(cp) > 0:
target['child_pid'] = int(cp)
@@ -199,8 +227,11 @@ elif agent == 'grok':
target['child_pid'] = int(cp)
elif agent == 'opencode':
target['pane']['cmd'] = 'opencode'
if uuid:
target['pane']['cmd_full'] = f'opencode --session {uuid} --auto --agent build'
target['opencode_session_id_own'] = uuid
else:
target['pane']['cmd_full'] = os.environ.get('CMD_FULL_OPT', '') or 'opencode --auto --agent build'
cp = os.environ.get('CHILD_PID', '0')
if cp.isdigit() and int(cp) > 0:
target['child_pid'] = int(cp)
+123
View File
@@ -1,5 +1,6 @@
import os
import json
import time
import uuid
import pytest
import subprocess
@@ -389,3 +390,125 @@ def test_t13_mam_workspace_key_equivalence(mam_sandbox):
abs_path = os.path.realpath(v)
py_key = abs_path.replace("/", "-").replace("_", "-")
assert sh_key == py_key, f"Mismatch for {v}: shell={sh_key}, py={py_key}"
def test_t14_resume_class_a_fresh_spawn_fallback(mam_sandbox, mock_herdr, mock_agents):
"""T-14 (Issue #3): agy (Class A) stopped at 0-turns has no conversation id to
resolve; resume_session.sh must fall back to a fresh spawn (RC=0, plain spawn
command) rather than hard-failing with 'No saved session'."""
ws = str(mam_sandbox / "resume_class_a")
os.makedirs(ws, exist_ok=True)
create_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-create" / "scripts" / "create_session.sh"
stop_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-stop" / "scripts" / "stop_session.sh"
resume_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resume_session.sh"
cmd = [
"bash", str(create_script),
"--workspace", ws,
"--agent", "agy",
"--role", "creator",
"--session", "test-t14-agy",
"--no-onboard"
]
subprocess.run(cmd, check=True, cwd=str(mam_sandbox))
# Stop the session without ever sending a prompt (0-turn stop).
subprocess.run(
["bash", str(stop_script), "--session", "test-t14-agy", "--agent", "agy"],
check=True, cwd=str(mam_sandbox)
)
yaml_path = mam_sandbox / ".mam" / "agent-sessions.yaml"
with open(yaml_path) as f:
d = yaml.safe_load(f)
session = [s for s in d["herdr_sessions"] if s["name"] == "test-t14-agy"][0]
assert session["status"] == "stopped"
assert not session.get("agy_conversation_id_own")
res = subprocess.run([
"bash", str(resume_script),
"--workspace", ws,
"--agent", "agy",
"--session", "test-t14-agy",
"--dry-run"
], capture_output=True, text=True, cwd=str(mam_sandbox))
assert res.returncode == 0, f"stderr: {res.stderr}"
assert "would spawn:" in res.stdout
assert "--dangerously-skip-permissions" in res.stdout
assert "--conversation" not in res.stdout
def test_t15_resume_class_b_still_hard_fails_when_truly_unresolvable(mam_sandbox, mock_herdr, mock_agents):
"""T-15 (Issue #3 regression guard): claude/grok (Class B) must NOT receive the
Class A fresh-spawn fallback - a claude session with no resolvable id at all
(never created) still hard-fails exactly as before."""
ws = str(mam_sandbox / "resume_class_b_missing")
os.makedirs(ws, exist_ok=True)
resume_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resume_session.sh"
res = subprocess.run([
"bash", str(resume_script),
"--workspace", ws,
"--agent", "claude",
"--session", "never-created-claude",
"--dry-run"
], capture_output=True, text=True, cwd=str(mam_sandbox))
assert res.returncode == 1
assert "ERROR: No saved session for" in res.stderr
def test_t16_resume_class_a_fresh_spawn_resets_discovery_epoch(mam_sandbox, mock_herdr, mock_agents):
"""T-16 (Issue #3 challenge refinement): a Class A (agy) 0-turn resume must reset
herdr_session_epoch to the resume timestamp so reconcile.sh does not auto-capture
stale conversations created between the original create and the resume."""
ws = str(mam_sandbox / "resume_class_a_epoch")
os.makedirs(ws, exist_ok=True)
create_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-create" / "scripts" / "create_session.sh"
stop_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-stop" / "scripts" / "stop_session.sh"
resume_script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resume_session.sh"
cmd = [
"bash", str(create_script),
"--workspace", ws,
"--agent", "agy",
"--role", "creator",
"--session", "test-t16-agy",
"--no-onboard"
]
subprocess.run(cmd, check=True, cwd=str(mam_sandbox))
yaml_path = mam_sandbox / ".mam" / "agent-sessions.yaml"
with open(yaml_path) as f:
d = yaml.safe_load(f)
initial_epoch = [s for s in d["herdr_sessions"] if s["name"] == "test-t16-agy"][0]["herdr_session_epoch"]
# Stop at 0-turns.
subprocess.run(
["bash", str(stop_script), "--session", "test-t16-agy", "--agent", "agy"],
check=True, cwd=str(mam_sandbox)
)
# Wait to guarantee the epoch advances by at least 1 second.
time.sleep(1.1)
# Resume the session (live execution, updating YAML).
res = subprocess.run([
"bash", str(resume_script),
"--workspace", ws,
"--agent", "agy",
"--session", "test-t16-agy"
], capture_output=True, text=True, cwd=str(mam_sandbox))
assert res.returncode == 0, f"stderr: {res.stderr}\nstdout: {res.stdout}"
with open(yaml_path) as f:
d = yaml.safe_load(f)
session = [s for s in d["herdr_sessions"] if s["name"] == "test-t16-agy"][0]
assert session["status"] == "running"
assert session["herdr_session_epoch"] > initial_epoch, (
f"herdr_session_epoch was not refreshed: {session['herdr_session_epoch']} <= {initial_epoch}"
)
assert session.get("session_id_source") == "pending-discovery"
assert session.get("session_id_verified") is False