Compare commits

...
Author SHA1 Message Date
BaaRai 6568a8eb44 fix(llm): record breaker health on every task LLM path
The /system/llm/ badge read llm_profile_health, which only the failover
Pool wrote. A task pinned to an explicit chain, a bound agent with
failover off, and the global provider all call the provider directly, so
a dead backend stayed "Healthy".

One policy now lives in llmpool.Registry.RecordCall, used by the Pool and
by the task runtimes: a success clears the state, a failover-eligible
failure trips it, context cancellation and deterministic 400s do not, and
a stream that already delivered output is left alone. The selection
carries the profile whose breaker the call updates, and a provider
already wrapped in the Pool reports nothing to avoid double counting.

The LLM recording page also explains an empty list while the llm_record
toggle is off, after the settings call resolves.

Originally reported in #257.
2026-11-18 01:31:18 +01:00
BaaRai 9a2a9e61ef Merge pull request #308 from Coff0xc/feat/finding-report-state
feat(findings): report stage separate from triage status
2026-11-18 00:17:48 +01:00
Coff0xc 6a0b79d34f feat(findings): add report_state separate from triage 2026-11-18 00:07:50 +01:00
BaaRai 729f935a21 Merge pull request #311 from BaaRai/fix/agent-frame-intent-summaries
fix(agent): frame intent summaries as data (issue #124)
2026-11-17 23:55:50 +01:00
27 changed files with 1170 additions and 108 deletions
+1
View File
@@ -36,6 +36,7 @@ All notable changes to this project are documented in this file. The format is b
### Fixed
- **LLM health badge on every task path**: `/system/llm/` read a breaker state that only the failover Pool wrote, so a profile used by an explicit task chain, by a bound agent with failover off, or as the global provider stayed "Healthy" through a real outage. Every task LLM attempt now records its outcome through the same policy the Pool uses, keyed by the profile that served it. The LLM recording page also explains an empty list when the `llm_record` toggle is off instead of looking broken.
- **Skill visibility survives restarts**: turning a skill off now writes an explicit disabled row instead of deleting it, so the startup seed, which only inserts missing rows, can no longer bring the skill back. Skills that were turned off before this version have no row to keep, so they come back once after the upgrade and must be turned off again.
- **Browser extension traffic capture**: CDP delivers `Network.responseReceived` response headers as an object, not an array of `{name, value}` records, so the extension threw on every response and never queued an exchange. It now accepts both shapes and lower-cases names before the body-capture gate, so captures work again and `content-type` correctly controls body capture.
- **bbot on arm64**: the image now builds on arm64 hosts such as Apple Silicon. bbot's `gowitness` downloaded an x86-64 Chrome and `retirejs` an x86-64 Node, which can't run there, so the build failed. On arm64 the image uses Debian's `chromium` for gowitness and an arm64 Node for retirejs, and the build check now covers `retirejs` too. amd64 builds are unchanged.
+102 -7
View File
@@ -31,6 +31,7 @@ type DBFinding struct {
Worker string
AssetIDs []int64
Status string
ReportState string // none / submitted / accepted / rejected / duplicate; separate from Status
PoELevel int // proof-of-exploitation level 0-4 (S1); 0 = none declared
VerificationStatus string // verified / unverified / blocked / not_assessed / false_positive
VerificationNote string // why the status was reached; empty when verified
@@ -65,6 +66,27 @@ func ValidFindingStatus(s string) bool {
return false
}
// Finding report stages (findings.report_state). A workflow stage after triage,
// not a verdict: a finding can be confirmed and submitted at the same time.
// RSI lesson extraction still reads findings.status.
const (
FindingReportNone = "none"
FindingReportSubmitted = "submitted"
FindingReportAccepted = "accepted"
FindingReportRejected = "rejected"
FindingReportDuplicate = "duplicate"
)
// ValidFindingReportState reports whether s is a known report stage.
func ValidFindingReportState(s string) bool {
switch s {
case FindingReportNone, FindingReportSubmitted, FindingReportAccepted,
FindingReportRejected, FindingReportDuplicate:
return true
}
return false
}
// Finding severity levels (findings.severity).
const (
SeverityCritical = "critical" // Critical
@@ -109,7 +131,7 @@ func (d *DB) AddFinding(taskID, nodeID int64, vulnclass, name, severity, summary
// findingSelectCols is the column list (with task_description join) every finding
// list query selects, so scanFinding stays in sync across callers.
const findingSelectCols = `f.id, f.task_id, f.node_id, f.vulnclass, COALESCE(f.name, ''), f.severity, f.summary,
f.evidence, f.worker, f.asset_ids, COALESCE(f.status, 'pending'), f.created_at,
f.evidence, f.worker, f.asset_ids, COALESCE(f.status, 'pending'), COALESCE(f.report_state, 'none'), f.created_at,
COALESCE(t.description, '') AS task_description, f.evidence_version, f.report_evidence_version,
f.poe_level, COALESCE(f.verification_status, 'unverified'), COALESCE(f.verification_note, ''),
f.counterevidence, f.confidence_rationale,
@@ -126,7 +148,7 @@ func scanFindings(rows interface {
f := &DBFinding{}
var aidsJSON string
if err := rows.Scan(&f.ID, &f.TaskID, &f.NodeID, &f.VulnClass, &f.Name, &f.Severity,
&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.CreatedAt, &f.TaskDescription, &f.EvidenceVersion, &f.ReportEvidenceVersion, &f.PoELevel, &f.VerificationStatus, &f.VerificationNote, &f.Counterevidence, &f.ConfidenceRationale, &f.TrafficCount); err != nil {
&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.ReportState, &f.CreatedAt, &f.TaskDescription, &f.EvidenceVersion, &f.ReportEvidenceVersion, &f.PoELevel, &f.VerificationStatus, &f.VerificationNote, &f.Counterevidence, &f.ConfidenceRationale, &f.TrafficCount); err != nil {
return nil, err
}
_ = json.Unmarshal([]byte(aidsJSON), &f.AssetIDs)
@@ -160,6 +182,7 @@ func (d *DB) ListFindings(limit int) ([]*DBFinding, error) {
type FindingFilter struct {
Severity string // high | medium | low
Status string // pending | false_positive | ignored | resolved
ReportState string // none | submitted | accepted | rejected | duplicate
VerificationStatus string // verified | unverified | blocked | not_assessed | false_positive
VulnClass string
TaskID string // task id (in string form; empty/invalid = do not filter by task)
@@ -197,6 +220,7 @@ func (f FindingFilter) where() (string, []any) {
}
add("severity", f.Severity)
add("status", f.Status)
add("report_state", f.ReportState)
add("verification_status", f.VerificationStatus)
add("vulnclass", f.VulnClass)
// task_id is a bigint column, compared as an integer (it cannot go through
@@ -528,7 +552,7 @@ func (d *DB) ListFindingsForExport(f FindingFilter, ids []int64) ([]*DBFinding,
f := &DBFinding{}
var aidsJSON string
if err := rows.Scan(&f.ID, &f.TaskID, &f.NodeID, &f.VulnClass, &f.Name, &f.Severity,
&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.CreatedAt,
&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.ReportState, &f.CreatedAt,
&f.TaskDescription, &f.EvidenceVersion, &f.ReportEvidenceVersion, &f.PoELevel, &f.VerificationStatus, &f.VerificationNote, &f.Counterevidence, &f.ConfidenceRationale, &f.TrafficCount, &f.Report); err != nil {
return nil, err
}
@@ -544,6 +568,8 @@ func (d *DB) ListFindingsForExport(f FindingFilter, ids []int64) ([]*DBFinding,
type FindingStats struct {
Total int `json:"total"`
Pending int `json:"pending"`
Submitted int `json:"submitted"`
Accepted int `json:"accepted"`
Critical int `json:"critical"`
High int `json:"high"`
Medium int `json:"medium"`
@@ -563,18 +589,20 @@ type FindingTaskOption struct {
Count int `json:"count"`
}
// FindingStats returns whole-table counts (by severity + pending) and the sorted
// set of distinct vuln classes.
// FindingStats returns whole-table counts (by severity, pending, and report
// stage) and the sorted set of distinct vuln classes.
func (d *DB) FindingStats() (*FindingStats, error) {
st := &FindingStats{VulnClasses: []string{}, Tasks: []FindingTaskOption{}}
err := d.QueryRow(`SELECT
COUNT(*),
COUNT(*) FILTER (WHERE status = 'pending'),
COUNT(*) FILTER (WHERE report_state = 'submitted'),
COUNT(*) FILTER (WHERE report_state = 'accepted'),
COUNT(*) FILTER (WHERE severity = 'critical'),
COUNT(*) FILTER (WHERE severity = 'high'),
COUNT(*) FILTER (WHERE severity = 'medium'),
COUNT(*) FILTER (WHERE severity = 'low')
FROM findings`).Scan(&st.Total, &st.Pending, &st.Critical, &st.High, &st.Medium, &st.Low)
FROM findings`).Scan(&st.Total, &st.Pending, &st.Submitted, &st.Accepted, &st.Critical, &st.High, &st.Medium, &st.Low)
if err != nil {
return nil, err
}
@@ -629,6 +657,8 @@ func (d *DB) FindingStats() (*FindingStats, error) {
cold := aggregate.FindingStats
st.Total += cold.Total
st.Pending += cold.Pending
st.Submitted += cold.Submitted
st.Accepted += cold.Accepted
st.Critical += cold.Critical
st.High += cold.High
st.Medium += cold.Medium
@@ -658,7 +688,7 @@ func (d *DB) GetFinding(id int64) (*DBFinding, error) {
LEFT JOIN tasks t ON f.task_id = t.id
WHERE f.id = $1`, id).Scan(
&f.ID, &f.TaskID, &f.NodeID, &f.VulnClass, &f.Name, &f.Severity,
&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.CreatedAt,
&f.Summary, &f.Evidence, &f.Worker, &aidsJSON, &f.Status, &f.ReportState, &f.CreatedAt,
&f.TaskDescription, &f.EvidenceVersion, &f.ReportEvidenceVersion, &f.PoELevel, &f.VerificationStatus, &f.VerificationNote, &f.Counterevidence, &f.ConfidenceRationale, &f.TrafficCount, &f.Report)
if err == sql.ErrNoRows {
return nil, nil
@@ -723,6 +753,71 @@ func (d *DB) SetFindingStatus(id int64, status string) (int64, error) {
return d.setFindingCol(id, "status", "status", status)
}
// SetFindingReportState updates one finding's report stage and mirrors it onto
// the originating exploration node's payload under "report_state". It does not
// register a status-change notification: report_state is a workflow stage, not
// a triage verdict. Returns rows affected.
func (d *DB) SetFindingReportState(id int64, state string) (int64, error) {
return d.setFindingCol(id, "report_state", "report_state", state)
}
// FindingReportStateBatchResult is one id's outcome in a bulk report-stage update.
type FindingReportStateBatchResult struct {
ID int64 `json:"id"`
From string `json:"from,omitempty"`
Found bool `json:"found"`
Changed bool `json:"changed"`
}
// SetFindingReportStates updates several findings to the same report stage in
// one transaction. Unknown ids come back as found=false instead of hiding the
// others; a database error rolls the whole batch back. No notification event
// is written — status-change pushes stay on findings.status.
func (d *DB) SetFindingReportStates(ctx context.Context, ids []int64, state string) ([]FindingReportStateBatchResult, error) {
out := make([]FindingReportStateBatchResult, 0, len(ids))
err := d.WithEvidenceTx(ctx, func(tx *sql.Tx) error {
for _, id := range ids {
from, found, changed, err := setFindingReportStateTx(ctx, tx, id, state)
if err != nil {
return err
}
out = append(out, FindingReportStateBatchResult{ID: id, From: from, Found: found, Changed: changed})
}
return nil
})
if err != nil {
return nil, err
}
return out, nil
}
func setFindingReportStateTx(ctx context.Context, tx *sql.Tx, id int64, state string) (from string, found bool, changed bool, err error) {
var nodeID sql.NullInt64
scanErr := tx.QueryRowContext(ctx, `SELECT COALESCE(report_state,'none'), node_id
FROM findings WHERE id=$1 FOR UPDATE`, id).Scan(&from, &nodeID)
if scanErr == sql.ErrNoRows {
return "", false, false, nil
}
if scanErr != nil {
return "", false, false, scanErr
}
found = true
if from == state {
return from, true, false, nil
}
if _, err := tx.ExecContext(ctx, `UPDATE findings SET report_state=$2 WHERE id=$1`, id, state); err != nil {
return from, true, false, err
}
if nodeID.Valid {
if _, err := tx.ExecContext(ctx, `UPDATE exploration_nodes
SET payload = jsonb_set(payload, '{report_state}', to_jsonb($1::text))
WHERE id = $2`, state, nodeID.Int64); err != nil {
return from, true, false, err
}
}
return from, true, true, nil
}
// SetFindingReportByNodeID sets the Markdown report on the standalone finding row
// whose node_id matches — report_finding returns that node id, so an agent tool
// can address the finding it just created. Returns rows affected (0 when no row).
+131
View File
@@ -13,6 +13,25 @@ import (
// TestDeleteFinding verifies that deleting a finding removes both the findings
// row and its originating exploration node (kind='finding').
func TestValidFindingReportState(t *testing.T) {
for _, s := range []string{
FindingReportNone, FindingReportSubmitted, FindingReportAccepted,
FindingReportRejected, FindingReportDuplicate,
} {
if !ValidFindingReportState(s) {
t.Fatalf("%q should be valid", s)
}
}
for _, s := range []string{"", "confirmed", "pending", "SUBMITTED", "submitted "} {
if ValidFindingReportState(s) {
t.Fatalf("%q should be invalid", s)
}
}
if ValidFindingStatus(FindingReportSubmitted) {
t.Fatal("submitted must not be a triage status")
}
}
func TestDeleteFinding(t *testing.T) {
d, err := Open(testDSN(t))
if err != nil {
@@ -122,6 +141,115 @@ func TestSetFindingStatusMirrorsNodePayload(t *testing.T) {
}
}
func TestFindingReportState(t *testing.T) {
d, err := Open(testDSN(t))
if err != nil {
t.Skipf("postgres unavailable (%v) — skipping", err)
}
defer d.Close()
tk, err := d.CreateTask("finding report state", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer d.DeleteTask(tk.ID)
es := d.Exploration(tk.ExplorationID)
nodeID, err := es.AddNode(KindFinding, map[string]any{"summary": "x", "severity": "high"}, 5, "confirmed", "worker", nil)
if err != nil {
t.Fatal(err)
}
fid, err := d.AddFinding(tk.ID, nodeID, "XSS", "reflected XSS", "high", "summary", "poc", "worker", nil)
if err != nil {
t.Fatal(err)
}
second, err := d.AddFinding(tk.ID, 0, "XSS", "other", "low", "summary", "poc", "worker", nil)
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { d.Exec(`DELETE FROM notification_events WHERE finding_id IN ($1,$2)`, fid, second) })
one, err := d.GetFinding(fid)
if err != nil || one == nil || one.ReportState != FindingReportNone {
t.Fatalf("new finding report_state: %+v err=%v", one, err)
}
n, err := d.SetFindingReportState(fid, FindingReportSubmitted)
if err != nil || n != 1 {
t.Fatalf("SetFindingReportState: n=%d err=%v", n, err)
}
one, _ = d.GetFinding(fid)
if one.ReportState != FindingReportSubmitted {
t.Fatalf("report_state=%q", one.ReportState)
}
if one.Status != FindingPending {
t.Fatalf("triage status changed: %q", one.Status)
}
var payload []byte
if err := d.QueryRow(`SELECT payload FROM exploration_nodes WHERE id=$1`, nodeID).Scan(&payload); err != nil {
t.Fatal(err)
}
var p map[string]any
if err := json.Unmarshal(payload, &p); err != nil {
t.Fatal(err)
}
if p["report_state"] != FindingReportSubmitted {
t.Fatalf("payload report_state=%v", p["report_state"])
}
listed, total, err := d.ListFindingsPage(FindingFilter{TaskID: fmt.Sprintf("%d", tk.ID), ReportState: FindingReportSubmitted}, 1, 20)
if err != nil || total != 1 || len(listed) != 1 || listed[0].ID != fid {
t.Fatalf("filter submitted: total=%d listed=%d err=%v", total, len(listed), err)
}
st, err := d.FindingStats()
if err != nil || st.Submitted < 1 {
t.Fatalf("stats submitted undercount: %+v err=%v", st, err)
}
const missingID = int64(999999999)
results, err := d.SetFindingReportStates(context.Background(), []int64{fid, second, missingID}, FindingReportAccepted)
if err != nil {
t.Fatal(err)
}
if len(results) != 3 {
t.Fatalf("want 3 results, got %d", len(results))
}
if !results[0].Found || !results[0].Changed || results[0].From != FindingReportSubmitted {
t.Fatalf("first: %+v", results[0])
}
if !results[1].Found || !results[1].Changed || results[1].From != FindingReportNone {
t.Fatalf("second: %+v", results[1])
}
if results[2].Found || results[2].Changed {
t.Fatalf("missing: %+v", results[2])
}
again, err := d.SetFindingReportStates(context.Background(), []int64{fid}, FindingReportAccepted)
if err != nil || len(again) != 1 || again[0].Changed {
t.Fatalf("noop batch: %v %v", again, err)
}
var events int
if err := d.QueryRow(`SELECT count(*) FROM notification_events WHERE finding_id IN ($1,$2)`, fid, second).Scan(&events); err != nil || events != 0 {
t.Fatalf("report_state must not write notification_events: %d (%v)", events, err)
}
one, _ = d.GetFinding(fid)
if one.Status != FindingPending {
t.Fatalf("bulk report_state mutated triage: %q", one.Status)
}
if one.ReportState != FindingReportAccepted {
t.Fatalf("bulk report_state=%q", one.ReportState)
}
st, err = d.FindingStats()
if err != nil || st.Accepted < 2 {
t.Fatalf("stats accepted undercount: %+v err=%v", st, err)
}
}
// TestFindingsPageAndStats exercises ListFindingsPage (filter/sort/paging) and
// FindingStats against the live dev PG. It tags its rows with a unique vulnclass
// so assertions are isolated from any pre-existing data, and cleans up after.
@@ -226,6 +354,9 @@ func TestFindingsPageAndStats(t *testing.T) {
if one == nil || one.ID != ids[0] || one.Severity != "critical" || one.Name != "Critical Vulnerability Title" {
t.Fatalf("GetFinding mismatch: %+v", one)
}
if one.ReportState != FindingReportNone {
t.Fatalf("new finding report_state should be none, got %q", one.ReportState)
}
if one.Report != "" {
t.Fatalf("new finding report should be empty, got %q", one.Report)
}
+2 -2
View File
@@ -137,9 +137,9 @@ func assertEmbeddedApplied(t *testing.T, d *DB) []appliedMigration {
version int
name string
}{{1, llmRecordsMigration}, {2, llmUsageMigration}, {3, "0003_task_scope_icp_index.sql"}, {4, "0004_llm_usage_ts_index.sql"},
{5, "0005_traffic_rewrite_rules.sql"}, {6, "0006_activity_outbox.sql"}}
{5, "0005_traffic_rewrite_rules.sql"}, {6, "0006_activity_outbox.sql"}, {7, "0007_finding_report_state.sql"}}
if len(applied) != len(want) {
t.Fatalf("schema_migrations = %+v, want versions 1 to 6", applied)
t.Fatalf("schema_migrations = %+v, want versions 1 to 7", applied)
}
for i, w := range want {
content, err := fs.ReadFile(migrationsFS(), w.name)
@@ -0,0 +1,5 @@
-- Report stage on findings, separate from triage status. Existing rows stay
-- 'none'. Values are validated server-side (no CHECK): none / submitted /
-- accepted / rejected / duplicate. RSI lessons still read findings.status.
ALTER TABLE findings ADD COLUMN IF NOT EXISTS report_state TEXT NOT NULL DEFAULT 'none';
CREATE INDEX IF NOT EXISTS idx_findings_report_state ON findings (report_state, created_at DESC);
+5
View File
@@ -1256,6 +1256,11 @@ ALTER TABLE findings ADD COLUMN IF NOT EXISTS verification_note TEXT NOT NULL DE
-- tested and did not work; confidence_rationale is why the conclusion holds.
ALTER TABLE findings ADD COLUMN IF NOT EXISTS counterevidence TEXT NOT NULL DEFAULT '';
ALTER TABLE findings ADD COLUMN IF NOT EXISTS confidence_rationale TEXT NOT NULL DEFAULT '';
-- Report stage after triage (none / submitted / accepted / rejected / duplicate).
-- Separate from status so a finding can be confirmed and submitted at once; RSI
-- lessons still read status. Server whitelist, no CHECK (same as status).
ALTER TABLE findings ADD COLUMN IF NOT EXISTS report_state TEXT NOT NULL DEFAULT 'none';
CREATE INDEX IF NOT EXISTS idx_findings_report_state ON findings (report_state, created_at DESC);
CREATE TABLE IF NOT EXISTS traffic_evidence_snapshots (
id TEXT PRIMARY KEY,
+2
View File
@@ -770,6 +770,8 @@ FROM skill_usage WHERE (task_id=$1 OR exploration_id=$2) AND NOT found GROUP BY
{"findings", `SELECT COALESCE(jsonb_object_agg(name,n),'{}'::jsonb) FROM (SELECT COALESCE(NULLIF(severity,''),'unknown') name,count(*) n FROM findings WHERE task_id=$1 GROUP BY severity) x`, []any{taskID}},
{"finding_stats", `SELECT jsonb_build_object(
'total',count(*),'pending',count(*) FILTER (WHERE status='pending'),
'submitted',count(*) FILTER (WHERE report_state='submitted'),
'accepted',count(*) FILTER (WHERE report_state='accepted'),
'critical',count(*) FILTER (WHERE severity='critical'),'high',count(*) FILTER (WHERE severity='high'),
'medium',count(*) FILTER (WHERE severity='medium'),'low',count(*) FILTER (WHERE severity='low'),
'vulnclasses',COALESCE(jsonb_agg(DISTINCT vulnclass) FILTER (WHERE vulnclass<>''),'[]'::jsonb))
+32 -20
View File
@@ -361,28 +361,40 @@ func rowExists(tx *sql.Tx, table string, id int64) bool {
return exists
}
func insertArchiveRows(tx *sql.Tx, table string, raw json.RawMessage) error {
// json_populate_recordset inserts NULL for absent columns, bypassing SQL
// defaults. Preserve compatibility with v3 archives predating side memory.
if (table == "side_question_sessions" || table == "side_question_requests") && len(raw) > 0 {
var rows []map[string]json.RawMessage
if err := json.Unmarshal(raw, &rows); err != nil {
return err
}
field := "memory"
if table == "side_question_requests" {
field = "context_info"
}
for _, row := range rows {
// archiveDefaults are the columns that did not exist when older archives were
// written. json_populate_recordset inserts NULL for an absent key, bypassing the
// SQL default, and a NOT NULL column then rejects the whole restore.
var archiveDefaults = map[string]map[string]string{
"side_question_sessions": {"memory": `{}`},
"side_question_requests": {"context_info": `{}`},
"findings": {"report_state": `"none"`},
}
// fillArchiveDefaults restores the default of a column an older archive
// predates, so a restore keeps working across schema additions.
func fillArchiveDefaults(table string, raw json.RawMessage) (json.RawMessage, error) {
defaults, ok := archiveDefaults[table]
if !ok || len(raw) == 0 {
return raw, nil
}
var rows []map[string]json.RawMessage
if err := json.Unmarshal(raw, &rows); err != nil {
return nil, err
}
for _, row := range rows {
for field, def := range defaults {
if len(row[field]) == 0 || string(row[field]) == "null" {
row[field] = json.RawMessage(`{}`)
row[field] = json.RawMessage(def)
}
}
var err error
raw, err = json.Marshal(rows)
if err != nil {
return err
}
}
return json.Marshal(rows)
}
func insertArchiveRows(tx *sql.Tx, table string, raw json.RawMessage) error {
raw, err := fillArchiveDefaults(table, raw)
if err != nil {
return err
}
allowed := map[string]bool{
"exploration_nodes": true, "exploration_edges": true, "exploration_anchors": true,
@@ -396,7 +408,7 @@ func insertArchiveRows(tx *sql.Tx, table string, raw json.RawMessage) error {
if rawRowCount(raw) == 0 {
return nil
}
_, err := tx.Exec(`INSERT INTO `+table+` SELECT * FROM json_populate_recordset(NULL::`+table+`,$1::json)`, string(raw))
_, err = tx.Exec(`INSERT INTO `+table+` SELECT * FROM json_populate_recordset(NULL::`+table+`,$1::json)`, string(raw))
return err
}
+31
View File
@@ -486,3 +486,34 @@ func assertArchiveGlobalStats(t *testing.T, d *DB, profileName, skillName, toolN
t.Fatalf("archive finding class %q missing", vulnclass)
}
}
// An archive written before report_state existed restores with the column's
// default instead of failing on the NOT NULL constraint.
func TestFillArchiveDefaults(t *testing.T) {
raw := json.RawMessage(`[{"id":1,"summary":"old"},{"id":2,"report_state":null},{"id":3,"report_state":"submitted"}]`)
out, err := fillArchiveDefaults("findings", raw)
if err != nil {
t.Fatal(err)
}
var rows []map[string]json.RawMessage
if err := json.Unmarshal(out, &rows); err != nil {
t.Fatal(err)
}
want := []string{`"none"`, `"none"`, `"submitted"`}
for i, row := range rows {
if string(row["report_state"]) != want[i] {
t.Fatalf("row %d report_state=%s, want %s", i, row["report_state"], want[i])
}
}
// Side memory keeps its own default, and other tables pass through.
out, err = fillArchiveDefaults("side_question_sessions", json.RawMessage(`[{"id":1}]`))
if err != nil || !strings.Contains(string(out), `"memory":{}`) {
t.Fatalf("side memory default missing: %s err=%v", out, err)
}
in := json.RawMessage(`[{"id":1}]`)
out, err = fillArchiveDefaults("activity", in)
if err != nil || string(out) != string(in) {
t.Fatalf("activity changed: %s err=%v", out, err)
}
}
+42
View File
@@ -10,6 +10,7 @@
package llmpool
import (
"context"
"sync"
"time"
)
@@ -159,6 +160,47 @@ func (r *Registry) Trip(id int64, errMsg string, hard bool) (tripped bool) {
return tripped
}
// RecordCall applies the failover health policy to one provider call. The Pool
// calls it for every member attempt, and a caller that runs a provider directly
// (a task whose selection bypasses the Pool) calls it too, so a profile's state
// does not depend on which path the request took.
//
// stream tells a streaming call from an atomic one; committed says whether any
// event already reached the caller (streaming only). The rules are the ones the
// Pool applies:
//
// - a completed call (err == nil) clears the state with Pass;
// - a failure that does not trigger failover (context cancellation, a
// deterministic 4xx) is ignored on a stream, and clears the state on an
// atomic call, matching Complete's non-failover path;
// - a failover-eligible stream failure trips only before the first event:
// after that the caller already has output and the state is left alone;
// - a failover-eligible atomic failure trips.
//
// id 0 is a no-op (no profile to key, e.g. an environment-only configuration).
// It returns whether this call opened the breaker, and the shortened error
// stored with it.
func (r *Registry) RecordCall(ctx context.Context, id int64, stream, committed bool, err error) (tripped bool, errMsg string) {
if id == 0 {
return false, ""
}
if err == nil {
r.Pass(id)
return false, ""
}
if !shouldFailover(ctx, err) {
if !stream {
r.Pass(id)
}
return false, ""
}
if stream && committed {
return false, ""
}
msg := trimErr(err)
return r.Trip(id, msg, isHardFailure(err)), msg
}
// Reset clears one profile's state — the UI's "Recover now" action.
func (r *Registry) Reset(id int64) {
r.mu.Lock()
+119
View File
@@ -0,0 +1,119 @@
package llmpool
import (
"context"
"fmt"
"testing"
"github.com/Autumn-27/norma/llm"
)
// RecordCall is the policy a provider call applies when no Pool wraps it. A
// completed call clears the state; a failover-eligible failure trips; a
// non-failover failure neither trips nor, on a stream, clears.
func TestRecordCallPolicy(t *testing.T) {
ctx := context.Background()
soft := fmt.Errorf("openai: status 500: down")
bad := fmt.Errorf("openai: status 400: bad request")
seed := func(t *testing.T) *Registry {
t.Helper()
reg := NewRegistry(nil, nil)
reg.RecordCall(ctx, 7, false, false, soft)
if st := reg.Get(7); st.Fails != 1 || st.Trips != 0 {
t.Fatalf("seed state=%+v, want one soft failure", st)
}
return reg
}
t.Run("success clears", func(t *testing.T) {
reg := seed(t)
reg.RecordCall(ctx, 7, false, false, nil)
if st := reg.Get(7); st.Fails != 0 || st.Trips != 0 || st.Open() {
t.Fatalf("state=%+v after a success, want clean", st)
}
})
t.Run("atomic non-failover clears", func(t *testing.T) {
reg := seed(t)
reg.RecordCall(ctx, 7, false, false, bad)
if st := reg.Get(7); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("state=%+v after an atomic 400, want clean (Pool.Complete policy)", st)
}
})
t.Run("stream non-failover keeps", func(t *testing.T) {
reg := seed(t)
reg.RecordCall(ctx, 7, true, false, bad)
if st := reg.Get(7); st.Fails != 1 || st.Trips != 0 {
t.Fatalf("state=%+v after a stream 400, want the failure kept", st)
}
})
t.Run("stream cancel keeps", func(t *testing.T) {
reg := seed(t)
reg.RecordCall(ctx, 7, true, false, context.Canceled)
if st := reg.Get(7); st.Fails != 1 || st.Trips != 0 {
t.Fatalf("state=%+v after a cancel, want the failure kept", st)
}
})
t.Run("committed stream failure keeps", func(t *testing.T) {
reg := seed(t)
reg.RecordCall(ctx, 7, true, true, soft)
if st := reg.Get(7); st.Fails != 1 || st.Trips != 0 {
t.Fatalf("state=%+v after a committed failure, want it ignored", st)
}
})
t.Run("pre-commit stream failure counts", func(t *testing.T) {
reg := NewRegistry(nil, nil)
reg.RecordCall(ctx, 7, true, false, soft)
if st := reg.Get(7); st.Fails != 1 || st.Trips != 0 || st.Open() {
t.Fatalf("state=%+v, want one soft failure", st)
}
})
t.Run("hard failure trips at once", func(t *testing.T) {
reg := NewRegistry(nil, nil)
reg.RecordCall(ctx, 7, false, false, fmt.Errorf("openai: status 401: bad key"))
if st := reg.Get(7); st.Trips != 1 || !st.Open() {
t.Fatalf("state=%+v, want an open breaker after one hard failure", st)
}
})
t.Run("profile 0 is a no-op", func(t *testing.T) {
reg := NewRegistry(nil, nil)
reg.RecordCall(ctx, 0, false, false, soft)
if n := len(reg.Snapshot()); n != 0 {
t.Fatalf("%d states recorded for profile 0", n)
}
})
}
// The Pool feeds the same policy: a failure before the first event counts on
// the member that failed, a failure after output reached the caller does not,
// because the Pool neither fails over nor blames the profile in that case.
func TestPoolFeedsRecordCallPolicy(t *testing.T) {
t.Run("pre-commit failure counts", func(t *testing.T) {
a := &fakeProv{name: "a", errs: []error{fmt.Errorf("status 500: down")}}
reg := NewRegistry(nil, nil)
p := New([]*Member{member(1, "a", 10, a), member(2, "b", 5, okProv("b", "full"))}, reg)
if _, err := drain(p.Stream(context.Background(), llm.CompletionRequest{})); err != nil {
t.Fatalf("stream error: %v", err)
}
if st := reg.Get(1); st.Fails != 1 || st.Trips != 0 {
t.Fatalf("state=%+v after a pre-commit failure, want one soft failure", st)
}
if st := reg.Get(2); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("state=%+v on the backup member, want clean", st)
}
})
t.Run("committed failure is ignored", func(t *testing.T) {
a := &fakeProv{
name: "a",
events: [][]llm.StreamEvent{{{Type: llm.SETextDelta, Text: "partial"}}},
errs: []error{fmt.Errorf("status 500: mid-stream drop")},
}
reg := NewRegistry(nil, nil)
p := New([]*Member{member(1, "a", 10, a), member(2, "b", 5, okProv("b", "full"))}, reg)
if _, err := drain(p.Stream(context.Background(), llm.CompletionRequest{})); err == nil {
t.Fatal("mid-stream error should surface")
}
if st := reg.Get(1); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("state=%+v after a committed failure, want clean", st)
}
})
}
+9 -10
View File
@@ -100,13 +100,12 @@ func (p *Pool) Stream(ctx context.Context, req llm.CompletionRequest) iter.Seq2[
}
}
if failed == nil {
p.health.Pass(m.ID) // completed (or failed in a non-failover way)
p.health.RecordCall(ctx, m.ID, true, false, nil) // completed: Pass
return
}
lastErr = failed
hard := isHardFailure(failed)
if p.health.Trip(m.ID, trimErr(failed), hard) {
log.Printf("[llmpool] profile %q(%s) breaker tripped: %s", m.Name, m.Model, trimErr(failed))
if tripped, msg := p.health.RecordCall(ctx, m.ID, true, false, failed); tripped {
log.Printf("[llmpool] profile %q(%s) breaker tripped: %s", m.Name, m.Model, msg)
}
if i+1 < len(order) {
n := order[i+1]
@@ -133,19 +132,19 @@ func (p *Pool) Complete(ctx context.Context, req llm.CompletionRequest) (llm.Mes
for i, m := range order {
msg, sr, usage, err := m.Prov.Complete(ctx, req)
if err == nil {
p.health.Pass(m.ID)
p.health.RecordCall(ctx, m.ID, false, false, nil) // Pass
return msg, sr, usage, nil
}
if !shouldFailover(ctx, err) {
// Non-failover error (e.g. ctx cancel, deterministic 4xx): surface as-is
// without tripping health, matching Stream's non-failover path.
p.health.Pass(m.ID)
// without tripping health. RecordCall clears the state on an atomic
// call, matching the historical Pass here.
p.health.RecordCall(ctx, m.ID, false, false, err)
return llm.Message{}, "", llm.Usage{}, err
}
lastErr = err
hard := isHardFailure(err)
if p.health.Trip(m.ID, trimErr(err), hard) {
log.Printf("[llmpool] profile %q(%s) breaker tripped: %s", m.Name, m.Model, trimErr(err))
if tripped, msg := p.health.RecordCall(ctx, m.ID, false, false, err); tripped {
log.Printf("[llmpool] profile %q(%s) breaker tripped: %s", m.Name, m.Model, msg)
}
if i+1 < len(order) {
n := order[i+1]
+4 -1
View File
@@ -70,6 +70,7 @@ func FindingsMarkdown(fs []*db.DBFinding, generatedAt time.Time) string {
fmt.Fprintf(&b, "- **Class**: %s\n", f.VulnClass)
}
fmt.Fprintf(&b, "- **Status**: %s\n", nz(f.Status, "pending"))
fmt.Fprintf(&b, "- **Report stage**: %s\n", nz(f.ReportState, "none"))
if desc := strings.TrimSpace(f.TaskDescription); desc != "" {
fmt.Fprintf(&b, "- **Task**: %s\n", desc)
}
@@ -99,6 +100,7 @@ func SingleFindingMarkdown(f *db.DBFinding, generatedAt time.Time) string {
}
fmt.Fprintf(&b, "- **Severity**: %s\n", nz(f.Severity, "info"))
fmt.Fprintf(&b, "- **Status**: %s\n", nz(f.Status, "pending"))
fmt.Fprintf(&b, "- **Report stage**: %s\n", nz(f.ReportState, "none"))
if desc := strings.TrimSpace(f.TaskDescription); desc != "" {
fmt.Fprintf(&b, "- **Task**: %s\n", desc)
}
@@ -148,7 +150,7 @@ func FindingsCSV(fs []*db.DBFinding) []byte {
var buf bytes.Buffer
buf.WriteString("\xEF\xBB\xBF") // UTF-8 BOM
w := csv.NewWriter(&buf)
_ = w.Write([]string{"ID", "Name", "Class", "Severity", "Status", "Task", "Found At", "Summary", "Traffic Evidence Count", "Traffic Evidence IDs"})
_ = w.Write([]string{"ID", "Name", "Class", "Severity", "Status", "Report Stage", "Task", "Found At", "Summary", "Traffic Evidence Count", "Traffic Evidence IDs"})
for _, f := range items {
_ = w.Write([]string{
fmt.Sprintf("%d", f.ID),
@@ -156,6 +158,7 @@ func FindingsCSV(fs []*db.DBFinding) []byte {
f.VulnClass,
nz(f.Severity, "info"),
nz(f.Status, "pending"),
nz(f.ReportState, "none"),
f.TaskDescription,
f.CreatedAt.Format("2006-01-02 15:04:05"),
strings.TrimSpace(f.Summary),
+34
View File
@@ -32,3 +32,37 @@ func TestBulkFindingStatusValidation(t *testing.T) {
})
}
}
func TestNormalizeFindingIDs(t *testing.T) {
got := normalizeFindingIDs([]int64{0, -1, 2, 2, 3, 4}, 2)
if len(got) != 2 || got[0] != 2 || got[1] != 3 {
t.Fatalf("got %v, want [2 3]", got)
}
if got := normalizeFindingIDs(nil, 500); len(got) != 0 {
t.Fatalf("empty input: %v", got)
}
}
func TestBulkFindingReportStateValidation(t *testing.T) {
s := &Server{m: &Manager{}}
cases := []struct {
name string
body string
want int
}{
{"bad json", `{`, http.StatusBadRequest},
{"unknown state", `{"ids":[1],"report_state":"confirmed"}`, http.StatusBadRequest},
{"empty ids", `{"ids":[],"report_state":"submitted"}`, http.StatusBadRequest},
{"non-positive ids", `{"ids":[0,-3],"report_state":"submitted"}`, http.StatusBadRequest},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
rec := httptest.NewRecorder()
req := httptest.NewRequest(http.MethodPost, "/api/exploration/findings/bulk-report-state", strings.NewReader(tc.body))
s.bulkPatchFindingReportState(rec, req)
if rec.Code != tc.want {
t.Fatalf("status %d, want %d: %s", rec.Code, tc.want, rec.Body.String())
}
})
}
}
+23 -17
View File
@@ -327,15 +327,16 @@ type FindingDTO struct {
Counterevidence string `json:"counterevidence,omitempty"` // what was tested and did not work; optional
ConfidenceRationale string `json:"confidence_rationale,omitempty"` // why the conclusion holds; optional
ID string `json:"id"`
FindingID string `json:"finding_id,omitempty"` // standalone findings-table id — the handle for status updates
VulnClass string `json:"vulnclass"`
Name string `json:"name,omitempty"` // vulnerability name; the frontend falls back to displaying vulnclass when empty
Severity string `json:"severity"` // critical | high | medium | low
Status string `json:"status"` // pending | in_progress | confirmed | resolved | fixed | false_positive | ignored | duplicate | risk_accepted
Summary string `json:"summary"`
Evidence string `json:"evidence"`
Report string `json:"report,omitempty"` // detailed report (Markdown); returned only by the detail endpoint, empty in lists
ID string `json:"id"`
FindingID string `json:"finding_id,omitempty"` // standalone findings-table id — the handle for status updates
VulnClass string `json:"vulnclass"`
Name string `json:"name,omitempty"` // vulnerability name; the frontend falls back to displaying vulnclass when empty
Severity string `json:"severity"` // critical | high | medium | low
Status string `json:"status"` // pending | in_progress | confirmed | resolved | fixed | false_positive | ignored | duplicate | risk_accepted
ReportState string `json:"report_state"` // none | submitted | accepted | rejected | duplicate; workflow stage, not a triage verdict
Summary string `json:"summary"`
Evidence string `json:"evidence"`
Report string `json:"report,omitempty"` // detailed report (Markdown); returned only by the detail endpoint, empty in lists
IntentID string `json:"intent_id,omitempty"`
ParamID string `json:"param_id,omitempty"`
@@ -461,6 +462,10 @@ func findingFromDB(f *db.DBFinding, assets map[int64]*db.Asset) FindingDTO {
if status == "" {
status = db.FindingPending
}
reportState := f.ReportState
if reportState == "" {
reportState = db.FindingReportNone
}
d := FindingDTO{
ID: i64s(f.ID),
FindingID: i64s(f.ID),
@@ -468,14 +473,15 @@ func findingFromDB(f *db.DBFinding, assets map[int64]*db.Asset) FindingDTO {
ReportStale: f.Report != "" && f.EvidenceVersion != f.ReportEvidenceVersion, TrafficBindings: f.TrafficBindings,
PoELevel: f.PoELevel, VerificationStatus: f.VerificationStatus, VerificationNote: f.VerificationNote,
Counterevidence: f.Counterevidence, ConfidenceRationale: f.ConfidenceRationale,
VulnClass: f.VulnClass,
Name: f.Name,
Severity: f.Severity,
Status: status,
Summary: f.Summary,
Evidence: f.Evidence,
Report: f.Report,
TS: rfc3339(f.CreatedAt),
VulnClass: f.VulnClass,
Name: f.Name,
Severity: f.Severity,
Status: status,
ReportState: reportState,
Summary: f.Summary,
Evidence: f.Evidence,
Report: f.Report,
TS: rfc3339(f.CreatedAt),
}
d.Assets = findingAssetDTOs(f.AssetIDs, assets)
if f.TaskID != nil {
+11
View File
@@ -26,6 +26,17 @@ func TestFindingDTOReasoningFields(t *testing.T) {
}
}
func TestFindingDTOReportState(t *testing.T) {
raw, _ := json.Marshal(findingFromDB(&db.DBFinding{ID: 1}, nil))
if !strings.Contains(string(raw), `"report_state":"none"`) {
t.Fatalf("empty report_state should serialize as none: %s", raw)
}
d := findingFromDB(&db.DBFinding{ID: 2, ReportState: db.FindingReportSubmitted}, nil)
if d.ReportState != db.FindingReportSubmitted {
t.Fatalf("report_state lost: %q", d.ReportState)
}
}
// PG: the detail endpoint returns the stored fields, and the report_finding
// schema upgrade adds both properties to an existing (customized) row without
// overwriting it.
+1
View File
@@ -31,6 +31,7 @@ func findingFilterFromQuery(q url.Values) db.FindingFilter {
return db.FindingFilter{
Severity: normFilter(q.Get("severity")),
Status: normFilter(q.Get("status")),
ReportState: normFilter(q.Get("report_state")),
VerificationStatus: normFilter(q.Get("verification_status")),
VulnClass: normFilter(q.Get("vulnclass")),
// task_id (independent of the task param, which switches to the per-task-node branch): filters the global table by task.
+89 -21
View File
@@ -80,8 +80,12 @@ type Server struct {
llmProv llm.Provider
llmDirect llm.Provider // concrete global provider, before any failover pool
llmCfg agent.Config // current LLM config (key not exposed)
llmOn bool
llmProf string // active LLM profile name (for llmrec tagging)
// llmHealthProfileID is the llm_profiles row llmProv was built from, or 0
// when it came from environment settings. Task selections use it to key
// breaker health for the global fallback path.
llmHealthProfileID int64
llmOn bool
llmProf string // active LLM profile name (for llmrec tagging)
// chatBusy guards the per-task main-agent run: the chat handler launches the
// agent on the server's background ctx (not the request ctx) and returns
@@ -504,11 +508,13 @@ func maxTokensResolver(cfg agent.Config) func() int {
// the running engine as the GLOBAL active pair. Safe to call at runtime (UI configures LLM).
func (s *Server) applyLLM(cfg agent.Config) error {
var prov llm.Provider
var healthID int64
// A persisted active profile must use the same cached provider as task chains
// and Agent bindings, otherwise each path owns a separate rate limiter.
if active, _ := s.m.pg.ActiveProfile(); active != nil {
if activeCfg, ok := s.loadProfileConfig(active.ID); ok && activeCfg == cfg {
prov, _, _ = s.providerForProfile(active.ID)
healthID = active.ID
}
}
if prov == nil {
@@ -555,6 +561,7 @@ func (s *Server) applyLLM(cfg agent.Config) error {
s.chatAgent.SetNoaEnabled(s.m.NoaCompactionEnabled) // experimental: noa context compaction (read per run)
s.llmProv = prov
s.llmCfg = cfg
s.llmHealthProfileID = healthID
s.llmOn = true
s.cfgMu.Unlock()
s.invalidateTaskAgents()
@@ -848,6 +855,7 @@ func (s *Server) Handler() http.Handler {
mux.HandleFunc("POST /api/exploration/findings/{id}/retests", s.startFindingRetest)
mux.HandleFunc("PATCH /api/exploration/findings/{id}", s.patchFinding)
mux.HandleFunc("POST /api/exploration/findings/bulk-status", s.bulkPatchFindingStatus)
mux.HandleFunc("POST /api/exploration/findings/bulk-report-state", s.bulkPatchFindingReportState)
mux.HandleFunc("DELETE /api/exploration/findings/{id}", s.deleteFinding)
mux.HandleFunc("GET /api/exploration/intents", s.intents)
mux.HandleFunc("GET /api/exploration/graph", s.explorationGraph)
@@ -2364,7 +2372,7 @@ func (s *Server) getFinding(w http.ResponseWriter, r *http.Request) {
// format = md-single (one combined .md) | md-zip (one .md per vulnerability, zipped)
// | csv | json
// ids = comma-separated finding ids (required when scope=selected)
// Filter params severity/status/vulnclass/task_id/q/sort match the list endpoint (used when scope=filtered).
// Filter params severity/status/report_state/vulnclass/task_id/q/sort match the list endpoint (used when scope=filtered).
func (s *Server) findingsExport(w http.ResponseWriter, r *http.Request) {
q := r.URL.Query()
scope := q.Get("scope")
@@ -2529,17 +2537,18 @@ func (s *Server) patchFinding(w http.ResponseWriter, r *http.Request) {
return
}
var body struct {
Status *string `json:"status"`
Severity *string `json:"severity"`
Name *string `json:"name"`
VulnClass *string `json:"vulnclass"`
Status *string `json:"status"`
ReportState *string `json:"report_state"`
Severity *string `json:"severity"`
Name *string `json:"name"`
VulnClass *string `json:"vulnclass"`
}
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
writeErr(w, 400, "bad json: "+err.Error())
return
}
if body.Status == nil && body.Severity == nil && body.Name == nil && body.VulnClass == nil {
writeErr(w, 400, "nothing to update: provide status/severity/name/vulnclass")
if body.Status == nil && body.ReportState == nil && body.Severity == nil && body.Name == nil && body.VulnClass == nil {
writeErr(w, 400, "nothing to update: provide status/severity/name/vulnclass/report_state")
return
}
if body.Status != nil {
@@ -2563,6 +2572,21 @@ func (s *Server) patchFinding(w http.ResponseWriter, r *http.Request) {
log.Printf("[notify] status-change event not registered finding=%d %s→%s (status was updated)", id, from, *body.Status)
}
}
if body.ReportState != nil {
if !db.ValidFindingReportState(*body.ReportState) {
writeErr(w, 400, "bad report_state: "+*body.ReportState)
return
}
n, err := s.m.pg.SetFindingReportState(id, *body.ReportState)
if err != nil {
writeErr(w, 500, err.Error())
return
}
if n == 0 {
writeErr(w, 404, "finding not found")
return
}
}
if body.Severity != nil {
if !db.ValidSeverity(*body.Severity) {
writeErr(w, 400, "bad severity: "+*body.Severity)
@@ -2634,18 +2658,7 @@ func (s *Server) bulkPatchFindingStatus(w http.ResponseWriter, r *http.Request)
writeErr(w, 400, "bad status: "+body.Status)
return
}
seen := make(map[int64]bool, len(body.IDs))
ids := make([]int64, 0, len(body.IDs))
for _, id := range body.IDs {
if id <= 0 || seen[id] {
continue
}
seen[id] = true
ids = append(ids, id)
if len(ids) == maxBulkFindingStatus {
break
}
}
ids := normalizeFindingIDs(body.IDs, maxBulkFindingStatus)
if len(ids) == 0 {
writeErr(w, 400, "no finding ids")
return
@@ -2667,6 +2680,61 @@ func (s *Server) bulkPatchFindingStatus(w http.ResponseWriter, r *http.Request)
writeJSON(w, 200, map[string]any{"updated": updated, "missing": missing, "results": results})
}
// normalizeFindingIDs drops non-positive and duplicate ids, then caps the list.
func normalizeFindingIDs(ids []int64, capN int) []int64 {
seen := make(map[int64]bool, len(ids))
out := make([]int64, 0, len(ids))
for _, id := range ids {
if id <= 0 || seen[id] {
continue
}
seen[id] = true
out = append(out, id)
if capN > 0 && len(out) == capN {
break
}
}
return out
}
// bulkPatchFindingReportState sets the same report stage on a batch of findings
// in one transaction. Same cap/dedupe/per-id result as bulk status, without
// status-change notifications: report_state is a workflow stage, not a verdict.
func (s *Server) bulkPatchFindingReportState(w http.ResponseWriter, r *http.Request) {
var body struct {
IDs []int64 `json:"ids"`
ReportState string `json:"report_state"`
}
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
writeErr(w, 400, "bad json: "+err.Error())
return
}
if !db.ValidFindingReportState(body.ReportState) {
writeErr(w, 400, "bad report_state: "+body.ReportState)
return
}
ids := normalizeFindingIDs(body.IDs, maxBulkFindingStatus)
if len(ids) == 0 {
writeErr(w, 400, "no finding ids")
return
}
results, err := s.m.pg.SetFindingReportStates(r.Context(), ids, body.ReportState)
if err != nil {
writeErr(w, 500, err.Error())
return
}
updated, missing := 0, 0
for _, res := range results {
if res.Changed {
updated++
}
if !res.Found {
missing++
}
}
writeJSON(w, 200, map[string]any{"updated": updated, "missing": missing, "results": results})
}
// deleteFinding removes a finding (findings row + originating exploration node).
// id is the standalone findings-table id (DTO finding_id).
func (s *Server) deleteFinding(w http.ResponseWriter, r *http.Request) {
+72 -13
View File
@@ -15,6 +15,7 @@ import (
"github.com/Autumn-27/artex/agent"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/artex/llmpool"
"github.com/Autumn-27/artex/llmrec"
"github.com/Autumn-27/norma/llm"
"github.com/Autumn-27/norma/transcript"
@@ -137,8 +138,15 @@ func resetLLMFailStreak(taskID string, profileID int64) {
type taskLLMSelection struct {
task *Task
profileID int64
revision int64
provider llm.Provider
// healthProfileID is the llm_profiles row whose breaker state the calls on
// this selection update. It equals profileID on the task-chain path; a bound
// agent or the global provider carries its own id here while profileID stays
// 0 (no chain semantics). 0 means "do not report": either there is no
// profile (environment-only configuration) or the provider is already the
// failover Pool, which updates the registry itself.
healthProfileID int64
revision int64
provider llm.Provider
// retry holds the retry parameters resolved for the profile selected this time (profile override → global policy → built-in default).
// Same-provider safe-window retries follow it, so switching profiles switches the retry cadence.
retry agent.RetryConfig
@@ -165,6 +173,24 @@ type taskLLMStreamHooks struct {
current func(context.Context) (taskLLMSelection, error)
exhaust func(taskLLMSelection, error) (db.TaskLLMTransition, error)
transition func(taskLLMSelection, db.TaskLLMTransition, error)
// recordHealth reports one physical provider attempt to the process-wide
// breaker. It keeps /system/llm/ truthful for every selection whose provider
// is not already the failover Pool (the chain, a bound agent, the global
// profile). Optional: test literals leave it nil and the call sites skip it.
recordHealth func(ctx context.Context, profileID int64, stream, committed bool, err error)
}
// healthIDFor returns the profile whose breaker state a call on this provider
// should update. It returns 0 when there is no profile, or when the provider is
// the failover Pool, which records every member attempt itself.
func healthIDFor(prov llm.Provider, profileID int64) int64 {
if profileID == 0 {
return 0
}
if _, pooled := prov.(*llmpool.Pool); pooled {
return 0
}
return profileID
}
// current resolves the provider this role runs on, by precedence:
@@ -196,8 +222,9 @@ func (r *taskLLMRuntime) currentFor(ctx context.Context) (taskLLMSelection, erro
r.s.syncTaskLLMState(pt)
t, _ := r.s.m.Task(r.taskID)
sel := taskLLMSelection{task: t, revision: pt.LLMChainRevision}
if prov, cfg, ok := r.s.agentBindingProvider(r.agentKey); ok {
if prov, cfg, id, ok := r.s.agentBindingProvider(r.agentKey); ok {
sel.provider, sel.retry = prov, cfg.Retry
sel.healthProfileID = healthIDFor(prov, id)
return sel, nil
}
if len(pt.LLMProfileIDs) > 0 {
@@ -218,13 +245,15 @@ func (r *taskLLMRuntime) currentFor(ctx context.Context) (taskLLMSelection, erro
return sel, fmt.Errorf("LLM profile #%d is missing or invalid", sel.profileID)
}
sel.provider, sel.retry = prov, cfg.Retry
sel.healthProfileID = sel.profileID // chain selections never go through the Pool
return sel, nil
}
prov, cfg, ok := r.s.globalProvider()
prov, cfg, id, ok := r.s.globalProviderWithProfile()
if !ok {
return sel, fmt.Errorf("task %s has no available fallback LLM provider", r.taskID)
}
sel.provider, sel.retry = prov, cfg.Retry
sel.healthProfileID = healthIDFor(prov, id)
return sel, nil
}
@@ -242,7 +271,7 @@ func (r *taskLLMRuntime) activeCfg() (agent.Config, bool) {
if err != nil || pt == nil {
return agent.Config{}, false
}
if _, cfg, ok := r.s.agentBindingProvider(r.agentKey); ok {
if _, cfg, _, ok := r.s.agentBindingProvider(r.agentKey); ok {
return cfg, true
}
if len(pt.LLMProfileIDs) > 0 && pt.ActiveLLMProfileID != nil {
@@ -280,10 +309,19 @@ func parseTaskID(id string) (int64, error) {
// streamHooks builds the failover/exhaustion callbacks shared by Stream and
// Complete: how to read the current profile selection, how to mark it quota
// exhausted, and how to emit a failover transition.
// exhausted, how to emit a failover transition, and how to report breaker
// health for providers that are not the failover Pool.
func (r *taskLLMRuntime) streamHooks() taskLLMStreamHooks {
return taskLLMStreamHooks{
current: func(ctx context.Context) (taskLLMSelection, error) { return r.currentFor(ctx) },
recordHealth: func(ctx context.Context, profileID int64, stream, committed bool, err error) {
if profileID == 0 || r.s == nil || r.s.llmHealth == nil {
return
}
if tripped, msg := r.s.llmHealth.RecordCall(ctx, profileID, stream, committed, err); tripped {
log.Printf("[task-llm] profile #%d breaker tripped: %s", profileID, msg)
}
},
exhaust: func(selection taskLLMSelection, cause error) (db.TaskLLMTransition, error) {
taskNum, _ := parseTaskID(r.taskID)
transition, err := r.s.m.pg.MarkTaskLLMProfileQuotaExhaustedAtRevision(taskNum, selection.profileID, selection.revision, cause.Error())
@@ -338,6 +376,9 @@ func completeTaskLLM(ctx context.Context, taskID string, req llm.CompletionReque
for attempt := 0; ; attempt++ {
msg, sr, usage, callErr = selection.provider.Complete(ctx, req)
total.Add(usage)
if hooks.recordHealth != nil {
hooks.recordHealth(ctx, selection.healthProfileID, false, false, callErr)
}
if callErr != nil && ctx.Err() == nil &&
attempt < retries && isRetryableStreamError(callErr) {
backoff := backoffOf(attempt)
@@ -462,6 +503,9 @@ func streamTaskLLM(ctx context.Context, taskID string, req llm.CompletionRequest
return
}
}
if hooks.recordHealth != nil {
hooks.recordHealth(ctx, selection.healthProfileID, true, committed, streamErr)
}
if streamErr != nil && !committed {
// A failed uncommitted attempt delivers none of its events (they are
// never replayed), but the usage it reported survives as one
@@ -635,7 +679,7 @@ func isRetryableStreamError(err error) bool {
// CompactionWindow mirrors current()'s precedence so the context window always
// matches the provider the role will actually stream on.
func (r *taskLLMRuntime) CompactionWindow() int {
if _, cfg, ok := r.s.agentBindingProvider(r.agentKey); ok {
if _, cfg, _, ok := r.s.agentBindingProvider(r.agentKey); ok {
return cfg.CompactionWindow()
}
taskNum, err := parseTaskID(r.taskID)
@@ -671,17 +715,19 @@ func (r *taskLLMRuntime) CompactionWindow() int {
// (agents.llm_profile_id) — the highest-precedence level for task agents. ok=false
// when the role has no binding or the bound profile no longer builds, so callers
// fall through to the task chain. A bound profile stays exclusive unless
// llm_pool_bind_fallback is on, which is what poolForBinding encodes.
func (s *Server) agentBindingProvider(agentKey string) (llm.Provider, agent.Config, bool) {
// llm_pool_bind_fallback is on, which is what poolForBinding encodes. The
// returned id is the binding's profile, used to key breaker health when the
// returned provider is not the Pool.
func (s *Server) agentBindingProvider(agentKey string) (llm.Provider, agent.Config, int64, bool) {
id := s.effectiveProfileForAgent(agentKey, nil)
if id == nil {
return nil, agent.Config{}, false
return nil, agent.Config{}, 0, false
}
prov, cfg, ok := s.providerForProfile(*id)
if !ok {
return nil, agent.Config{}, false
return nil, agent.Config{}, 0, false
}
return s.poolForBinding(*id, prov, cfg), cfg, true
return s.poolForBinding(*id, prov, cfg), cfg, *id, true
}
// globalProvider returns the process-wide provider (persisted active profile or
@@ -696,6 +742,19 @@ func (s *Server) globalProvider() (llm.Provider, agent.Config, bool) {
return s.llmProv, s.llmCfg, true
}
// globalProviderWithProfile is globalProvider plus the llm_profiles row the
// provider was built from (0 for an environment-only configuration). The id is
// read under the same lock as the provider so the pair cannot be mixed across
// an applyLLM rebuild.
func (s *Server) globalProviderWithProfile() (llm.Provider, agent.Config, int64, bool) {
s.cfgMu.Lock()
defer s.cfgMu.Unlock()
if !s.llmOn || s.llmProv == nil {
return nil, agent.Config{}, 0, false
}
return s.llmProv, s.llmCfg, s.llmHealthProfileID, true
}
// taskRuntimeAvailable reports whether every listed role can resolve a provider
// under the runtime precedence in current(): the role's own binding first, then
// the task chain, then global. An exhausted chain is a hard stop for unbound
@@ -714,7 +773,7 @@ func (s *Server) taskRuntimeAvailable(t *Task, agentKeys ...string) bool {
_, _, unboundReady = s.globalProvider()
}
for _, key := range agentKeys {
if _, _, ok := s.agentBindingProvider(key); ok {
if _, _, _, ok := s.agentBindingProvider(key); ok {
continue
}
if !unboundReady {
+150
View File
@@ -0,0 +1,150 @@
package server
import (
"context"
"fmt"
"testing"
"github.com/Autumn-27/artex/agent"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/artex/llmpool"
"github.com/Autumn-27/norma/llm"
)
// healthHooks runs one logical call on p and reports every physical attempt to
// reg under healthID, the way streamHooks wires it in production.
func healthHooks(reg *llmpool.Registry, p llm.Provider, healthID int64) taskLLMStreamHooks {
return taskLLMStreamHooks{
current: func(context.Context) (taskLLMSelection, error) {
return taskLLMSelection{provider: p, healthProfileID: healthID, retry: agent.RetryConfig{StreamAttempts: 1}}, nil
},
exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) { return db.TaskLLMTransition{}, nil },
recordHealth: func(ctx context.Context, profileID int64, stream, committed bool, err error) {
reg.RecordCall(ctx, profileID, stream, committed, err)
},
}
}
// A successful call clears a degraded profile, on both call shapes.
func TestTaskLLMRecordsHealthOnSuccess(t *testing.T) {
resetLLMFailStreaksForTest()
ctx := context.Background()
for _, stream := range []bool{false, true} {
t.Run(fmt.Sprintf("stream=%v", stream), func(t *testing.T) {
reg := llmpool.NewRegistry(nil, nil)
reg.RecordCall(ctx, 7, false, false, fmt.Errorf("openai: status 500: down"))
if st := reg.Get(7); st.Fails != 1 {
t.Fatalf("seed state=%+v, want one failure", st)
}
runLogicalCall(ctx, stream, "health-success", healthHooks(reg, &scriptedLLMProvider{}, 7))
if st := reg.Get(7); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("state=%+v after success, want clean", st)
}
})
}
}
// A deterministic 400 on an atomic call clears the streak, matching
// Pool.Complete; the same error on a stream leaves it alone, matching
// Pool.Stream.
func TestTaskLLMRecordsNonFailoverErrorLikeThePool(t *testing.T) {
resetLLMFailStreaksForTest()
ctx := context.Background()
bad := fmt.Errorf("openai: status 400: bad request")
reg := llmpool.NewRegistry(nil, nil)
reg.RecordCall(ctx, 7, false, false, fmt.Errorf("openai: status 500: down"))
runLogicalCall(ctx, false, "health-atomic-400", healthHooks(reg, &scriptedLLMProvider{err: bad}, 7))
if st := reg.Get(7); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("atomic state=%+v after 400, want clean", st)
}
reg2 := llmpool.NewRegistry(nil, nil)
reg2.RecordCall(ctx, 8, false, false, fmt.Errorf("openai: status 500: down"))
runLogicalCall(ctx, true, "health-stream-400", healthHooks(reg2, &scriptedLLMProvider{err: bad}, 8))
if st := reg2.Get(8); st.Fails != 1 || st.Trips != 0 {
t.Fatalf("stream state=%+v after 400, want the failure kept", st)
}
}
// Every physical stream attempt is reported: a retried pre-commit 500 counts
// twice and does not trip a three-strike breaker.
func TestTaskLLMRecordsEveryStreamAttempt(t *testing.T) {
resetLLMFailStreaksForTest()
defer withZeroRetryBackoff()()
ctx := context.Background()
reg := llmpool.NewRegistry(nil, nil)
p := &scriptedLLMProvider{err: fmt.Errorf("openai: status 500: down")}
runLogicalCall(ctx, true, "health-retry", healthHooks(reg, p, 7))
if p.calls != 2 {
t.Fatalf("provider calls=%d, want one attempt plus one retry", p.calls)
}
if st := reg.Get(7); st.Fails != 2 || st.Trips != 0 {
t.Fatalf("state=%+v, want two soft failures and no trip", st)
}
}
// A failure after output reached the caller is committed: the Pool does not
// fail over there and the breaker is left alone.
func TestTaskLLMRecordsCommittedFailureAsNoHealthEvent(t *testing.T) {
resetLLMFailStreaksForTest()
reg := llmpool.NewRegistry(nil, nil)
p := &scriptedLLMProvider{
events: []llm.StreamEvent{{Type: llm.SETextDelta, Text: "partial"}},
err: fmt.Errorf("openai: status 500: mid-stream drop"),
}
runLogicalCall(context.Background(), true, "health-committed", healthHooks(reg, p, 7))
if st := reg.Get(7); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("state=%+v after a committed failure, want clean", st)
}
}
// Without a health profile (environment-only configuration) nothing is
// recorded.
func TestTaskLLMRecordsNoHealthWithoutProfile(t *testing.T) {
resetLLMFailStreaksForTest()
reg := llmpool.NewRegistry(nil, nil)
p := &scriptedLLMProvider{err: fmt.Errorf("openai: status 500: down")}
runLogicalCall(context.Background(), true, "health-none", healthHooks(reg, p, 0))
if n := len(reg.Snapshot()); n != 0 {
t.Fatalf("%d states recorded with healthProfileID 0", n)
}
}
// healthIDFor keys the registry only for a provider that does not record by
// itself: a Pool already updates every member attempt.
func TestHealthIDForSkipsPooledProviders(t *testing.T) {
bare := &scriptedLLMProvider{}
pool := llmpool.New([]*llmpool.Member{{ID: 9, Prov: bare}}, nil)
if pool == nil {
t.Fatal("pool should be non-nil")
}
if got := healthIDFor(bare, 7); got != 7 {
t.Fatalf("bare provider id=%d, want 7", got)
}
if got := healthIDFor(pool, 7); got != 0 {
t.Fatalf("pooled provider id=%d, want 0", got)
}
if got := healthIDFor(bare, 0); got != 0 {
t.Fatalf("profile 0 id=%d, want 0", got)
}
}
// The production hook closure records under the given profile and no-ops on 0.
func TestStreamHooksRecordBreakerHealth(t *testing.T) {
reg := llmpool.NewRegistry(nil, nil)
r := &taskLLMRuntime{s: &Server{llmHealth: reg}}
h := r.streamHooks()
ctx := context.Background()
h.recordHealth(ctx, 0, false, false, fmt.Errorf("openai: status 500: down"))
if n := len(reg.Snapshot()); n != 0 {
t.Fatalf("%d states recorded for profile 0", n)
}
h.recordHealth(ctx, 7, false, false, fmt.Errorf("openai: status 500: down"))
if st := reg.Get(7); st.Fails != 1 {
t.Fatalf("state=%+v, want one failure", st)
}
h.recordHealth(ctx, 7, false, false, nil)
if st := reg.Get(7); st.Fails != 0 || st.Trips != 0 {
t.Fatalf("state=%+v after success, want clean", st)
}
}
@@ -37,7 +37,14 @@ import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from "@
import { Spinner } from "@/components/ui/spinner";
import { Table, TableBody, TableCell, TableHead, TableHeader, TableRow } from "@/components/ui/table";
import { statusMeta, type Tone, toneClasses } from "@/lib/status";
import type { ActiveFindingRetest, Finding, FindingStatus, Severity, VerificationStatus } from "@/lib/types";
import type {
ActiveFindingRetest,
Finding,
FindingReportState,
FindingStatus,
Severity,
VerificationStatus,
} from "@/lib/types";
import { cn } from "@/lib/utils";
export const SEVERITIES: Severity[] = ["critical", "high", "medium", "low"];
@@ -62,6 +69,37 @@ export const VERIFICATION_STATUSES: VerificationStatus[] = [
"false_positive",
];
export const FINDING_REPORT_STATES: FindingReportState[] = [
"none",
"submitted",
"accepted",
"rejected",
"duplicate",
];
export const REPORT_STATE_META: Record<FindingReportState, { label: string; tone: Tone }> = {
none: { label: "Not submitted", tone: "slate" },
submitted: { label: "Submitted", tone: "amber" },
accepted: { label: "Accepted", tone: "green" },
rejected: { label: "Rejected", tone: "red" },
duplicate: { label: "Duplicate report", tone: "neutral" },
};
export function ReportStateBadge({ state }: { state?: FindingReportState }) {
const value = state ?? "none";
const meta = REPORT_STATE_META[value] ?? REPORT_STATE_META.none;
return (
<span
className={cn(
"inline-flex w-fit items-center whitespace-nowrap rounded-md border px-2 py-0.5 font-medium text-xs",
toneClasses[meta.tone],
)}
>
{meta.label}
</span>
);
}
// Verification outcome → label/tone, on the same tone tokens StatusBadge uses.
export const VERIFICATION_META: Record<VerificationStatus, { label: string; tone: Tone }> = {
verified: { label: "Verified", tone: "green" },
@@ -125,7 +163,7 @@ export function fmtTime(ts: string) {
});
}
const COLUMN_COUNT = 9;
const COLUMN_COUNT = 10;
interface FindingsTableProps {
items: Finding[];
@@ -141,6 +179,7 @@ interface FindingsTableProps {
saving: boolean;
onSave: (finding: Finding) => void;
onStatusChange: (finding: Finding, next: FindingStatus) => void;
onReportStateChange: (finding: Finding, next: FindingReportState) => void;
onRetest: (finding: Finding) => void;
activeRetests: Record<string, ActiveFindingRetest>;
onDeepen: (finding: Finding) => void;
@@ -164,6 +203,7 @@ export function FindingsTable({
saving,
onSave,
onStatusChange,
onReportStateChange,
onRetest,
activeRetests,
onDeepen,
@@ -181,7 +221,7 @@ export function FindingsTable({
return (
/* Fixed column widths keep expanded content from stretching the table; narrow screens scroll horizontally inside the table only. */
<Table className="min-w-[60rem] table-fixed">
<Table className="min-w-[68rem] table-fixed">
<TableHeader>
<TableRow>
<TableHead className="w-8">
@@ -196,6 +236,7 @@ export function FindingsTable({
<TableHead>Vulnerability</TableHead>
<TableHead className="w-44">Asset</TableHead>
<TableHead className="w-28">Status</TableHead>
<TableHead className="w-36">Report</TableHead>
<TableHead className="w-32">Task</TableHead>
<TableHead className="w-24">Time</TableHead>
<TableHead className="w-48">Actions</TableHead>
@@ -296,6 +337,27 @@ export function FindingsTable({
<StatusBadge domain="finding" value={f.status} dot />
)}
</TableCell>
<TableCell onClick={(e) => e.stopPropagation()}>
{f.finding_id ? (
<Select
value={f.report_state ?? "none"}
onValueChange={(v) => onReportStateChange(f, v as FindingReportState)}
>
<SelectTrigger size="sm" className="h-7 w-full border-none px-1 shadow-none focus-visible:ring-0">
<ReportStateBadge state={f.report_state} />
</SelectTrigger>
<SelectContent position="popper" align="end">
{FINDING_REPORT_STATES.map((st) => (
<SelectItem key={st} value={st}>
{REPORT_STATE_META[st].label}
</SelectItem>
))}
</SelectContent>
</Select>
) : (
<ReportStateBadge state={f.report_state} />
)}
</TableCell>
<TableCell>
{f.task_id ? (
<Link
@@ -23,9 +23,14 @@ import { SidebarTrigger } from "@/components/ui/sidebar";
import { Tabs, TabsContent, TabsList, TabsTrigger } from "@/components/ui/tabs";
import { api } from "@/lib/api";
import { statusMeta } from "@/lib/status";
import type { Finding, FindingStatus, Severity } from "@/lib/types";
import type { Finding, FindingReportState, FindingStatus, Severity } from "@/lib/types";
import { VerificationBadge } from "../_components/findings-table";
import {
FINDING_REPORT_STATES,
REPORT_STATE_META,
ReportStateBadge,
VerificationBadge,
} from "../_components/findings-table";
import { FindingLineageView } from "./lineage";
const SEVERITIES: Severity[] = ["critical", "high", "medium", "low"];
@@ -112,6 +117,23 @@ function FindingDetailInner() {
[finding, id],
);
const changeReportState = React.useCallback(
async (next: FindingReportState) => {
if (!finding || finding.inherited || next === (finding.report_state ?? "none")) return;
const prev = finding.report_state ?? "none";
setFinding({ ...finding, report_state: next });
try {
const updated = await api.setFindingReportState(id, next);
setFinding(updated);
toast.success(`Report stage changed to "${REPORT_STATE_META[next].label}"`);
} catch (e) {
setFinding((cur) => (cur ? { ...cur, report_state: prev } : cur));
toast.error("Update failed: " + (e as Error).message);
}
},
[finding, id],
);
if (!finding) {
return (
<div className="flex flex-1 flex-col items-center justify-center gap-3 p-10 text-center">
@@ -148,6 +170,7 @@ function FindingDetailInner() {
<Separator orientation="vertical" className="mx-1 h-4" />
<StatusBadge domain="severity" value={finding.severity} dot />
<StatusBadge domain="finding" value={finding.status} dot />
<ReportStateBadge state={finding.report_state} />
<VerificationBadge finding={finding} />
{finding.inherited && finding.source_task_id && (
<Badge variant="outline">Source task #{finding.source_task_id} · read-only</Badge>
@@ -294,6 +317,30 @@ function FindingDetailInner() {
)}
</FieldRow>
<FieldRow label="Report stage">
{finding.inherited ? (
<ReportStateBadge state={finding.report_state} />
) : (
<Select
value={finding.report_state ?? "none"}
onValueChange={(v) => changeReportState(v as FindingReportState)}
>
<SelectTrigger size="sm" className="h-7 w-auto border-none px-1 shadow-none focus-visible:ring-0">
<ReportStateBadge state={finding.report_state} />
</SelectTrigger>
<SelectContent position="popper" align="end">
<SelectGroup>
{FINDING_REPORT_STATES.map((st) => (
<SelectItem key={st} value={st}>
{REPORT_STATE_META[st].label}
</SelectItem>
))}
</SelectGroup>
</SelectContent>
</Select>
)}
</FieldRow>
{/* Verification: automated outcome; the note explains any state other than verified */}
{finding.verification_status && (
<FieldRow label="Verification">
+136 -9
View File
@@ -8,8 +8,10 @@ import {
ArrowUpRightIcon,
BugIcon,
ChevronRightIcon,
CircleCheckIcon,
ClockIcon,
DownloadIcon,
FileUpIcon,
InfoIcon,
SearchIcon,
ShieldAlertIcon,
@@ -46,6 +48,7 @@ import type {
Finding,
FindingAssetNode,
FindingGroup,
FindingReportState,
FindingStats,
FindingStatus,
Severity,
@@ -55,6 +58,7 @@ import { cn } from "@/lib/utils";
import { AssetTree, assetPathOf } from "./_components/asset-tree";
import {
FINDING_REPORT_STATES,
FINDING_STATUSES,
type FindingEdit,
type FindingReport,
@@ -62,6 +66,7 @@ import {
findingRowKey,
fmtTime,
isSameFinding,
REPORT_STATE_META,
SEVERITIES,
UNASSIGNED_TASK,
VERIFICATION_META,
@@ -124,6 +129,8 @@ function findingGroupKey(group: FindingGroup) {
const EMPTY_STATS: FindingStats = {
total: 0,
pending: 0,
submitted: 0,
accepted: 0,
critical: 0,
high: 0,
medium: 0,
@@ -136,6 +143,7 @@ export default function FindingsPage() {
const [view, setView] = React.useState<FindingView>("flat");
const [severity, setSeverity] = React.useState<"all" | Severity>("all");
const [status, setStatus] = React.useState<"all" | FindingStatus>("all");
const [reportState, setReportState] = React.useState<"all" | FindingReportState>("all");
const [verificationStatus, setVerificationStatus] = React.useState<"all" | VerificationStatus>("all");
const [vulnclass, setVulnclass] = React.useState<string>("all");
const [task, setTask] = React.useState<string>("all");
@@ -167,7 +175,16 @@ export default function FindingsPage() {
const retestRefreshVersion = React.useRef(0);
const [deepenDescription, setDeepenDescription] = React.useState("");
const [deepening, setDeepening] = React.useState(false);
const filterFingerprint = JSON.stringify([severity, status, verificationStatus, vulnclass, task, sort, query]);
const filterFingerprint = JSON.stringify([
severity,
status,
reportState,
verificationStatus,
vulnclass,
task,
sort,
query,
]);
const activeFilterFingerprint = React.useRef(filterFingerprint);
activeFilterFingerprint.current = filterFingerprint;
@@ -205,6 +222,7 @@ export default function FindingsPage() {
view?: unknown;
severity?: unknown;
status?: unknown;
reportState?: unknown;
verificationStatus?: unknown;
vulnclass?: unknown;
task?: unknown;
@@ -217,6 +235,9 @@ export default function FindingsPage() {
if (parsed.status === "all" || FINDING_STATUSES.includes(parsed.status as FindingStatus)) {
setStatus(parsed.status as "all" | FindingStatus);
}
if (parsed.reportState === "all" || FINDING_REPORT_STATES.includes(parsed.reportState as FindingReportState)) {
setReportState(parsed.reportState as "all" | FindingReportState);
}
if (
parsed.verificationStatus === "all" ||
VERIFICATION_STATUSES.includes(parsed.verificationStatus as VerificationStatus)
@@ -237,9 +258,9 @@ export default function FindingsPage() {
if (!preferencesHydrated) return;
setLocalStorageValue(
FINDING_LIST_PREFERENCE_KEY,
JSON.stringify({ view, severity, status, verificationStatus, vulnclass, task, sort }),
JSON.stringify({ view, severity, status, reportState, verificationStatus, vulnclass, task, sort }),
);
}, [preferencesHydrated, severity, sort, status, task, verificationStatus, view, vulnclass]);
}, [preferencesHydrated, reportState, severity, sort, status, task, verificationStatus, view, vulnclass]);
React.useEffect(() => {
const timer = window.setTimeout(() => setQuery(search.trim()), 300);
@@ -299,7 +320,7 @@ export default function FindingsPage() {
await api.exportFindings({
format: exportFormat,
scope: exportScope,
filters: { severity, status, verificationStatus, vulnclass, task, query, sort },
filters: { severity, status, reportState, verificationStatus, vulnclass, task, query, sort },
ids: [...selectedIds],
});
setExportOpen(false);
@@ -339,6 +360,7 @@ export default function FindingsPage() {
pageSize: flatPageSize,
severity,
status,
reportState,
verificationStatus,
vulnclass,
task,
@@ -360,6 +382,7 @@ export default function FindingsPage() {
flatPageSize,
severity,
status,
reportState,
verificationStatus,
vulnclass,
task,
@@ -375,7 +398,16 @@ export default function FindingsPage() {
const request = ++assetTreeRequest.current;
setAssetTree((current) => ({ ...current, loading: true }));
try {
const result = await api.findingAssetTree({ severity, status, verificationStatus, vulnclass, task, query, sort });
const result = await api.findingAssetTree({
severity,
status,
reportState,
verificationStatus,
vulnclass,
task,
query,
sort,
});
if (request !== assetTreeRequest.current || activeFilterFingerprint.current !== requestFilter) return;
setAssetTree({
nodes: result.nodes ?? [],
@@ -390,7 +422,7 @@ export default function FindingsPage() {
setAssetTree((current) => ({ ...current, loading: false }));
toast.error(`Failed to load the asset tree: ${(e as Error).message}`);
}
}, [filterFingerprint, severity, status, verificationStatus, vulnclass, task, query, sort]);
}, [filterFingerprint, reportState, severity, status, verificationStatus, vulnclass, task, query, sort]);
const refreshGroups = React.useCallback(async () => {
const requestFilter = filterFingerprint;
@@ -402,6 +434,7 @@ export default function FindingsPage() {
pageSize,
severity,
status,
reportState,
verificationStatus,
vulnclass,
task,
@@ -415,7 +448,7 @@ export default function FindingsPage() {
} catch {
// Polling keeps the last successful snapshot visible.
}
}, [filterFingerprint, page, pageSize, severity, status, verificationStatus, vulnclass, task, query, sort]);
}, [filterFingerprint, page, pageSize, reportState, severity, status, verificationStatus, vulnclass, task, query, sort]);
const loadGroup = React.useCallback(
async (key: string, groupPage: number, groupPageSize: number) => {
@@ -440,6 +473,7 @@ export default function FindingsPage() {
pageSize: groupPageSize,
severity,
status,
reportState,
verificationStatus,
vulnclass,
task: key,
@@ -475,7 +509,7 @@ export default function FindingsPage() {
}));
}
},
[filterFingerprint, severity, status, verificationStatus, vulnclass, query, sort],
[filterFingerprint, reportState, severity, status, verificationStatus, vulnclass, query, sort],
);
React.useEffect(() => {
@@ -657,6 +691,34 @@ export default function FindingsPage() {
[refreshAfterMutation, setFindings, status],
);
const updateReportState = React.useCallback(
async (f: Finding, next: FindingReportState) => {
if (!f.finding_id || next === (f.report_state ?? "none")) return;
const prev = f.report_state ?? "none";
setFindings((cur) => cur.map((x) => (isSameFinding(x, f) ? { ...x, report_state: next } : x)));
try {
await api.setFindingReportState(f.finding_id, next);
toast.success(`Report stage set to "${REPORT_STATE_META[next].label}"`);
api
.findingStats()
.then(setStats)
.catch(() => {
// The row update remains valid even if the aggregate refresh fails.
});
if (reportState !== "all" && next !== reportState) {
setFindings((cur) => cur.filter((x) => !isSameFinding(x, f)));
setTotal((t) => Math.max(0, t - 1));
setFlat((cur) => ({ ...cur, total: Math.max(0, cur.total - 1) }));
}
refreshAfterMutation(f);
} catch (e) {
setFindings((cur) => cur.map((x) => (isSameFinding(x, f) ? { ...x, report_state: prev } : x)));
toast.error(`Update failed: ${(e as Error).message}`);
}
},
[refreshAfterMutation, reportState, setFindings],
);
// applyBulkStatus sets the same triage status on every selected finding, then
// reloads the current view so rows hidden by the status filter disappear.
const applyBulkStatus = React.useCallback(
@@ -687,6 +749,36 @@ export default function FindingsPage() {
[bulkBusy, loadAssetTree, loadFlat, selectedIds, view],
);
const applyBulkReportState = React.useCallback(
async (next: FindingReportState) => {
if (selectedIds.size === 0 || bulkBusy) return;
setBulkBusy(true);
try {
const r = await api.bulkSetFindingReportState([...selectedIds], next);
setSelectedIds(new Set());
const label = REPORT_STATE_META[next].label;
toast.success(
r.updated === 1
? `Set report stage of 1 finding to "${label}"`
: `Set report stage of ${r.updated} findings to "${label}"`,
);
void loadFlat();
if (view === "asset") void loadAssetTree();
api
.findingStats()
.then(setStats)
.catch(() => {
// The list reload remains valid even if the aggregate refresh fails.
});
} catch (e) {
toast.error(`Bulk update failed: ${(e as Error).message}`);
} finally {
setBulkBusy(false);
}
},
[bulkBusy, loadAssetTree, loadFlat, selectedIds, view],
);
// The inline expanded detailed-report cache is keyed by the globally stable row key. report is long-form
// Markdown the list query doesn't include, so it is fetched once by finding_id on expand; done with empty text = no report yet.
const [reports, setReports] = React.useState<Record<string, FindingReport>>({});
@@ -812,6 +904,8 @@ export default function FindingsPage() {
const statCards = [
{ label: "Total findings", value: stats.total, icon: BugIcon },
{ label: "Pending", value: stats.pending, tone: "text-amber-500", icon: ClockIcon },
{ label: "Submitted", value: stats.submitted, tone: "text-amber-500", icon: FileUpIcon },
{ label: "Accepted", value: stats.accepted, tone: "text-emerald-500", icon: CircleCheckIcon },
{ label: "Critical", value: stats.critical, tone: "text-rose-600", icon: ShieldAlertIcon },
{ label: "High", value: stats.high, tone: "text-red-500", icon: TriangleAlertIcon },
{ label: "Medium", value: stats.medium, tone: "text-amber-500", icon: TriangleAlertIcon },
@@ -838,6 +932,7 @@ export default function FindingsPage() {
saving,
onSave: saveEdit,
onStatusChange: updateStatus,
onReportStateChange: updateReportState,
onRetest: setRetestFinding,
activeRetests,
onDeepen: openDeepen,
@@ -888,7 +983,7 @@ export default function FindingsPage() {
</Tabs>
</div>
<div className="flex flex-1 flex-col gap-4 md:gap-6">
<div className="grid grid-cols-2 gap-4 sm:grid-cols-3 lg:grid-cols-6">
<div className="grid grid-cols-2 gap-4 sm:grid-cols-4 lg:grid-cols-8">
{statCards.map((stat) => {
const StatIcon = stat.icon;
return (
@@ -956,6 +1051,20 @@ export default function FindingsPage() {
</SelectContent>
</Select>
<Select value={reportState} onValueChange={(v) => setReportState(v as "all" | FindingReportState)}>
<SelectTrigger size="sm" className="w-40">
<SelectValue placeholder="Report stage" />
</SelectTrigger>
<SelectContent>
<SelectItem value="all">All report stages</SelectItem>
{FINDING_REPORT_STATES.map((st) => (
<SelectItem key={st} value={st}>
{REPORT_STATE_META[st].label}
</SelectItem>
))}
</SelectContent>
</Select>
<Select
value={verificationStatus}
onValueChange={(v) => setVerificationStatus(v as "all" | VerificationStatus)}
@@ -1046,6 +1155,24 @@ export default function FindingsPage() {
))}
</SelectContent>
</Select>
<Select
value=""
onValueChange={(v) => {
if (v) void applyBulkReportState(v as FindingReportState);
}}
disabled={bulkBusy}
>
<SelectTrigger size="sm" className="w-44" aria-label="Set the report stage of the selected findings">
<SelectValue placeholder={bulkBusy ? "Updating…" : "Set report stage"} />
</SelectTrigger>
<SelectContent>
{FINDING_REPORT_STATES.map((st) => (
<SelectItem key={st} value={st}>
{REPORT_STATE_META[st].label}
</SelectItem>
))}
</SelectContent>
</Select>
</div>
)}
<Button size="sm" variant="outline" onClick={openExport}>
@@ -133,6 +133,9 @@ export default function LLMRecordsPage() {
// Recording on/off toggle (settings.llm_record; default off). When off the
// backend records nothing.
const [recEnabled, setRecEnabled] = React.useState(false);
// recKnown flips once the settings call resolves, so the empty state does not
// claim "recording is off" while the answer is still in flight.
const [recKnown, setRecKnown] = React.useState(false);
const [recBusy, setRecBusy] = React.useState(false);
// Inline detail panel (Burp-style split, not a dialog)
@@ -167,7 +170,11 @@ export default function LLMRecordsPage() {
let alive = true;
api
.settings()
.then((s) => { if (alive) setRecEnabled(!!s.llm_record); })
.then((s) => {
if (!alive) return;
setRecEnabled(!!s.llm_record);
setRecKnown(true);
})
.catch(() => {});
return () => { alive = false; };
}, []);
@@ -400,7 +407,9 @@ export default function LLMRecordsPage() {
) : records.length === 0 ? (
<TableRow>
<TableCell colSpan={8} className="py-12 text-center text-sm text-muted-foreground">
No LLM call records yet
{!recKnown || recEnabled
? "No LLM call records yet"
: "LLM call recording is off. Turn on the Recording toggle above to start capturing calls."}
</TableCell>
</TableRow>
) : (
+10 -1
View File
@@ -43,6 +43,7 @@ import type {
FindingGroupsPage,
FindingQuery,
FindingRetest,
FindingReportState,
FindingStats,
FindingStatus,
FindingsPage,
@@ -241,6 +242,7 @@ function findingFilterParams(q: Omit<FindingQuery, "page" | "pageSize">): URLSea
const p = new URLSearchParams();
if (q.severity && q.severity !== "all") p.set("severity", q.severity);
if (q.status && q.status !== "all") p.set("status", q.status);
if (q.reportState && q.reportState !== "all") p.set("report_state", q.reportState);
if (q.verificationStatus && q.verificationStatus !== "all") p.set("verification_status", q.verificationStatus);
if (q.vulnclass && q.vulnclass !== "all") p.set("vulnclass", q.vulnclass);
if (q.task && q.task !== "all") p.set("task_id", q.task);
@@ -703,6 +705,8 @@ export const api = {
// Vulnerability lineage: the subgraph (nodes + edges) from the vulnerability node back to the task's initial node.
findingLineage: (id: string) => get<{ nodes: TaskNode[]; edges: Edge[] }>(`/exploration/findings/${id}/lineage`),
setFindingStatus: (id: string, status: FindingStatus) => patch<Finding>(`/exploration/findings/${id}`, { status }),
setFindingReportState: (id: string, report_state: FindingReportState) =>
patch<Finding>(`/exploration/findings/${id}`, { report_state }),
// Bulk triage: one transaction server-side, with a per-id result. Unknown ids
// come back as found=false instead of failing the whole batch.
bulkSetFindingStatus: (ids: string[], status: FindingStatus) =>
@@ -710,11 +714,16 @@ export const api = {
"/exploration/findings/bulk-status",
{ ids: ids.map(Number), status },
),
bulkSetFindingReportState: (ids: string[], report_state: FindingReportState) =>
post<{ updated: number; missing: number; results: { id: number; found: boolean; changed: boolean }[] }>(
"/exploration/findings/bulk-report-state",
{ ids: ids.map(Number), report_state },
),
setFindingSeverity: (id: string, severity: Severity) => patch<Finding>(`/exploration/findings/${id}`, { severity }),
// Save a vulnerability's name/class/severity in one call (inline edit in the findings list); only present fields are sent.
updateFinding: (
id: string,
fields: { name?: string; vulnclass?: string; severity?: Severity; status?: FindingStatus },
fields: { name?: string; vulnclass?: string; severity?: Severity; status?: FindingStatus; report_state?: FindingReportState },
) => patch<Finding>(`/exploration/findings/${id}`, fields),
// Delete a vulnerability: removes the findings record + the source exploration node (gone from the findings list, the task findings tab, and the exploration graph together).
deleteFinding: (id: string) => del<{ deleted: boolean; id: number }>(`/exploration/findings/${id}`),
+27
View File
@@ -471,6 +471,8 @@ function mockFilterFindings(q: URLSearchParams): (typeof mockFindings)[number][]
const taskID = q.get("task_id");
if (severity) list = list.filter((finding) => finding.severity === severity);
if (status) list = list.filter((finding) => finding.status === status);
const reportState = q.get("report_state");
if (reportState) list = list.filter((finding) => (finding.report_state ?? "none") === reportState);
if (vulnclass) list = list.filter((finding) => finding.vulnclass === vulnclass);
if (taskID === "__unassigned__") list = list.filter((finding) => !finding.task_id);
else if (taskID) list = list.filter((finding) => finding.task_id === taskID);
@@ -1866,6 +1868,8 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
return {
total: mockFindings.length,
pending: mockFindings.filter((f) => f.status === "pending").length,
submitted: mockFindings.filter((f) => (f.report_state ?? "none") === "submitted").length,
accepted: mockFindings.filter((f) => f.report_state === "accepted").length,
critical: mockFindings.filter((f) => f.severity === "critical").length,
high: mockFindings.filter((f) => f.severity === "high").length,
medium: mockFindings.filter((f) => f.severity === "medium").length,
@@ -1874,6 +1878,28 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
tasks,
};
}
if (path === "/exploration/findings/bulk-report-state" && m === "POST") {
const ids = Array.isArray(b.ids) ? b.ids : [];
const state = typeof b.report_state === "string" ? b.report_state : "";
let updated = 0;
let missing = 0;
const results = ids.map((id) => {
const key = String(id);
const f = mockFindings.find((item) => item.id === key);
if (!f) {
missing++;
return { id, found: false, changed: false };
}
const from = f.report_state ?? "none";
const changed = from !== state;
if (changed) {
f.report_state = state as typeof f.report_state;
updated++;
}
return { id, found: true, changed, from };
});
return { updated, missing, results };
}
if (path === "/exploration/findings/asset-tree") {
const list = mockApplyAssetScope(mockFilterFindings(q), null);
return { nodes: mockBuildAssetTree(list), finding_total: list.length, truncated: false };
@@ -2081,6 +2107,7 @@ function route(m: string, path: string, seg: string[], q: URLSearchParams, b: Re
if (!f) return {};
if (m === "PATCH") {
if (typeof b.status === "string") f.status = b.status as typeof f.status;
if (typeof b.report_state === "string") f.report_state = b.report_state as typeof f.report_state;
if (typeof b.severity === "string") f.severity = b.severity as typeof f.severity;
if (typeof b.name === "string") f.name = b.name;
if (typeof b.vulnclass === "string") f.vulnclass = b.vulnclass;
+7
View File
@@ -486,6 +486,9 @@ export type FindingStatus =
// human triage status above. "blocked" is produced by no path yet but is displayed when present.
export type VerificationStatus = "verified" | "unverified" | "blocked" | "not_assessed" | "false_positive";
// Report stage after triage: a finding can be confirmed and submitted at the same time.
export type FindingReportState = "none" | "submitted" | "accepted" | "rejected" | "duplicate";
// FindingAsset is an asset bound to a vulnerability (label pre-rendered by the backend).
export interface FindingAsset {
id: string;
@@ -509,6 +512,7 @@ export interface Finding {
name?: string; // vulnerability name; display falls back to vulnclass when empty
severity: Severity;
status: FindingStatus;
report_state?: FindingReportState; // workflow stage after triage; default none
summary: string;
evidence: string;
report?: string; // detailed report (Markdown); returned only by the detail endpoint, empty in lists
@@ -562,6 +566,8 @@ export interface FindingDeepenResponse {
export interface FindingStats {
total: number;
pending: number;
submitted: number;
accepted: number;
critical: number;
high: number;
medium: number;
@@ -585,6 +591,7 @@ export interface FindingQuery {
pageSize: number;
severity?: "all" | Severity;
status?: "all" | FindingStatus;
reportState?: "all" | FindingReportState;
verificationStatus?: "all" | VerificationStatus;
vulnclass?: string;
task?: string; // task id; "all"/empty = no task filter