fix(#23, #24): 교차 서버 herdr 스코프와 에이전트 무관 시작 다이얼로그, 실제 에이전트가 충족하는 온보딩 전달 판정 #35

Merged
Godopu merged 15 commits from fix/issue-23-24 into main 2026-09-23 23:36:37 +00:00
3 changed files with 372 additions and 2 deletions
Showing only changes of commit 3bcaa39ab0 - Show all commits

No files matched your search

+42 -1
View File
@@ -1373,9 +1373,50 @@ EOF
fi
}
# _herdr_call_is_cross_server
#
# ISSUE-23: a herdr pane exports HERDR_WORKSPACE_ID / HERDR_PANE_ID /
# HERDR_TAB_ID, and those ids only mean something on the server that issued
# them. When a MAM script runs inside one server's pane (the orchestrator's)
# but targets another (the isolated MAM server named by HERDR_SESSION_NAME),
# the inherited ids reach the scope filters (_herdr_ws_scope,
# _herdr_agent_get_scoped, _resolve_herdr_pane_id) as a foreign workspace, and
# their fail-closed miss (lib.sh explicit-env miss, ISSUE-3 / F-2b) reports a
# live session as absent: has-session, kill-session, agent prompt and the
# delegate-job preflight all answer "no such session".
#
# True only for a call aimed at a *named* server other than the caller's own:
#
# * target unset or `default` (the lib.sh:97 fallback) is the shim's sentinel
# for "pass no session flag", so herdr answers on the ambient server, which
# for a caller inside a pane is that same caller's server - not a crossing;
# * no HERDR_SESSION means the caller is not in a pane at all, so an exported
# HERDR_WORKSPACE_ID is a deliberate scope, not a leak (H-18/H-21/H-23);
# * caller == target is the same server.
_herdr_call_is_cross_server() {
local caller="${HERDR_SESSION:-}" target="${HERDR_SESSION_NAME:-}"
case "$target" in
""|default) return 1 ;;
esac
if [ -n "$caller" ] && [ "$caller" != "$target" ]; then
return 0
fi
return 1
}
mam_herdr() {
_init_herdr_isolation
"$WORKSPACE_ROOT/.mam/shim/herdr" "$@"
# Built as an argv array and decided inside `if`: this runs under
# `set -euo pipefail` in most callers, where a bare failing test would abort
# the whole script instead of this one call.
local -a _scope_strip=()
if _herdr_call_is_cross_server; then
_scope_strip=(env -u HERDR_WORKSPACE_ID -u HERDR_PANE_ID -u HERDR_TAB_ID)
if [ "${MAM_SHIM_DEBUG:-0}" = "1" ]; then
echo "mam_herdr: dropping caller pane scope ids (caller server '${HERDR_SESSION:-}' != target '${HERDR_SESSION_NAME:-}')" >&2
fi
fi
${_scope_strip[@]+"${_scope_strip[@]}"} "$WORKSPACE_ROOT/.mam/shim/herdr" "$@"
}
_herdr() {
+26 -1
View File
@@ -19,6 +19,7 @@ from __future__ import annotations
import json
import logging
import os
import subprocess
from typing import Any, Dict, Optional, Sequence
@@ -38,6 +39,30 @@ TRANSPORT_ERROR = "transport_error"
BACKOFF_START_S = 1.0
BACKOFF_MAX_S = 16.0
# ISSUE-23: ids a herdr pane exports about ITS OWN server. Passing them to a
# call aimed at a different server makes herdr scope the lookup to a workspace
# that does not exist there, and the target is reported as missing. The shell
# side drops them in mam_herdr; this is the one python path that runs herdr
# without going through it.
CALLER_SCOPE_VARS = ("HERDR_WORKSPACE_ID", "HERDR_PANE_ID", "HERDR_TAB_ID")
def scoped_env(session=None, environ=None):
"""The environment for a wait aimed at `session`.
Unchanged unless this process runs inside a pane of another server: only
then are the caller's own workspace/pane/tab ids meaningless for the call
and dropped. No pane (no HERDR_SESSION) and same-server keep every value.
"""
src = os.environ if environ is None else environ
caller = (src.get("HERDR_SESSION") or "").strip()
target = (session or "").strip()
# `default` is the same sentinel build_argv reads: no --session flag, so
# herdr answers on the ambient server - the caller's own.
if not caller or not target or target == "default" or caller == target:
return dict(src)
return dict((k, v) for k, v in src.items() if k not in CALLER_SCOPE_VARS)
class WaitOutcome(object):
__slots__ = ("kind", "agent", "error_code", "detail", "raw")
@@ -92,7 +117,7 @@ def wait_once(target: str, until: Sequence[str] = DEFAULT_UNTIL,
"""One barrier call. `run` is injectable so tests never touch a real herdr."""
argv = build_argv(target, until, timeout_ms, session=session)
try:
proc = run(argv, capture_output=True, text=True)
proc = run(argv, capture_output=True, text=True, env=scoped_env(session))
except Exception as exc:
return WaitOutcome(TRANSPORT_ERROR, detail="herdr could not be executed: {}".format(exc))
+304
View File
@@ -0,0 +1,304 @@
"""Issue #23 + #24 — the create_session start path.
Four regressions live here, in the order of the issues:
1. a herdr call aimed at another server must not carry the caller pane's own
workspace/pane/tab ids (#23 comments: reproduced on Linux and macOS);
2. blocking start-up modals are agent-adapter knowledge and are answered while
waiting for the TUI (#24);
3. onboarding delivery is judged by the agent's own `started` event (#23);
4. a rollback verifies the kill and leaves no `running` row or dangling job (#23).
What is mocked and what is real is stated per test. The dialog tests fix OUR
contract - which pane string gets which keys - not that the keys drive a real
agy/grok/codex/opencode TUI; no such TUI can be started here.
"""
import json
import os
import subprocess
import textwrap
import pytest
# --------------------------------------------------------------------------
# helpers
# --------------------------------------------------------------------------
def _run_lib(tmp_path, body, env=None):
"""Source the sandbox lib.sh and run `body`. Unlike the shim-contract
helper this keeps HERDR_* exactly as `env` sets it: these tests are about
what the caller's pane leaks in."""
lib_path = tmp_path / ".agents" / "skills" / "lib.sh"
script = 'source "%s"\n_init_herdr_isolation\n%s\n' % (lib_path, body)
run_env = dict(os.environ)
for var in ("HERDR_SESSION", "HERDR_SESSION_NAME", "HERDR_WORKSPACE_ID",
"HERDR_PANE_ID", "HERDR_TAB_ID"):
run_env.pop(var, None)
run_env.update(env or {})
return subprocess.run(["bash", "-c", script], capture_output=True,
text=True, cwd=str(tmp_path), env=run_env)
def _mock_state(mock_herdr):
with open(mock_herdr) as f:
return json.load(f)
def _write_mock_state(mock_herdr, state):
with open(mock_herdr, "w") as f:
json.dump(state, f, indent=2)
def _seed_agent(mock_herdr, tmp_path, name, workspace_id, pane_id):
state = _mock_state(mock_herdr)
state.setdefault("workspaces", [])
if not any(w.get("workspace_id") == workspace_id for w in state["workspaces"]):
state["workspaces"].append(
{"workspace_id": workspace_id, "label": workspace_id, "cwd": str(tmp_path)})
state.setdefault("agents", {})[name] = {
"name": name, "agent": "agy", "status": "running", "cwd": str(tmp_path),
"workspace_id": workspace_id, "pane_id": pane_id, "pid": 9999,
"command": "agy", "buffer": "Ready",
}
state.setdefault("panes", []).append(
{"pane_id": pane_id, "workspace_id": workspace_id, "label": name, "buffer": "Ready"})
_write_mock_state(mock_herdr, state)
def _install_env_recorder(tmp_path, monkeypatch):
"""A herdr earlier in PATH than the mock that dumps the env it was called
with, then execs the mock. The shim resolves the first non-shim herdr on
PATH, so this observes exactly what reached the real binary."""
rec_dir = tmp_path / "recorder"
rec_dir.mkdir(parents=True, exist_ok=True)
env_log = tmp_path / "herdr_env.ndjson"
real = tmp_path / "bin" / "herdr"
script = rec_dir / "herdr"
script.write_text(textwrap.dedent('''\
#!/usr/bin/env python3
import json, os, sys
with open(%r, "a") as f:
f.write(json.dumps({
"argv": sys.argv[1:],
"HERDR_WORKSPACE_ID": os.environ.get("HERDR_WORKSPACE_ID"),
"HERDR_PANE_ID": os.environ.get("HERDR_PANE_ID"),
"HERDR_TAB_ID": os.environ.get("HERDR_TAB_ID"),
"HERDR_SESSION": os.environ.get("HERDR_SESSION"),
"HERDR_SESSION_NAME": os.environ.get("HERDR_SESSION_NAME"),
}) + "\\n")
os.execv(%r, [%r] + sys.argv[1:])
''') % (str(env_log), str(real), str(real)))
script.chmod(0o755)
monkeypatch.setenv("PATH", "%s:%s" % (rec_dir, os.environ["PATH"]))
return env_log
def _recorded(env_log):
if not os.path.exists(env_log):
return []
return [json.loads(l) for l in open(env_log) if l.strip()]
# --------------------------------------------------------------------------
# item 1 — caller pane scope must not leak across servers
# --------------------------------------------------------------------------
CROSS = {"HERDR_SESSION": "default", "HERDR_SESSION_NAME": "mam-isolated",
"HERDR_WORKSPACE_ID": "w5", "HERDR_PANE_ID": "w5:p2",
"HERDR_TAB_ID": "w5:t1"}
def test_s1_cross_server_call_drops_the_caller_panes_scope_ids(
mam_sandbox, mock_herdr, mock_agents, monkeypatch):
"""S1 (mock herdr): the caller sits in another server's pane. has-session,
kill-session and agent prompt must find the agent, and the three ids must
not reach herdr."""
tmp_path = mam_sandbox
env_log = _install_env_recorder(tmp_path, monkeypatch)
_seed_agent(mock_herdr, tmp_path, "creator-agy-01", "w1", "w1:p5")
res = _run_lib(tmp_path, 'herdr has-session -t creator-agy-01', env=CROSS)
assert res.returncode == 0, res.stdout + res.stderr
res = _run_lib(tmp_path, 'herdr agent prompt creator-agy-01 "hello"', env=CROSS)
assert res.returncode == 0, res.stdout + res.stderr
res = _run_lib(tmp_path, 'herdr kill-session -t creator-agy-01', env=CROSS)
assert res.returncode == 0, res.stdout + res.stderr
calls = _recorded(env_log)
assert calls, "the recorder saw no herdr call"
for call in calls:
assert call["HERDR_WORKSPACE_ID"] is None, call
assert call["HERDR_PANE_ID"] is None, call
assert call["HERDR_TAB_ID"] is None, call
# the kill really happened, i.e. this is not a silent no-op
assert "creator-agy-01" not in _mock_state(mock_herdr).get("agents", {})
def test_s1b_the_delegate_job_wrappers_bare_herdr_is_the_lib_sh_function(
mam_sandbox, mock_herdr, mock_agents, monkeypatch):
"""S1b (mock herdr): the wrapper sources lib.sh and then calls a bare
`herdr` (wrapper :918 preflight). A shell function beats PATH, so that call
is mam_herdr and is fixed by item 1 too - this is why #23's preflight
failure is cured by the same change."""
tmp_path = mam_sandbox
_install_env_recorder(tmp_path, monkeypatch)
_seed_agent(mock_herdr, tmp_path, "creator-agy-01", "w1", "w1:p5")
res = _run_lib(tmp_path, 'echo "kind=$(type -t herdr)"\nherdr has-session -t creator-agy-01',
env=CROSS)
assert "kind=function" in res.stdout, res.stdout + res.stderr
assert res.returncode == 0, res.stdout + res.stderr
def test_s2_same_server_keeps_the_explicit_workspace_scope(
mam_sandbox, mock_herdr, mock_agents, monkeypatch):
"""S2 (mock herdr): on the target's own server the exported scope still
scopes - a same-named agent in another workspace is not addressed (H-22)."""
tmp_path = mam_sandbox
_seed_agent(mock_herdr, tmp_path, "creator-agy-01", "w1", "w1:p5")
same = {"HERDR_SESSION": "mam-isolated", "HERDR_SESSION_NAME": "mam-isolated",
"HERDR_WORKSPACE_ID": "w2"}
before = len(_mock_state(mock_herdr).get("calls", []))
res = _run_lib(tmp_path, 'herdr agent prompt creator-agy-01 "hello from w2"', env=same)
assert res.returncode != 0, res.stdout + res.stderr
new = _mock_state(mock_herdr).get("calls", [])[before:]
prompts = [c for c in new if len(c) >= 2 and c[0] == "agent" and c[1] == "prompt"]
assert prompts == [], new
def test_s2b_a_caller_outside_any_pane_keeps_the_explicit_scope(
mam_sandbox, mock_herdr, mock_agents):
"""S2b (mock herdr): no HERDR_SESSION means the caller is not in a pane at
all, so an exported HERDR_WORKSPACE_ID is a deliberate scope, not a leak.
This is the case H-18/H-21/H-23 fix."""
tmp_path = mam_sandbox
_seed_agent(mock_herdr, tmp_path, "creator-agy-01", "w1", "w1:p5")
res = _run_lib(tmp_path, 'herdr has-session -t creator-agy-01',
env={"HERDR_SESSION_NAME": "mam-isolated", "HERDR_WORKSPACE_ID": "w2"})
assert res.returncode == 1, "an explicit out-of-scope id must still hide the agent"
res = _run_lib(tmp_path, 'herdr has-session -t creator-agy-01',
env={"HERDR_SESSION_NAME": "mam-isolated", "HERDR_WORKSPACE_ID": "w1"})
assert res.returncode == 0, res.stdout + res.stderr
def test_s3_shim_contract_regressions_still_pass(mam_sandbox):
"""S3: H-18 / H-21 / H-23 are the tests that fix today's scoping meaning.
Run them explicitly as this change's regression guard."""
repo = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
res = subprocess.run(
[os.path.join(repo, ".venv", "bin", "python"), "-m", "pytest", "-q",
"-p", "no:cacheprovider",
"tests/test_herdr_shim_contract.py", "-k",
"h18 or h21 or h22 or h23"],
capture_output=True, text=True, cwd=repo,
env=dict(os.environ, PYTHONDONTWRITEBYTECODE="1"))
assert res.returncode == 0, res.stdout[-4000:] + res.stderr[-2000:]
def test_s11_this_branch_does_not_change_the_generated_shim(mam_sandbox):
"""S11: the item-1 fix lives in mam_herdr, never in the heredoc that
lib.sh materialises into $WORKSPACE_ROOT/.mam/shim/herdr. That file is
shared by every checkout bound to one control plane, so keeping it
byte-identical to base is what makes "the control plane is not modified"
true at runtime."""
repo = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
base = os.environ.get("MAM_BASE_SHA", "1e78dae")
show = subprocess.run(["git", "show", "%s:.agents/skills/lib.sh" % base],
capture_output=True, text=True, cwd=repo)
if show.returncode != 0:
pytest.skip("base %s not available in this checkout" % base)
with open(os.path.join(repo, ".agents", "skills", "lib.sh")) as f:
head = f.read()
assert _heredoc_body(head) == _heredoc_body(show.stdout), (
"the shim heredoc changed: the generated .mam/shim/herdr of this branch "
"would differ from the control root's, and the two would overwrite "
"each other on every herdr call")
def _heredoc_body(text):
lines = text.splitlines(True)
start = None
for i, line in enumerate(lines):
if line.strip() == "cat <<'EOF' > \"$tmp_file\"":
start = i + 1
break
assert start is not None, "shim heredoc not found in lib.sh"
for j in range(start, len(lines)):
if lines[j].rstrip("\n") == "EOF":
return "".join(lines[start:j])
raise AssertionError("unterminated shim heredoc")
@pytest.mark.parametrize("env,expected_rc", [(CROSS, 1), ({"HERDR_SESSION_NAME": "mam-isolated"}, 1)])
def test_s12_mam_herdr_never_aborts_its_caller_under_set_e(
mam_sandbox, mock_herdr, mock_agents, env, expected_rc):
"""S12 (mock herdr): mam_herdr runs inside scripts using `set -euo
pipefail`. Neither branch of the cross-server test may abort the shell,
and the child's rc must be the function's rc."""
tmp_path = mam_sandbox
body = ('set -euo pipefail\n'
'rc=0\n'
'mam_herdr has-session -t nobody-here || rc=$?\n'
'echo "missing_rc=$rc"\n'
'_seen=0\n'
'mam_herdr has-session -t creator-agy-01 || _seen=$?\n'
'echo "present_rc=$_seen"\n'
'echo REACHED_END\n')
_seed_agent(mock_herdr, tmp_path, "creator-agy-01", "w1", "w1:p5")
lib_path = tmp_path / ".agents" / "skills" / "lib.sh"
script = 'source "%s"\n%s' % (lib_path, body)
run_env = dict(os.environ)
for var in ("HERDR_SESSION", "HERDR_SESSION_NAME", "HERDR_WORKSPACE_ID",
"HERDR_PANE_ID", "HERDR_TAB_ID"):
run_env.pop(var, None)
run_env.update(env)
res = subprocess.run(["bash", "-c", script], capture_output=True, text=True,
cwd=str(tmp_path), env=run_env)
assert "REACHED_END" in res.stdout, res.stdout + res.stderr
assert "missing_rc=%d" % expected_rc in res.stdout, res.stdout + res.stderr
assert "present_rc=0" in res.stdout, res.stdout + res.stderr
def test_s13_the_loop_watcher_applies_the_same_rule(mam_sandbox):
"""S13 (pure unit): watcher.py builds its own herdr argv and never goes
through mam_herdr, so it needs the same rule."""
import sys
repo = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.insert(0, os.path.join(repo, ".agents", "skills"))
from lib_py.loop import watcher
caller_pane = {"HERDR_SESSION": "default", "HERDR_WORKSPACE_ID": "w5",
"HERDR_PANE_ID": "w5:p2", "HERDR_TAB_ID": "w5:t1", "PATH": "/usr/bin"}
cross = watcher.scoped_env("mam-isolated", environ=caller_pane)
assert "HERDR_WORKSPACE_ID" not in cross and "HERDR_PANE_ID" not in cross
assert "HERDR_TAB_ID" not in cross and cross["PATH"] == "/usr/bin"
same = watcher.scoped_env("default", environ=caller_pane)
assert same["HERDR_WORKSPACE_ID"] == "w5"
outside = watcher.scoped_env("mam-isolated", environ={"HERDR_WORKSPACE_ID": "w5"})
assert outside["HERDR_WORKSPACE_ID"] == "w5"
none_target = watcher.scoped_env(None, environ=caller_pane)
assert none_target["HERDR_WORKSPACE_ID"] == "w5"
seen = {}
def _run(argv, **kw):
seen["env"] = kw.get("env")
class P:
returncode, stdout, stderr = 0, "{}", ""
return P()
watcher.wait_once("agt", session="iso-x", run=_run)
assert seen["env"] is not None and "HERDR_SESSION" not in ("",)