mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-11-17 19:32:45 +03:00
30b8d5a2d3
do_update reinstalled every update via do_install(force=True). That one flag meant both "replace the installed copy" and the Skills Guard override, so a caution-verdict community update was shown "Decision: BLOCKED" and installed anyway (CLI, /skills update, and the dashboard which shells out to it). Split the two: do_install gains reinstall_existing (replace an installed copy only) and do_update passes force=False with it, plus skip_confirm and the installed name so the non-interactive update path keeps working. A blocked update now counts as "Not updated" and exits non-zero. `update --force` still only means "overwrite local edits". Fixes #38041 Co-authored-by: Coy Geek <65363919+coygeek@users.noreply.github.com>
128 lines
6.2 KiB
Python
128 lines
6.2 KiB
Python
"""A failed ``hermes skills install|update|uninstall|snapshot import`` exits non-zero (#118301).
|
|
|
|
The Desktop spawns these as detached actions and learns their outcome only from the
|
|
exit code (``apps/desktop/src/store/hub-actions.ts`` toasts on non-zero), so a refused
|
|
action that exits 0 reads as "the button did nothing" while the reason sits unread in
|
|
the action log. The scan gate is the common case: a blocked install used to print
|
|
"Not installed: ..." and exit 0.
|
|
"""
|
|
|
|
import sys
|
|
import pytest
|
|
|
|
import hermes_cli.skills_hub as cli_hub
|
|
|
|
|
|
def _scan_gate_env(monkeypatch, tmp_path, *, verdict="dangerous", installed=None):
|
|
"""Wire ``do_install`` down to the real scan-policy decision for a community skill."""
|
|
import tools.skills_guard as guard
|
|
import tools.skills_hub as hub
|
|
import tools.skills_hub_install as hub_install
|
|
|
|
bundle = type("Bundle", (), {
|
|
"name": "risky-skill", "files": {"SKILL.md": "---\nname: risky-skill\n---\n"},
|
|
"source": "skills-sh", "identifier": "org/repo/risky-skill", "trust_level": "community",
|
|
"metadata": {},
|
|
})()
|
|
q_path = tmp_path / "skills" / ".hub" / "quarantine" / "risky-skill"
|
|
q_path.mkdir(parents=True)
|
|
audit: list = []
|
|
|
|
monkeypatch.setattr(hub, "ensure_hub_dirs", lambda: None)
|
|
monkeypatch.setattr(hub, "append_audit_log", lambda *a, **k: audit.append(a))
|
|
monkeypatch.setattr(hub, "HubLockFile", lambda: type(
|
|
"Lock", (), {"get_installed": lambda self, n: installed})())
|
|
monkeypatch.setattr(cli_hub, "_sources", lambda: [type("Source", (), {"source_id": lambda self: "skills-sh"})()])
|
|
monkeypatch.setattr(cli_hub, "_resolve_source_meta_and_bundle",
|
|
lambda identifier, sources: (None, bundle, sources[0]))
|
|
monkeypatch.setattr(cli_hub, "_record_skill_install", lambda *a, **k: None)
|
|
monkeypatch.setattr(hub_install, "quarantine_bundle", lambda b: q_path)
|
|
finding = guard.Finding("remote_fetch", "critical", "supply_chain", "SKILL.md", 40,
|
|
'curl -s "https://example.com"', "remote fetch")
|
|
monkeypatch.setattr(cli_hub, "_scan_quarantined", lambda *a, **k: guard.ScanResult(
|
|
skill_name="risky-skill", source="org/repo/risky-skill", trust_level="community",
|
|
verdict=verdict, findings=[finding, finding]))
|
|
installs: list = []
|
|
|
|
def _install(q, name, category, b, result):
|
|
installs.append(name)
|
|
d = tmp_path / "skills" / name
|
|
d.mkdir(parents=True, exist_ok=True)
|
|
return d
|
|
|
|
monkeypatch.setattr(hub_install, "install_from_quarantine", _install)
|
|
monkeypatch.setattr(hub, "SKILLS_DIR", tmp_path / "skills")
|
|
return installs, audit
|
|
|
|
|
|
# --- the process exit code, through the real `hermes` entry point ---
|
|
|
|
|
|
@pytest.mark.parametrize("argv, target, outcome, code", [
|
|
(["skills", "install", "org/repo/x", "--yes"], "do_install", False, 1),
|
|
(["skills", "install", "org/repo/x", "--yes"], "do_install", True, 0),
|
|
(["skills", "install", "org/repo/x", "--yes"], "do_install", None, 0),
|
|
(["skills", "update"], "do_update", False, 1),
|
|
(["skills", "update"], "do_update", None, 0),
|
|
(["skills", "uninstall", "x", "--yes"], "do_uninstall", False, 1),
|
|
(["skills", "uninstall", "x", "--yes"], "do_uninstall", True, 0),
|
|
(["skills", "snapshot", "import", "snap.json"], "do_snapshot_import", False, 1),
|
|
(["skills", "snapshot", "import", "snap.json"], "do_snapshot_import", None, 0),
|
|
# The real scan gate, not a stubbed outcome: a refused community skill installs nothing and exits 1.
|
|
(["skills", "install", "org/repo/risky-skill", "--yes"], "scan_gate", "dangerous", 1),
|
|
(["skills", "install", "org/repo/risky-skill", "--yes"], "scan_gate", "caution", 1),
|
|
])
|
|
def test_cli_exit_code_follows_the_action_outcome(monkeypatch, tmp_path, capsys, argv, target, outcome, code):
|
|
from hermes_cli.main import main
|
|
|
|
if target == "scan_gate":
|
|
installs, audit = _scan_gate_env(monkeypatch, tmp_path, verdict=outcome)
|
|
else:
|
|
monkeypatch.setattr(cli_hub, target, lambda *a, **k: outcome)
|
|
if argv[1] == "update":
|
|
# Never run hermes_cli.main with `update` in argv (it can turn into a real self-update): drive
|
|
# the same router `hermes skills` returns its exit code from.
|
|
from argparse import Namespace
|
|
exit_code = cli_hub.skills_command(Namespace(skills_action="update", name=None, force=False)) or 0
|
|
else:
|
|
monkeypatch.setattr(sys, "argv", ["hermes", *argv])
|
|
try:
|
|
main()
|
|
exit_code = 0
|
|
except SystemExit as exc:
|
|
exit_code = exc.code or 0
|
|
assert exit_code == code
|
|
if target == "scan_gate":
|
|
assert installs == [] and "Not installed:" in capsys.readouterr().out
|
|
assert audit and audit[0][0] == "BLOCKED"
|
|
|
|
|
|
@pytest.mark.parametrize("verdict", ["caution", "dangerous"])
|
|
def test_update_enforces_the_scan_verdict_it_prints(monkeypatch, tmp_path, capsys, verdict):
|
|
"""#38041: `skills update` printed "Decision: BLOCKED" yet re-ran the gate with force=True and
|
|
installed a caution-verdict community update. The update must honour the displayed decision."""
|
|
import tools.skills_hub_install as hub_install
|
|
installs, audit = _scan_gate_env(monkeypatch, tmp_path, verdict=verdict,
|
|
installed={"install_path": "risky-skill"})
|
|
monkeypatch.setattr(hub_install, "check_for_skill_updates", lambda **_kw: [{
|
|
"name": "risky-skill", "identifier": "org/repo/risky-skill", "source": "skills-sh",
|
|
"status": "update_available"}])
|
|
|
|
assert cli_hub.do_update() is False
|
|
assert installs == []
|
|
out = capsys.readouterr().out
|
|
assert "Not installed:" in out and "Updated 1 skill" not in out
|
|
assert audit and audit[0][0] == "BLOCKED"
|
|
|
|
|
|
def test_update_still_replaces_an_installed_skill_the_scan_allows(monkeypatch, tmp_path):
|
|
import tools.skills_hub_install as hub_install
|
|
installs, _audit = _scan_gate_env(monkeypatch, tmp_path, verdict="safe",
|
|
installed={"install_path": "risky-skill"})
|
|
monkeypatch.setattr(hub_install, "check_for_skill_updates", lambda **_kw: [{
|
|
"name": "risky-skill", "identifier": "org/repo/risky-skill", "source": "skills-sh",
|
|
"status": "update_available"}])
|
|
|
|
assert cli_hub.do_update() is None
|
|
assert installs == ["risky-skill"]
|