fix(resume): no guess with two same-type rows in a cwd; refuse another row's id before spawning; the running branch uses the same decision (Issue #63)

This commit is contained in:
Godopu committed 2026-10-21 16:28:34 +00:00
1 parent 3a2219454e
commit fa93cc5622
4 files changed
+290 -27

No files matched your search

@@ -4,9 +4,9 @@
# bash resolve_session_id.sh --workspace <path> --agent <claude|agy|hermes|grok|opencode|codex>
# 출력: stdout 으로 UUID 한 줄 (없으면 빈 줄 + exit 0)
#
# P0-C: 전역 agent_identities 를 즉시 반환하지 않는다. lib.sh::find_workspace_uuid
# 가 워크스페이스 격리된 해결 경로(per-row own id -> 디스크 스캔 -> cwd 일치하는
# cache)만 사용. 다른 워크스페이스의 UUID 를 절대 반환하지 않음.
# P0-C / Issue #63: 전역 agent_identities 를 즉시 반환하지 않는다. lib.sh::mam_session_ids
# 가 워크스페이스 격리된 해결 경로(per-row own id -> #40 active capture / 디스크 스캔)만 사용.
# 다른 워크스페이스나 다른 행이 기록한 UUID 를 절대 반환하지 않음.
set -euo pipefail
source "$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)/lib.sh"
@@ -22,13 +22,17 @@ EOF
WORKSPACE=""
AGENT=""
SESSION_NAME=""
HERDR_ALIVE="0"
PANE_PID=""
while [ $# -gt 0 ]; do
case "$1" in
--workspace) WORKSPACE="$2"; shift 2 ;;
--agent) AGENT="$2"; shift 2 ;;
--session) SESSION_NAME="$2"; shift 2 ;;
-h|--help) usage; exit 0 ;;
--workspace) WORKSPACE="$2"; shift 2 ;;
--agent) AGENT="$2"; shift 2 ;;
--session) SESSION_NAME="$2"; shift 2 ;;
--herdr-alive) HERDR_ALIVE="$2"; shift 2 ;;
--pane-pid) PANE_PID="$2"; shift 2 ;;
-h|--help) usage; exit 0 ;;
*) echo "ERROR: unknown arg: $1" >&2; exit 2 ;;
esac
done
@@ -40,4 +44,6 @@ case "$AGENT" in
*) echo "ERROR: --agent must be claude, agy, hermes, grok, opencode, or codex" >&2; exit 2 ;;
esac
find_workspace_uuid "$WORKSPACE" "$AGENT" "$SESSION_NAME"
mam_session_ids resume --session "$SESSION_NAME" --agent "$AGENT" \
--workspace "$(mam_abs_workspace "$WORKSPACE")" \
--herdr-alive "$HERDR_ALIVE" --pane-pid "$PANE_PID"
@@ -107,9 +107,33 @@ if [ "$cp_rc" -ne 0 ] || [ -z "$CP_JSON" ]; then
CP_JSON=""
fi
# 1. Resolve the session id
UUID=$(bash "$(dirname "${BASH_SOURCE[0]}")/resolve_session_id.sh" \
--workspace "$WORKSPACE" --agent "$AGENT" --session "$SESSION_NAME")
if [ -n "$HERDR_SERVER_OPT" ]; then
export HERDR_SESSION_NAME="$HERDR_SERVER_OPT"
else
HERDR_SESSION_NAME="$(resolve_herdr_session "$SESSION_NAME" "$WORKSPACE")"
export HERDR_SESSION_NAME
fi
if [ -n "$HERDR_WORKSPACE_OPT" ]; then
export MAM_WS_LABEL="$HERDR_WORKSPACE_OPT"
else
export MAM_WS_LABEL="$(resolve_herdr_workspace "$SESSION_NAME" "$WORKSPACE")"
fi
HERDR_ALIVE=0
PANE_PID_NOW=""
if herdr has-session -t "$SESSION_NAME" 2>/dev/null; then
HERDR_ALIVE=1
PANE_PID_NOW=$(herdr list-panes -t "$SESSION_NAME" -F '#{pane_pid}' 2>/dev/null | head -1 || true)
fi
# 1. Resolve the session id (Issue #63: own id first, then #40 capture when
# herdr is alive, or fresh spawn when 2+ same-type rows share the cwd)
RESUME_OUT=$(mam_session_ids resume --shell \
--workspace "$(mam_abs_workspace "$WORKSPACE")" --agent "$AGENT" --session "$SESSION_NAME" \
--herdr-alive "$HERDR_ALIVE" --pane-pid "$PANE_PID_NOW")
eval "$RESUME_OUT"
UUID="$RESUME_UUID"
# ISSUE-3: agy/hermes/opencode (Class A) are spawned without a UUID - their
# own conversation id is only assigned by the agent itself on its first turn.
@@ -130,21 +154,8 @@ if [ -z "$UUID" ]; then
esac
fi
if [ -n "$HERDR_SERVER_OPT" ]; then
export HERDR_SESSION_NAME="$HERDR_SERVER_OPT"
else
HERDR_SESSION_NAME="$(resolve_herdr_session "$SESSION_NAME" "$WORKSPACE")"
export HERDR_SESSION_NAME
fi
if [ -n "$HERDR_WORKSPACE_OPT" ]; then
export MAM_WS_LABEL="$HERDR_WORKSPACE_OPT"
else
export MAM_WS_LABEL="$(resolve_herdr_workspace "$SESSION_NAME" "$WORKSPACE")"
fi
# 2. If herdr is alive, print warning or attach.
if herdr has-session -t "$SESSION_NAME" 2>/dev/null; then
if [ "$HERDR_ALIVE" = "1" ]; then
if [ "${DRY_RUN:-0}" = "1" ]; then
echo "[dry-run] herdr '$SESSION_NAME' already running — nothing to validate"
exit 0
@@ -153,7 +164,7 @@ if herdr has-session -t "$SESSION_NAME" 2>/dev/null; then
# Just update YAML to make sure it's set to running
CONTROL_PLANE_JSON="$CP_JSON" 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" --already-running \
${HERDR_WORKSPACE_OPT:+--herdr-workspace "$HERDR_WORKSPACE_OPT"}
exit 0
fi
@@ -252,9 +263,12 @@ handle_startup_dialogs "$SESSION_NAME" 20 "$AGENT"
sleep 2
# 5. Update agent-sessions.yaml: status running, last_visible_status
DROP_FLAG=""
[ "${DROP_STALE_ID:-0}" = "1" ] && DROP_FLAG="--drop-stale-id"
CONTROL_PLANE_JSON="$CP_JSON" bash "$(dirname "${BASH_SOURCE[0]}")/update_yaml_resumed.sh" \
--session "$SESSION_NAME" --uuid "$UUID" --agent "$AGENT" --workspace "$WORKSPACE" \
--herdr-session "$HERDR_SESSION_NAME" --cmd-full "$CMD_FULL" \
${DROP_FLAG:+$DROP_FLAG} \
${HERDR_WORKSPACE_OPT:+--herdr-workspace "$HERDR_WORKSPACE_OPT"}
echo "Successfully resumed $SESSION_NAME ($AGENT)"
@@ -27,6 +27,8 @@ ROLE=""
HERDR_SERVER_OPT=""
HERDR_WORKSPACE_OPT=""
CMD_FULL_OPT=""
ALREADY_RUNNING=0
DROP_STALE_ID=0
while [ $# -gt 0 ]; do
case "$1" in
@@ -38,6 +40,8 @@ while [ $# -gt 0 ]; do
--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 ;;
--already-running) ALREADY_RUNNING=1; shift ;;
--drop-stale-id) DROP_STALE_ID=1; shift ;;
-h|--help) usage; exit 0 ;;
*) echo "ERROR: unknown arg: $1" >&2; exit 2 ;;
esac
@@ -110,6 +114,7 @@ atomic_dump_yaml "$AGENT_SESSIONS_YAML" \
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="${MAM_WS_LABEL_EXPLICIT:-0}" \
ALREADY_RUNNING="$ALREADY_RUNNING" DROP_STALE_ID="$DROP_STALE_ID" \
CMD_FULL_OPT="$CMD_FULL_OPT" <<'PYEOF'
name = os.environ['SESSION_NAME']
uuid = os.environ['UUID']
@@ -173,7 +178,17 @@ if _cp_raw and not isinstance(target.get('control_plane'), dict):
target['control_plane'] = {k: _cp[k] for k in ('workspace_root', 'agent_sessions_yaml', 'mam_jobs_dir', 'delegate_job_logs_dir')}
except Exception:
pass
if not uuid:
stale_dropped = ''
if not uuid and os.environ.get('DROP_STALE_ID') == '1':
from lib_py.agents.autopin import claimed_elsewhere
from lib_py.agents.registry import get_adapter
_adapter = get_adapter(agent)
_own_key = _adapter.own_key if _adapter else ''
_old = (target.get(_own_key) or '').strip() if _own_key else ''
if _old and _old not in set(claimed_elsewhere(target, d)):
target[_own_key] = None
stale_dropped = _old
if not uuid and os.environ.get('ALREADY_RUNNING') != '1':
# 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
@@ -196,6 +211,10 @@ target.pop('stop_reason', None)
target.pop('resumable', None)
if uuid:
target['last_visible_status'] = f'resumed conversation {uuid} at {now}'
elif os.environ.get('ALREADY_RUNNING') == '1':
target['last_visible_status'] = f'already running; no id recorded at {now}'
elif stale_dropped:
target['last_visible_status'] = f'resumed (fresh spawn, recorded id {stale_dropped} no longer verified) 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}'
+224
View File
@@ -389,3 +389,227 @@ def test_c2_purge_never_deletes_another_rows_conversation(mam_sandbox, mock_herd
assert rows[name_a]["agy_conversation_id_own"] == UA
# --------------------------------------------------------------------------- Done Criteria 3, 4, 7 & rev 2 (resume)
def _run_resume(sandbox: Path, ws: Path, session: str, agent: str = "agy",
dry_run: bool = False) -> subprocess.CompletedProcess:
script = sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resume_session.sh"
cmd = ["bash", str(script), "--workspace", str(ws), "--agent", agent, "--session", session]
if dry_run:
cmd.append("--dry-run")
return subprocess.run(cmd, capture_output=True, text=True, cwd=str(sandbox), timeout=120)
def test_c3_resume_unpinned_second_agy_row_starts_fresh_and_running_branch_keeps_epoch(
mam_sandbox, mock_herdr, mock_agents
):
"""Done Criterion 3: two stopped agy rows in one cwd (A=UA, B=None):
dry-run and real resume of B spawn without --conversation UA and emit NOTE;
when B's herdr is already alive without an id, resume does not record UA and
preserves B's herdr_session_epoch."""
ws = mam_sandbox / "same-ws"
ws.mkdir(parents=True, exist_ok=True)
epoch_a = int(time.time()) - 300
epoch_b = int(time.time()) - 200
_agy_full_artifacts(mam_sandbox, UA)
_agy_cache(mam_sandbox, {ws: UA})
name_a, name_b = "same-ws-creator-agy", "same-ws-reviewer-agy"
_write_rows(mam_sandbox, [
_row(name_a, "agy", ws, epoch_a, status="stopped", agy_conversation_id_own=UA,
session_id_source="discovered", session_id_verified=True),
_row(name_b, "agy", ws, epoch_b, status="stopped", agy_conversation_id_own=None),
])
dry = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=True)
assert dry.returncode == 0, dry.stdout + dry.stderr
assert "--conversation" not in dry.stdout
assert f"NOTE: {name_b} has no recorded agy id and 2 agy sessions share" in dry.stderr
real = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=False)
assert real.returncode == 0, real.stdout + real.stderr
rows = _read_rows(mam_sandbox)
assert rows[name_b]["status"] == "running"
assert rows[name_b]["agy_conversation_id_own"] is None
assert "--conversation" not in rows[name_b]["pane"]["cmd_full"]
# Running branch: B is already alive in herdr without an own id; epoch must stay fixed.
fixed_epoch = 1700000123
_write_rows(mam_sandbox, [
_row(name_a, "agy", ws, epoch_a, status="stopped", agy_conversation_id_own=UA,
session_id_source="discovered", session_id_verified=True),
_row(name_b, "agy", ws, fixed_epoch, status="running", agy_conversation_id_own=None),
])
_mock_herdr_agent(mock_herdr, name_b, ws, 9999, fixed_epoch, pane_id="w1:p_b")
alive_res = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=False)
assert alive_res.returncode == 0, alive_res.stdout + alive_res.stderr
rows = _read_rows(mam_sandbox)
assert rows[name_b]["agy_conversation_id_own"] is None
assert rows[name_b]["herdr_session_epoch"] == fixed_epoch
assert rows[name_b]["last_visible_status"].startswith("already running; no id recorded at ")
def test_c4_pinned_post_discovered_and_preassigned_agents_unchanged(mam_sandbox, mock_herdr, mock_agents):
"""Done Criterion 4: agy/opencode rows with own_id keep it on stop and resume with it;
claude/grok stop/resume and find_workspace_uuid behave as before."""
ws = mam_sandbox / "c4-ws"
ws.mkdir(parents=True, exist_ok=True)
epoch = int(time.time()) - 100
_agy_full_artifacts(mam_sandbox, UA)
_claude_transcript(mam_sandbox / ".claude" / "projects", ws, UB)
# Seed opencode DB for ses_c4
oc_db = mam_sandbox / ".local" / "share" / "opencode" / "opencode.db"
oc_db.parent.mkdir(parents=True, exist_ok=True)
conn = sqlite3.connect(str(oc_db))
conn.execute("CREATE TABLE IF NOT EXISTS session_v2 (id TEXT PRIMARY KEY, parent_id TEXT, directory TEXT, time_created INTEGER)")
conn.execute("INSERT OR REPLACE INTO session_v2 VALUES (?, NULL, ?, ?)", ("ses_c4", str(ws), epoch * 1000))
conn.commit()
conn.close()
_write_rows(mam_sandbox, [
_row("c4-creator-agy", "agy", ws, epoch, status="running", agy_conversation_id_own=UA,
session_id_source="discovered", session_id_verified=True),
_row("c4-creator-opencode", "opencode", ws, epoch, status="running", opencode_session_id_own="ses_c4",
session_id_source="discovered", session_id_verified=True),
_row("c4-creator-claude", "claude", ws, epoch, status="running", claude_session_id_own=UB,
session_id_source="assigned", session_id_verified=True),
])
for sname, agent in (("c4-creator-agy", "agy"), ("c4-creator-opencode", "opencode"), ("c4-creator-claude", "claude")):
st = _run_stop(mam_sandbox, sname, agent)
assert st.returncode == 0, st.stdout + st.stderr
rows = _read_rows(mam_sandbox)
assert rows["c4-creator-agy"]["agy_conversation_id_own"] == UA and rows["c4-creator-agy"]["resumable"] is True
assert rows["c4-creator-opencode"]["opencode_session_id_own"] == "ses_c4" and rows["c4-creator-opencode"]["resumable"] is True
assert rows["c4-creator-claude"]["claude_session_id_own"] == UB and rows["c4-creator-claude"]["resumable"] is True
r_agy = _run_resume(mam_sandbox, ws, "c4-creator-agy", "agy", dry_run=True)
assert r_agy.returncode == 0 and f"--conversation {UA}" in r_agy.stdout
r_oc = _run_resume(mam_sandbox, ws, "c4-creator-opencode", "opencode", dry_run=True)
assert r_oc.returncode == 0 and "--session ses_c4" in r_oc.stdout
r_cl = _run_resume(mam_sandbox, ws, "c4-creator-claude", "claude", dry_run=True)
assert r_cl.returncode == 0 and f"-r {UB}" in r_cl.stdout
@pytest.mark.parametrize("db_exists", [True, False])
def test_c7_resume_refuses_when_recorded_id_is_claimed_by_another_row(
mam_sandbox, mock_herdr, mock_agents, db_exists
):
"""Done Criterion 7 & rev 2 (A): when both A and B record UA (whether UA.db
exists on disk or is missing), resume B refuses with exit 1 before spawning,
names A on stderr, and emits no fresh-spawn NOTE."""
ws = mam_sandbox / "c7-ws"
ws.mkdir(parents=True, exist_ok=True)
epoch = int(time.time()) - 100
if db_exists:
_agy_full_artifacts(mam_sandbox, UA)
_agy_cache(mam_sandbox, {ws: UA})
name_a, name_b = "c7-ws-creator-agy", "c7-ws-reviewer-agy"
_write_rows(mam_sandbox, [
_row(name_a, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA,
session_id_source="discovered", session_id_verified=True),
_row(name_b, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA,
session_id_source="discovered", session_id_verified=True),
])
before = _read_rows(mam_sandbox)
res = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=False)
assert res.returncode == 1, res.stdout + res.stderr
assert name_a in res.stderr and UA in res.stderr
assert "NOTE:" not in res.stderr
state = json.loads(Path(mock_herdr).read_text())
assert name_b not in state.get("agents", {})
assert _read_rows(mam_sandbox) == before
def test_rev2_stale_own_id_single_row_never_overwrites_with_cache_and_clears_on_resume(
mam_sandbox, mock_herdr, mock_agents
):
"""rev 2 (A)-3: single agy row with stale own_id UX (UX.db missing) while cache
and disk hold UC. Stop keeps UX; resume starts a fresh conversation without UC,
emits NOTE, and clears agy_conversation_id_own to None so #40 can pin later."""
ws = mam_sandbox / "stale-ws"
ws.mkdir(parents=True, exist_ok=True)
epoch = int(time.time()) - 100
_agy_full_artifacts(mam_sandbox, UC)
_agy_cache(mam_sandbox, {ws: UC})
name = "stale-ws-creator-agy"
_write_rows(mam_sandbox, [
_row(name, "agy", ws, epoch, status="running", agy_conversation_id_own=UX,
session_id_source="discovered", session_id_verified=True),
])
st = _run_stop(mam_sandbox, name, "agy")
assert st.returncode == 0, st.stdout + st.stderr
row_after_stop = _read_rows(mam_sandbox)[name]
assert row_after_stop["agy_conversation_id_own"] == UX
assert "no longer verifies; kept as is" in row_after_stop["last_visible_status"]
res = _run_resume(mam_sandbox, ws, name, "agy", dry_run=False)
assert res.returncode == 0, res.stdout + res.stderr
assert f"NOTE: {name}'s recorded agy id {UX} no longer verifies" in res.stderr
row_after_resume = _read_rows(mam_sandbox)[name]
assert row_after_resume["agy_conversation_id_own"] is None
assert row_after_resume["session_id_source"] == "pending-discovery"
assert row_after_resume["session_id_verified"] is False
assert f"recorded id {UX} no longer verified" in row_after_resume["last_visible_status"]
assert UC not in row_after_resume["pane"]["cmd_full"]
def test_rev2_drop_stale_id_refuses_to_clear_id_claimed_by_another_row(mam_sandbox):
"""rev 2: update_yaml_resumed.sh --drop-stale-id keeps own_id if another row also claims it."""
ws = mam_sandbox / "dup-drop-ws"
ws.mkdir(parents=True, exist_ok=True)
epoch = int(time.time()) - 100
name_a, name_b = "dup-creator-agy", "dup-reviewer-agy"
_write_rows(mam_sandbox, [
_row(name_a, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA),
_row(name_b, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA),
])
upd = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "update_yaml_resumed.sh"
res = subprocess.run(
["bash", str(upd), "--session", name_b, "--uuid", "", "--agent", "agy",
"--workspace", str(ws), "--drop-stale-id"],
capture_output=True, text=True, cwd=str(mam_sandbox), timeout=60,
)
assert res.returncode == 0, res.stdout + res.stderr
assert _read_rows(mam_sandbox)[name_b]["agy_conversation_id_own"] == UA
def test_rev2_claude_unpinned_sibling_row_skips_creator_claimed_id(mam_sandbox, mock_herdr, mock_agents):
"""rev 2 (B): ws-creator-claude is stopped with UA; ws-reviewer-claude has no
own_id (--wrapper row shape); disk has UA.jsonl and UB.jsonl. Resume and stop
of ws-reviewer-claude pick UB instead of creator's UA."""
ws = mam_sandbox / "claude-sib-ws"
ws.mkdir(parents=True, exist_ok=True)
now = time.time()
epoch = int(now) - 100
cdir = mam_sandbox / ".claude" / "projects"
_claude_transcript(cdir, ws, UA, mtime=now)
_claude_transcript(cdir, ws, UB, mtime=now - 10)
name_a, name_b = "claude-sib-ws-creator-claude", "claude-sib-ws-reviewer-claude"
_write_rows(mam_sandbox, [
_row(name_a, "claude", ws, epoch, status="stopped", claude_session_id_own=UA,
session_id_source="discovered", session_id_verified=True),
_row(name_b, "claude", ws, epoch, status="running", claude_session_id_own=None),
])
dry = _run_resume(mam_sandbox, ws, name_b, "claude", dry_run=True)
assert dry.returncode == 0, dry.stdout + dry.stderr
st = _run_stop(mam_sandbox, name_b, "claude")
assert st.returncode == 0, st.stdout + st.stderr
assert _read_rows(mam_sandbox)[name_b]["claude_session_id_own"] == UB
dry_after = _run_resume(mam_sandbox, ws, name_b, "claude", dry_run=True)
assert dry_after.returncode == 0, dry_after.stdout + dry_after.stderr
assert f"-r {UB}" in dry_after.stdout