mirror of
https://github.com/usestrix/strix.git
synced 2026-10-12 03:47:56 +00:00
Delete: only the filing agent may withdraw, rewrite artifacts before persistence with rollback, dedicated viewer renderer
This commit is contained in:
parent
1ac6fe740c
commit
7baa127e93
8 changed files with 315 additions and 89 deletions
|
|
@ -0,0 +1,46 @@
|
|||
"use client";
|
||||
|
||||
import type { ToolRendererProps } from "@/types/events";
|
||||
import { TruncatedText } from "./ToolCard";
|
||||
|
||||
const SEVERITY_COLORS: Record<string, string> = {
|
||||
critical: "text-red-400", high: "text-orange-400", medium: "text-yellow-400",
|
||||
low: "text-blue-400", info: "text-cyan-400",
|
||||
};
|
||||
|
||||
/** A withdrawn finding: which report went, and what disproved it. */
|
||||
export default function VulnReportDeleteRenderer({ args, result, status }: ToolRendererProps) {
|
||||
const reportId = (args.report_id as string) ?? "";
|
||||
const reason = (args.delete_reason as string) ?? "";
|
||||
|
||||
const res = result && typeof result === "object" ? (result as Record<string, unknown>) : null;
|
||||
const ok = res?.success === true;
|
||||
const error = typeof res?.error === "string" ? res.error : "";
|
||||
const title = typeof res?.title === "string" ? res.title : "";
|
||||
const severity = typeof res?.severity === "string" ? res.severity.toLowerCase() : "";
|
||||
const failed = res != null ? !ok : status === "failed" || status === "error";
|
||||
|
||||
return (
|
||||
<div className="space-y-2">
|
||||
<div className="flex items-center gap-2 flex-wrap">
|
||||
<span className={`font-semibold text-sm ${failed ? "text-red-400/80" : "text-yellow-400/80"}`}>
|
||||
{failed ? "report not withdrawn" : "report withdrawn"}
|
||||
</span>
|
||||
{reportId && <span className="text-[#555] font-mono text-[13px]">{reportId}</span>}
|
||||
{severity && (
|
||||
<span className={`text-[13px] line-through ${SEVERITY_COLORS[severity] ?? "text-[#888]"}`}>
|
||||
{severity.toUpperCase()}
|
||||
</span>
|
||||
)}
|
||||
</div>
|
||||
{title && <div className="text-[15px] text-white/60 line-through">{title}</div>}
|
||||
{reason && (
|
||||
<div>
|
||||
<span className="text-emerald-400/60 text-sm font-semibold">Why</span>
|
||||
<div className="mt-1"><TruncatedText text={reason} maxLines={10} /></div>
|
||||
</div>
|
||||
)}
|
||||
{error && <div className="text-red-400/80 text-[13px]">{error}</div>}
|
||||
</div>
|
||||
);
|
||||
}
|
||||
|
|
@ -12,6 +12,7 @@ import FileEditRenderer from "./FileEditRenderer";
|
|||
import ApplyPatchRenderer from "./ApplyPatchRenderer";
|
||||
import ViewImageRenderer from "./ViewImageRenderer";
|
||||
import VulnReportRenderer from "./VulnReportRenderer";
|
||||
import VulnReportDeleteRenderer from "./VulnReportDeleteRenderer";
|
||||
import ReportListRenderer from "./ReportListRenderer";
|
||||
import ProxyRenderer from "./ProxyRenderer";
|
||||
import ThinkRenderer from "./ThinkRenderer";
|
||||
|
|
@ -151,6 +152,7 @@ const RENDERER_OVERRIDES: Partial<Record<string, ComponentType<ToolRendererProps
|
|||
view_image: ViewImageRenderer,
|
||||
list_reports: ReportListRenderer,
|
||||
get_report: ReportListRenderer,
|
||||
delete_vulnerability_report: VulnReportDeleteRenderer,
|
||||
};
|
||||
|
||||
/**
|
||||
|
|
|
|||
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
|
|
@ -6,8 +6,8 @@
|
|||
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
||||
<meta name="color-scheme" content="dark" />
|
||||
<title>Strix Results</title>
|
||||
<script type="module" crossorigin src="./assets/index-B94ANU8d.js"></script>
|
||||
<link rel="stylesheet" crossorigin href="./assets/index-DN__rVv3.css">
|
||||
<script type="module" crossorigin src="./assets/index-DNPiqLUu.js"></script>
|
||||
<link rel="stylesheet" crossorigin href="./assets/index-CrLE9rrV.css">
|
||||
</head>
|
||||
<body>
|
||||
<div id="root"></div>
|
||||
|
|
|
|||
|
|
@ -36,6 +36,19 @@ _global_report_state: Optional["ReportState"] = None
|
|||
_CONTROL_CHARS = re.compile(r"[\x00-\x1f\x7f]+")
|
||||
|
||||
|
||||
class ReportOwnershipError(PermissionError):
|
||||
"""Raised when an agent tries to withdraw a report another agent filed."""
|
||||
|
||||
def __init__(self, report_id: str, *, filed_by: str, caller: str | None) -> None:
|
||||
self.report_id = report_id
|
||||
self.filed_by = filed_by
|
||||
self.caller = caller
|
||||
super().__init__(
|
||||
f"Report '{report_id}' was filed by agent {filed_by}, not by "
|
||||
f"{caller or 'the caller'}; only the agent that filed a report can withdraw it"
|
||||
)
|
||||
|
||||
|
||||
def _strix_version() -> str | None:
|
||||
"""Best-effort package version for the SARIF tool.driver.version field."""
|
||||
try:
|
||||
|
|
@ -545,15 +558,27 @@ class ReportState:
|
|||
) -> dict[str, Any] | None:
|
||||
"""Withdraw a report from the run, keeping a record of the withdrawal.
|
||||
|
||||
Returns the removed report, or ``None`` when the id is unknown. The
|
||||
report leaves ``vulnerability_reports`` and its rendered artifacts; the
|
||||
run record keeps who withdrew it and why, so the id is never reissued.
|
||||
Returns the removed report, or ``None`` when the id is unknown. Only the
|
||||
agent that filed a report may withdraw it (:class:`ReportOwnershipError`
|
||||
otherwise). The report leaves ``vulnerability_reports`` and its rendered
|
||||
artifacts; the run record keeps who withdrew it and why, so the id is
|
||||
never reissued.
|
||||
|
||||
The rewritten artifacts and the ``vulnerability_deleted_callback`` must
|
||||
both accept the deletion first: if either fails the report is put back
|
||||
and the error propagates, so the deletion can be retried.
|
||||
"""
|
||||
report = next((r for r in self.vulnerability_reports if r.get("id") == report_id), None)
|
||||
if report is None:
|
||||
logger.warning("cannot delete unknown vulnerability report %s", report_id)
|
||||
return None
|
||||
|
||||
filed_by = report.get("agent_id")
|
||||
if filed_by and deleted_by_agent_id != filed_by:
|
||||
raise ReportOwnershipError(
|
||||
report_id, filed_by=str(filed_by), caller=deleted_by_agent_id
|
||||
)
|
||||
|
||||
entry: dict[str, Any] = {
|
||||
"id": report_id,
|
||||
"title": report.get("title"),
|
||||
|
|
@ -569,17 +594,30 @@ class ReportState:
|
|||
if deleted_by_agent_name:
|
||||
entry["agent_name"] = deleted_by_agent_name
|
||||
|
||||
# Persistence must accept the deletion before local state changes. A
|
||||
# failed callback leaves the report on file and the deletion retryable.
|
||||
if self.vulnerability_deleted_callback:
|
||||
self.vulnerability_deleted_callback({**report, "deletion": entry})
|
||||
position = self.vulnerability_reports.index(report)
|
||||
was_saved = report_id in self._saved_vuln_ids
|
||||
history = self._deleted_vulnerability_reports()
|
||||
|
||||
self.vulnerability_reports.remove(report)
|
||||
self._saved_vuln_ids.discard(report_id)
|
||||
self.run_record["deleted_vulnerability_reports"] = [
|
||||
*self._deleted_vulnerability_reports(),
|
||||
entry,
|
||||
]
|
||||
self.run_record["deleted_vulnerability_reports"] = [*history, entry]
|
||||
try:
|
||||
# Local artifacts first: they can be put back if persistence then
|
||||
# refuses, whereas a row deleted elsewhere cannot.
|
||||
self._sync_llm_usage_record()
|
||||
self._write_artifacts()
|
||||
if self.vulnerability_deleted_callback:
|
||||
self.vulnerability_deleted_callback({**report, "deletion": entry})
|
||||
except Exception:
|
||||
self.vulnerability_reports.insert(position, report)
|
||||
if was_saved:
|
||||
self._saved_vuln_ids.add(report_id)
|
||||
if history:
|
||||
self.run_record["deleted_vulnerability_reports"] = history
|
||||
else:
|
||||
self.run_record.pop("deleted_vulnerability_reports", None)
|
||||
self._save_artifacts()
|
||||
raise
|
||||
|
||||
md_path = self.get_run_dir() / "vulnerabilities" / f"{report_id}.md"
|
||||
try:
|
||||
|
|
@ -588,8 +626,6 @@ class ReportState:
|
|||
logger.exception("could not remove %s", md_path)
|
||||
|
||||
logger.info("Deleted vulnerability report %s - %s", report_id, report.get("title"))
|
||||
|
||||
self.save_run_data()
|
||||
return report
|
||||
|
||||
def get_existing_vulnerabilities(self) -> list[dict[str, Any]]:
|
||||
|
|
@ -775,49 +811,54 @@ class ReportState:
|
|||
return None
|
||||
|
||||
def _save_artifacts(self) -> None:
|
||||
"""Write scan artifacts under ``run_dir``."""
|
||||
run_dir = self.get_run_dir()
|
||||
"""Write scan artifacts under ``run_dir``; a write failure is logged."""
|
||||
try:
|
||||
run_dir.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
coverage = self._coverage_document()
|
||||
if coverage is not None:
|
||||
try:
|
||||
write_coverage(run_dir, coverage)
|
||||
except OSError:
|
||||
logger.exception("coverage.json write failed (non-fatal)")
|
||||
|
||||
if self.final_scan_result:
|
||||
write_executive_report(run_dir, self.final_scan_result)
|
||||
|
||||
# An index is written for an empty list too once a report was
|
||||
# deleted, or the CSV/JSON on disk would still list it.
|
||||
if self.vulnerability_reports or self._deleted_vulnerability_reports():
|
||||
write_vulnerabilities(run_dir, self.vulnerability_reports, self._saved_vuln_ids)
|
||||
|
||||
# SARIF 2.1.0 emitter for CI / ASPM integration. Always emit (even
|
||||
# empty) so a clean run overwrites a prior findings.sarif rather than
|
||||
# leaving a stale one — codeql-action's "absent from new submission →
|
||||
# fixed" needs the fresh empty doc to auto-resolve alerts. Isolated
|
||||
# in its own try: a SARIF-build error must NEVER break the CSV/MD/
|
||||
# run-record path (the emitter's own contract).
|
||||
try:
|
||||
write_sarif(
|
||||
run_dir,
|
||||
self.vulnerability_reports,
|
||||
tool_version=_strix_version(),
|
||||
repository_context=self._sarif_repository_context(),
|
||||
coverage=coverage,
|
||||
)
|
||||
except Exception:
|
||||
logger.exception("SARIF emit failed (non-fatal; CSV/MD unaffected)")
|
||||
|
||||
write_run_record(run_dir, self.run_record)
|
||||
|
||||
logger.info("Essential scan data saved to: %s", run_dir)
|
||||
self._write_artifacts()
|
||||
except (OSError, RuntimeError):
|
||||
logger.exception("Failed to save scan data")
|
||||
|
||||
def _write_artifacts(self) -> None:
|
||||
"""Write scan artifacts under ``run_dir``, raising when the index or
|
||||
run record cannot be written."""
|
||||
run_dir = self.get_run_dir()
|
||||
run_dir.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
coverage = self._coverage_document()
|
||||
if coverage is not None:
|
||||
try:
|
||||
write_coverage(run_dir, coverage)
|
||||
except OSError:
|
||||
logger.exception("coverage.json write failed (non-fatal)")
|
||||
|
||||
if self.final_scan_result:
|
||||
write_executive_report(run_dir, self.final_scan_result)
|
||||
|
||||
# An index is written for an empty list too once a report was
|
||||
# deleted, or the CSV/JSON on disk would still list it.
|
||||
if self.vulnerability_reports or self._deleted_vulnerability_reports():
|
||||
write_vulnerabilities(run_dir, self.vulnerability_reports, self._saved_vuln_ids)
|
||||
|
||||
# SARIF 2.1.0 emitter for CI / ASPM integration. Always emit (even
|
||||
# empty) so a clean run overwrites a prior findings.sarif rather than
|
||||
# leaving a stale one — codeql-action's "absent from new submission →
|
||||
# fixed" needs the fresh empty doc to auto-resolve alerts. Isolated
|
||||
# in its own try: a SARIF-build error must NEVER break the CSV/MD/
|
||||
# run-record path (the emitter's own contract).
|
||||
try:
|
||||
write_sarif(
|
||||
run_dir,
|
||||
self.vulnerability_reports,
|
||||
tool_version=_strix_version(),
|
||||
repository_context=self._sarif_repository_context(),
|
||||
coverage=coverage,
|
||||
)
|
||||
except Exception:
|
||||
logger.exception("SARIF emit failed (non-fatal; CSV/MD unaffected)")
|
||||
|
||||
write_run_record(run_dir, self.run_record)
|
||||
|
||||
logger.info("Essential scan data saved to: %s", run_dir)
|
||||
|
||||
def _sarif_repository_context(self) -> dict[str, Any] | None:
|
||||
"""Repo/commit/branch context for SARIF provenance (repo scans only).
|
||||
|
||||
|
|
|
|||
|
|
@ -720,7 +720,7 @@ def _do_delete(
|
|||
),
|
||||
}
|
||||
|
||||
from strix.report.state import get_global_report_state
|
||||
from strix.report.state import ReportOwnershipError, get_global_report_state
|
||||
|
||||
report_state = get_global_report_state()
|
||||
if report_state is None:
|
||||
|
|
@ -736,6 +736,15 @@ def _do_delete(
|
|||
deleted_by_agent_id=agent_id,
|
||||
deleted_by_agent_name=agent_name,
|
||||
)
|
||||
except ReportOwnershipError as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": (
|
||||
f"{e!s}. If you have disproved another agent's finding, send that agent "
|
||||
"your counterevidence so it can withdraw the report itself."
|
||||
),
|
||||
"report_id": report_id,
|
||||
}
|
||||
except Exception as e:
|
||||
logger.exception("delete_vulnerability_report persistence failed")
|
||||
return {
|
||||
|
|
@ -1645,7 +1654,8 @@ async def delete_vulnerability_report(
|
|||
impact and the counterevidence come down together. Do NOT rewrite a
|
||||
report into a "retracted" or "false positive" note either: delete it.
|
||||
|
||||
Only withdraw a report you can name and have re-tested yourself. Call
|
||||
You can only withdraw a report you filed; a report filed by another
|
||||
agent is refused, so send that agent your counterevidence instead. Call
|
||||
``get_report`` first to read what it claims, then state in
|
||||
``delete_reason`` exactly what disproved it, so the scan history shows
|
||||
why the finding is gone. A withdrawn report cannot be restored; if new
|
||||
|
|
|
|||
|
|
@ -13,7 +13,7 @@ from strix.report.dedupe import (
|
|||
_prepare_report_for_comparison,
|
||||
check_duplicate,
|
||||
)
|
||||
from strix.report.state import ReportState, set_global_report_state
|
||||
from strix.report.state import ReportOwnershipError, ReportState, set_global_report_state
|
||||
from strix.tools.finish.tool import finish_scan
|
||||
from strix.tools.reporting import tool as reporting_tool
|
||||
from strix.tools.reporting.tool import (
|
||||
|
|
@ -2003,8 +2003,8 @@ def test_delete_withdraws_a_disproved_report_and_its_artifacts(
|
|||
result = _do_delete(
|
||||
report_id="vuln-0009",
|
||||
delete_reason="The cross-tenant read came from the harness reusing the victim's session.",
|
||||
agent_id="834f79fb",
|
||||
agent_name="Validation Agent",
|
||||
agent_id="aaaa1111",
|
||||
agent_name="Recon Agent",
|
||||
)
|
||||
|
||||
assert result["success"] is True
|
||||
|
|
@ -2020,8 +2020,8 @@ def test_delete_withdraws_a_disproved_report_and_its_artifacts(
|
|||
assert persisted[0]["id"] == "vuln-0009"
|
||||
assert persisted[0]["agent_id"] == "aaaa1111", "the callback sees the finding as filed"
|
||||
deletion = persisted[0]["deletion"]
|
||||
assert deletion["agent_id"] == "834f79fb"
|
||||
assert deletion["agent_name"] == "Validation Agent"
|
||||
assert deletion["agent_id"] == "aaaa1111"
|
||||
assert deletion["agent_name"] == "Recon Agent"
|
||||
assert deletion["filed_by_agent_id"] == "aaaa1111"
|
||||
assert deletion["reason"].startswith("The cross-tenant read")
|
||||
|
||||
|
|
@ -2035,7 +2035,9 @@ def test_delete_withdraws_a_disproved_report_and_its_artifacts(
|
|||
def test_delete_never_reissues_the_withdrawn_id(report_state: ReportState) -> None:
|
||||
"""The next finding must not inherit a deleted report's id, file or history."""
|
||||
_seed_saved_report(report_state)
|
||||
assert report_state.delete_vulnerability_report("vuln-0009", delete_reason="Disproved.")
|
||||
assert report_state.delete_vulnerability_report(
|
||||
"vuln-0009", delete_reason="Disproved.", deleted_by_agent_id="aaaa1111"
|
||||
)
|
||||
|
||||
new_id = report_state.add_vulnerability_report(title="A real finding", severity="high")
|
||||
|
||||
|
|
@ -2043,26 +2045,151 @@ def test_delete_never_reissues_the_withdrawn_id(report_state: ReportState) -> No
|
|||
assert [r["id"] for r in report_state.vulnerability_reports] == ["vuln-0010"]
|
||||
|
||||
|
||||
def _assert_report_still_filed(
|
||||
report_state: ReportState, run_dir: Path, original: dict[str, Any]
|
||||
) -> None:
|
||||
"""The report is in memory and in every artifact exactly as it was filed."""
|
||||
assert report_state.vulnerability_reports == [original]
|
||||
assert "vuln-0009" in report_state._saved_vuln_ids
|
||||
assert (run_dir / "vulnerabilities" / "vuln-0009.md").exists()
|
||||
indexed = json.loads((run_dir / "vulnerabilities.json").read_text(encoding="utf-8"))
|
||||
assert [r["id"] for r in indexed] == ["vuln-0009"]
|
||||
assert "vuln-0009" in (run_dir / "vulnerabilities.csv").read_text(encoding="utf-8")
|
||||
run = json.loads((run_dir / "run.json").read_text(encoding="utf-8"))
|
||||
assert "deleted_vulnerability_reports" not in run
|
||||
assert "deleted_vulnerability_reports" not in report_state.run_record
|
||||
assert report_state._next_report_id() == "vuln-0010"
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
("agent_id", "agent_name"),
|
||||
[("834f79fb", "Validation Agent"), (None, None)],
|
||||
ids=["another agent", "no caller identity"],
|
||||
)
|
||||
def test_delete_refuses_a_report_the_caller_did_not_file(
|
||||
report_state: ReportState, agent_id: str | None, agent_name: str | None
|
||||
) -> None:
|
||||
"""Report ids are sequential and listed to every agent; knowing one is not owning it."""
|
||||
run_dir = _seed_saved_report(report_state)
|
||||
original = dict(report_state.vulnerability_reports[0])
|
||||
calls: list[dict[str, Any]] = []
|
||||
report_state.vulnerability_deleted_callback = calls.append
|
||||
|
||||
result = _do_delete(
|
||||
report_id="vuln-0009",
|
||||
delete_reason="Disproved.",
|
||||
agent_id=agent_id,
|
||||
agent_name=agent_name,
|
||||
)
|
||||
|
||||
assert result["success"] is False
|
||||
assert "filed by agent aaaa1111" in result["error"]
|
||||
assert "send that agent your counterevidence" in result["error"]
|
||||
assert calls == [], "a refused deletion never reaches persistence"
|
||||
_assert_report_still_filed(report_state, run_dir, original)
|
||||
|
||||
|
||||
def test_delete_refuses_a_foreign_report_at_the_state_layer(report_state: ReportState) -> None:
|
||||
_seed_saved_report(report_state)
|
||||
|
||||
with pytest.raises(ReportOwnershipError) as excinfo:
|
||||
report_state.delete_vulnerability_report(
|
||||
"vuln-0009", delete_reason="Disproved.", deleted_by_agent_id="834f79fb"
|
||||
)
|
||||
|
||||
assert excinfo.value.filed_by == "aaaa1111"
|
||||
assert excinfo.value.caller == "834f79fb"
|
||||
assert len(report_state.vulnerability_reports) == 1
|
||||
|
||||
|
||||
def test_delete_allows_a_report_filed_without_an_agent(report_state: ReportState) -> None:
|
||||
"""A report with no filer on record (host-side or legacy) has no owner to check."""
|
||||
_seed_saved_report(report_state)
|
||||
del report_state.vulnerability_reports[0]["agent_id"]
|
||||
|
||||
deleted = report_state.delete_vulnerability_report(
|
||||
"vuln-0009", delete_reason="Disproved.", deleted_by_agent_id="834f79fb"
|
||||
)
|
||||
|
||||
assert deleted is not None
|
||||
assert report_state.vulnerability_reports == []
|
||||
history = report_state.run_record["deleted_vulnerability_reports"]
|
||||
assert history[0]["agent_id"] == "834f79fb"
|
||||
assert "filed_by_agent_id" not in history[0]
|
||||
|
||||
|
||||
def test_delete_reports_persistence_failure_and_keeps_the_report(
|
||||
report_state: ReportState,
|
||||
) -> None:
|
||||
"""Persistence refused after the local indexes were rewritten: they are rewritten back."""
|
||||
run_dir = _seed_saved_report(report_state)
|
||||
original = dict(report_state.vulnerability_reports[0])
|
||||
|
||||
def fail_persistence(_report: dict[str, Any]) -> None:
|
||||
assert report_state.vulnerability_reports == [], "indexes are rewritten first"
|
||||
raise RuntimeError("persistence failed")
|
||||
|
||||
report_state.vulnerability_deleted_callback = fail_persistence
|
||||
|
||||
result = _do_delete(report_id="vuln-0009", delete_reason="Disproved.")
|
||||
result = _do_delete(report_id="vuln-0009", delete_reason="Disproved.", agent_id="aaaa1111")
|
||||
|
||||
assert result["success"] is False
|
||||
assert "persistence failed" in result["error"]
|
||||
assert "still on file" in result["error"]
|
||||
assert report_state.vulnerability_reports == [original]
|
||||
assert (run_dir / "vulnerabilities" / "vuln-0009.md").exists()
|
||||
assert "deleted_vulnerability_reports" not in report_state.run_record
|
||||
assert report_state._next_report_id() == "vuln-0010"
|
||||
_assert_report_still_filed(report_state, run_dir, original)
|
||||
|
||||
|
||||
def test_delete_reports_an_artifact_write_failure_and_keeps_the_report(
|
||||
report_state: ReportState, monkeypatch: pytest.MonkeyPatch
|
||||
) -> None:
|
||||
"""A disk that cannot take the rewritten run record is a failed deletion, not a
|
||||
success with stale artifacts; persistence is never told about it."""
|
||||
run_dir = _seed_saved_report(report_state)
|
||||
original = dict(report_state.vulnerability_reports[0])
|
||||
calls: list[dict[str, Any]] = []
|
||||
report_state.vulnerability_deleted_callback = calls.append
|
||||
attempts = 0
|
||||
|
||||
def write_run_record_once_failing(run_dir_: Path, record: dict[str, Any]) -> None:
|
||||
nonlocal attempts
|
||||
attempts += 1
|
||||
if attempts == 1:
|
||||
raise OSError(28, "No space left on device")
|
||||
(run_dir_ / "run.json").write_text(json.dumps(record), encoding="utf-8")
|
||||
|
||||
monkeypatch.setattr("strix.report.state.write_run_record", write_run_record_once_failing)
|
||||
|
||||
result = _do_delete(report_id="vuln-0009", delete_reason="Disproved.", agent_id="aaaa1111")
|
||||
|
||||
assert result["success"] is False
|
||||
assert "No space left on device" in result["error"]
|
||||
assert "still on file" in result["error"]
|
||||
assert calls == []
|
||||
assert attempts == 2, "the run record is written back after the failed rewrite"
|
||||
_assert_report_still_filed(report_state, run_dir, original)
|
||||
|
||||
|
||||
def test_delete_can_be_retried_after_persistence_recovers(report_state: ReportState) -> None:
|
||||
run_dir = _seed_saved_report(report_state)
|
||||
outcomes = iter([RuntimeError("persistence failed"), None])
|
||||
|
||||
def flaky_persistence(_report: dict[str, Any]) -> None:
|
||||
outcome = next(outcomes)
|
||||
if outcome is not None:
|
||||
raise outcome
|
||||
|
||||
report_state.vulnerability_deleted_callback = flaky_persistence
|
||||
|
||||
first = _do_delete(report_id="vuln-0009", delete_reason="Disproved.", agent_id="aaaa1111")
|
||||
second = _do_delete(report_id="vuln-0009", delete_reason="Disproved.", agent_id="aaaa1111")
|
||||
|
||||
assert first["success"] is False
|
||||
assert second["success"] is True
|
||||
assert report_state.vulnerability_reports == []
|
||||
assert not (run_dir / "vulnerabilities" / "vuln-0009.md").exists()
|
||||
assert json.loads((run_dir / "vulnerabilities.json").read_text(encoding="utf-8")) == []
|
||||
run = json.loads((run_dir / "run.json").read_text(encoding="utf-8"))
|
||||
assert [e["id"] for e in run["deleted_vulnerability_reports"]] == ["vuln-0009"]
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
|
|
@ -2083,7 +2210,7 @@ def test_delete_rejects_a_call_it_cannot_act_on(
|
|||
calls: list[dict[str, Any]] = []
|
||||
report_state.vulnerability_deleted_callback = calls.append
|
||||
|
||||
result = _do_delete(report_id=report_id, delete_reason=delete_reason)
|
||||
result = _do_delete(report_id=report_id, delete_reason=delete_reason, agent_id="aaaa1111")
|
||||
|
||||
assert result["success"] is False
|
||||
assert expected in result["error"]
|
||||
|
|
@ -2093,9 +2220,9 @@ def test_delete_rejects_a_call_it_cannot_act_on(
|
|||
|
||||
async def test_delete_tool_carries_the_caller_identity(report_state: ReportState) -> None:
|
||||
_seed_saved_report(report_state)
|
||||
coordinator = type("Coordinator", (), {"names": {"834f79fb": "Validation Agent"}})()
|
||||
coordinator = type("Coordinator", (), {"names": {"aaaa1111": "Recon Agent"}})()
|
||||
ctx = ToolContext(
|
||||
context={"agent_id": "834f79fb", "coordinator": coordinator},
|
||||
context={"agent_id": "aaaa1111", "coordinator": coordinator},
|
||||
tool_name="delete_vulnerability_report",
|
||||
tool_call_id="call-1",
|
||||
tool_arguments="{}",
|
||||
|
|
@ -2109,6 +2236,6 @@ async def test_delete_tool_carries_the_caller_identity(report_state: ReportState
|
|||
|
||||
assert result["success"] is True
|
||||
history = report_state.run_record["deleted_vulnerability_reports"]
|
||||
assert history[0]["agent_id"] == "834f79fb"
|
||||
assert history[0]["agent_name"] == "Validation Agent"
|
||||
assert history[0]["agent_id"] == "aaaa1111"
|
||||
assert history[0]["agent_name"] == "Recon Agent"
|
||||
assert "update_vulnerability_report" in delete_vulnerability_report.description
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue