@clear-capabilities/agentic-security-scanner 0.150.1 → 0.151.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +137 -0
- package/bin/agentic-security.js +73 -10
- package/dist/1122.index.js +16 -16
- package/dist/1208.index.js +23 -0
- package/dist/1301.index.js +3 -3
- package/dist/1379.index.js +3 -3
- package/dist/1444.index.js +3 -3
- package/dist/1660.index.js +1 -1
- package/dist/1700.index.js +5 -5
- package/dist/1905.index.js +12 -12
- package/dist/1920.index.js +10 -10
- package/dist/2238.index.js +4 -4
- package/dist/2271.index.js +3 -3
- package/dist/2432.index.js +8 -8
- package/dist/2659.index.js +1 -1
- package/dist/2826.index.js +2 -2
- package/dist/2830.index.js +3 -3
- package/dist/2923.index.js +1 -1
- package/dist/3001.index.js +4 -4
- package/dist/3117.index.js +4 -4
- package/dist/3180.index.js +9 -9
- package/dist/3276.index.js +1 -1
- package/dist/3415.index.js +1 -1
- package/dist/3499.index.js +1 -1
- package/dist/3518.index.js +4 -4
- package/dist/3736.index.js +7 -7
- package/dist/3839.index.js +4 -4
- package/dist/4113.index.js +441 -418
- package/dist/4265.index.js +4 -4
- package/dist/4384.index.js +8 -3
- package/dist/4399.index.js +29 -3
- package/dist/4547.index.js +2 -2
- package/dist/4863.index.js +6 -6
- package/dist/4970.index.js +2 -2
- package/dist/5051.index.js +15 -15
- package/dist/5144.index.js +4 -4
- package/dist/5333.index.js +8 -8
- package/dist/5343.index.js +2 -2
- package/dist/5350.index.js +6 -6
- package/dist/5561.index.js +1 -1
- package/dist/5637.index.js +10 -5
- package/dist/5724.index.js +21642 -0
- package/dist/5756.index.js +99 -489
- package/dist/5830.index.js +3 -3
- package/dist/6257.index.js +21 -8
- package/dist/6626.index.js +5 -5
- package/dist/6662.index.js +1 -1
- package/dist/6675.index.js +4 -4
- package/dist/6730.index.js +10 -10
- package/dist/6829.index.js +4 -4
- package/dist/6944.index.js +2 -2
- package/dist/6994.index.js +5 -5
- package/dist/7039.index.js +83 -19
- package/dist/7178.index.js +11 -6
- package/dist/7227.index.js +3 -3
- package/dist/7552.index.js +2 -2
- package/dist/7709.index.js +2 -2
- package/dist/7838.index.js +1037 -0
- package/dist/8218.index.js +4 -4
- package/dist/8476.index.js +4 -4
- package/dist/8513.index.js +8 -8
- package/dist/8520.index.js +1 -1
- package/dist/8752.index.js +4 -4
- package/dist/9207.index.js +2 -2
- package/dist/9220.index.js +5 -5
- package/dist/9390.index.js +6 -6
- package/dist/9503.index.js +2 -2
- package/dist/9560.index.js +3232 -0
- package/dist/957.index.js +4 -4
- package/dist/9801.index.js +1 -1
- package/dist/9824.index.js +5 -5
- package/dist/agentic-security.mjs +23 -23
- package/dist/agentic-security.mjs.sha256 +1 -1
- package/package.json +21 -3
- package/src/dataflow/catalog.js +52 -0
- package/src/egress/redact.js +1 -1
- package/src/engine.js +13 -2
- package/src/lineage/source-registry.js +8 -0
- package/src/llm-validator/agent-loop.js +85 -5
- package/src/llm-validator/fix-proposal.js +16 -3
- package/src/llm-validator/model-capabilities.js +26 -1
- package/src/llm-validator/model-probe.js +46 -8
- package/src/llm-validator/ollama-provider.js +26 -1
- package/src/llm-validator/oom-feedback.js +69 -0
- package/src/llm-validator/providers.js +50 -0
- package/src/llm-validator/redact.js +177 -11
- package/src/posture/deterministic-fix.js +11 -0
- package/src/report/index.js +14 -3
- package/src/sast/java-ast-folding.js +18 -2
package/dist/4113.index.js
CHANGED
|
@@ -1,8 +1,235 @@
|
|
|
1
1
|
export const id = 4113;
|
|
2
|
-
export const ids = [4113,2238,
|
|
2
|
+
export const ids = [4113,2238,7838];
|
|
3
3
|
export const modules = {
|
|
4
4
|
|
|
5
|
-
/***/
|
|
5
|
+
/***/ 27785:
|
|
6
|
+
/***/ ((__unused_webpack___webpack_module__, __webpack_exports__, __webpack_require__) => {
|
|
7
|
+
|
|
8
|
+
/* harmony export */ __webpack_require__.d(__webpack_exports__, {
|
|
9
|
+
/* harmony export */ _u: () => (/* binding */ gateFixOutput)
|
|
10
|
+
/* harmony export */ });
|
|
11
|
+
/* unused harmony exports checkResidualHonesty, requireCitedEvidence, computeFixTier, checkMechanicalTierEvidence, _internals */
|
|
12
|
+
// Deterministic honesty gates on fix / finding output (#7).
|
|
13
|
+
//
|
|
14
|
+
// The project's verification discipline (scanner/CLAUDE.md) exists because
|
|
15
|
+
// several releases shipped broken or false because work was reported as done
|
|
16
|
+
// without confirming the artifact changed. Two of those failure modes are
|
|
17
|
+
// *textual* — they live in the prose an agent emits alongside a fix — and can
|
|
18
|
+
// be caught deterministically, with no LLM and no network:
|
|
19
|
+
//
|
|
20
|
+
// 1. Hand-wave residual-risk prose. "The input is adequately handled",
|
|
21
|
+
// "future work", "tbd", "later" — vague assurances that claim safety
|
|
22
|
+
// without naming a concrete remaining vector. A residual you can't name
|
|
23
|
+
// is a residual you're guessing about; reject the guess.
|
|
24
|
+
//
|
|
25
|
+
// 2. An unbacked "this is a false positive / provably safe" verdict. Marking
|
|
26
|
+
// a finding safe is a coverage *reduction* — it must cite a `file:line`
|
|
27
|
+
// that shows why, exactly like the rules-override gate refuses to silently
|
|
28
|
+
// shrink coverage.
|
|
29
|
+
//
|
|
30
|
+
// Plus a conservative fix-tier classifier so a partial remediation can never be
|
|
31
|
+
// labelled FULL: any workaround-only signal (rate-limit, docs, log-without-
|
|
32
|
+
// reject) is WORKAROUND; anything short of (sink signature changed + all callers
|
|
33
|
+
// routed + a discriminating test) is at most MITIGATION; only the full set with
|
|
34
|
+
// no partial-sanitization caveat earns FULL.
|
|
35
|
+
//
|
|
36
|
+
// Pure functions, no side effects, no throwing — safe to call from a command,
|
|
37
|
+
// a hook, or the MCP verify_fix path.
|
|
38
|
+
|
|
39
|
+
// Vague-assurance phrases that a real residual must never hide behind. Matched
|
|
40
|
+
// case-insensitively with word boundaries so "later" doesn't trip on
|
|
41
|
+
// "collateral" and "tbd" doesn't trip on a longer token.
|
|
42
|
+
const BANNED_RESIDUAL_PHRASES = Object.freeze([
|
|
43
|
+
'adequately handled',
|
|
44
|
+
'adequately handles',
|
|
45
|
+
'properly validated',
|
|
46
|
+
'properly handled',
|
|
47
|
+
'handled properly',
|
|
48
|
+
'handled safely',
|
|
49
|
+
'future work',
|
|
50
|
+
'more work needed',
|
|
51
|
+
'to be done',
|
|
52
|
+
'tbd',
|
|
53
|
+
'later',
|
|
54
|
+
]);
|
|
55
|
+
|
|
56
|
+
// A citation shaped like `file:line` — one or more non-space, non-colon chars,
|
|
57
|
+
// a colon, then digits. Unanchored: it need only appear somewhere in the item.
|
|
58
|
+
const CITATION_RE = /[^\s:]+:\d+/;
|
|
59
|
+
|
|
60
|
+
// Verdicts that assert the finding is not real and therefore demand a citation.
|
|
61
|
+
// Compared after normalizing separators (`_`/space → `-`) and lowercasing, so
|
|
62
|
+
// FALSE_POSITIVE, false-positive, and "provably safe" all land here.
|
|
63
|
+
const FP_VERDICTS = Object.freeze(new Set(['false-positive', 'provably-safe', 'safe']));
|
|
64
|
+
|
|
65
|
+
function _escapeRe(s) {
|
|
66
|
+
return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
/**
|
|
70
|
+
* Reject vague-assurance / hand-wave residual-risk prose.
|
|
71
|
+
*
|
|
72
|
+
* An empty or whitespace-only residual is ok — there is no residual to lie
|
|
73
|
+
* about. A non-empty residual is rejected when it contains any banned phrase;
|
|
74
|
+
* each match yields one violation naming the offending phrase.
|
|
75
|
+
*
|
|
76
|
+
* @param {string} residualText
|
|
77
|
+
* @returns {{ ok: boolean, violations: string[] }}
|
|
78
|
+
*/
|
|
79
|
+
function checkResidualHonesty(residualText) {
|
|
80
|
+
const text = typeof residualText === 'string' ? residualText : '';
|
|
81
|
+
if (text.trim() === '') return { ok: true, violations: [] };
|
|
82
|
+
|
|
83
|
+
const violations = [];
|
|
84
|
+
for (const phrase of BANNED_RESIDUAL_PHRASES) {
|
|
85
|
+
const re = new RegExp(`\\b${_escapeRe(phrase)}\\b`, 'i');
|
|
86
|
+
if (re.test(text)) {
|
|
87
|
+
violations.push(`vague-assurance phrase: "${phrase}"`);
|
|
88
|
+
}
|
|
89
|
+
}
|
|
90
|
+
return { ok: violations.length === 0, violations };
|
|
91
|
+
}
|
|
92
|
+
|
|
93
|
+
function _isCitation(item) {
|
|
94
|
+
if (typeof item === 'string') return CITATION_RE.test(item);
|
|
95
|
+
if (item && typeof item === 'object' && typeof item.location === 'string') {
|
|
96
|
+
return CITATION_RE.test(item.location);
|
|
97
|
+
}
|
|
98
|
+
return false;
|
|
99
|
+
}
|
|
100
|
+
|
|
101
|
+
function _normalizeVerdict(verdict) {
|
|
102
|
+
return String(verdict).trim().toLowerCase().replace(/[_\s]+/g, '-');
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
/**
|
|
106
|
+
* Require a file:line citation behind a "this is not real" verdict.
|
|
107
|
+
*
|
|
108
|
+
* For a false-positive / provably-safe / safe verdict (case-insensitive; also
|
|
109
|
+
* accepts FALSE_POSITIVE), at least one evidence item must be a `file:line`
|
|
110
|
+
* citation — either a string matching /[^\s:]+:\d+/ or an object
|
|
111
|
+
* `{ location: "file:line" }`. Any other verdict passes unconditionally.
|
|
112
|
+
*
|
|
113
|
+
* @param {string} verdict
|
|
114
|
+
* @param {Array|string|object} evidence
|
|
115
|
+
* @returns {{ ok: boolean, violations: string[] }}
|
|
116
|
+
*/
|
|
117
|
+
function requireCitedEvidence(verdict, evidence) {
|
|
118
|
+
if (typeof verdict !== 'string' || !FP_VERDICTS.has(_normalizeVerdict(verdict))) {
|
|
119
|
+
return { ok: true, violations: [] };
|
|
120
|
+
}
|
|
121
|
+
const items = Array.isArray(evidence)
|
|
122
|
+
? evidence
|
|
123
|
+
: evidence == null
|
|
124
|
+
? []
|
|
125
|
+
: [evidence];
|
|
126
|
+
if (items.some(_isCitation)) return { ok: true, violations: [] };
|
|
127
|
+
return {
|
|
128
|
+
ok: false,
|
|
129
|
+
violations: ['false-positive/safe verdict requires a file:line citation'],
|
|
130
|
+
};
|
|
131
|
+
}
|
|
132
|
+
|
|
133
|
+
/**
|
|
134
|
+
* Classify a fix into FULL | MITIGATION | WORKAROUND, conservative-first.
|
|
135
|
+
*
|
|
136
|
+
* @param {object} signals
|
|
137
|
+
* @param {boolean} signals.sinkSignatureChanged
|
|
138
|
+
* @param {boolean} signals.allCallersRouted
|
|
139
|
+
* @param {boolean} signals.testDiscriminates - a test that fails pre-fix, passes post-fix
|
|
140
|
+
* @param {boolean} [signals.rateLimitOnly]
|
|
141
|
+
* @param {boolean} [signals.docsOnly]
|
|
142
|
+
* @param {boolean} [signals.logOnlyNoReject]
|
|
143
|
+
* @param {boolean} [signals.partialSanitization]
|
|
144
|
+
* @returns {'FULL'|'MITIGATION'|'WORKAROUND'}
|
|
145
|
+
*/
|
|
146
|
+
function computeFixTier(signals) {
|
|
147
|
+
const s = signals && typeof signals === 'object' ? signals : {};
|
|
148
|
+
if (s.rateLimitOnly || s.docsOnly || s.logOnlyNoReject) return 'WORKAROUND';
|
|
149
|
+
const complete = s.sinkSignatureChanged && s.allCallersRouted && s.testDiscriminates;
|
|
150
|
+
if (s.partialSanitization || !complete) return 'MITIGATION';
|
|
151
|
+
return 'FULL';
|
|
152
|
+
}
|
|
153
|
+
|
|
154
|
+
/**
|
|
155
|
+
* FR-308: cross-check a fix-completeness TIER against MECHANICAL evidence,
|
|
156
|
+
* when any is available. `signals` (computeFixTier's input) is agent-
|
|
157
|
+
* self-reported — this module cannot compute sinkSignatureChanged /
|
|
158
|
+
* allCallersRouted / testDiscriminates itself (see the header above: "the
|
|
159
|
+
* gate can only run against claims the AGENT self-reports... nothing here
|
|
160
|
+
* is server-computable"). `pocLeg` is different: fix-verify.js's PoC leg is
|
|
161
|
+
* a REAL execution result (posture/CLAUDE.md's execution-proof tiers), not
|
|
162
|
+
* a claim. When it is available and shows the proof-of-concept STILL
|
|
163
|
+
* demonstrates the vulnerability against the patch, a self-reported FULL
|
|
164
|
+
* tier is not merely internally inconsistent — it is REFUTED by fact. This
|
|
165
|
+
* is the literal "a mitigation or workaround cannot be represented as a
|
|
166
|
+
* full fix" acceptance criterion, now backed by mechanical evidence where
|
|
167
|
+
* it exists rather than by self-report consistency alone.
|
|
168
|
+
*
|
|
169
|
+
* A `pocLeg` of `not-requested` or `inconclusive` carries no mechanical
|
|
170
|
+
* signal either way and is a no-op here — this check can only ever ADD a
|
|
171
|
+
* violation on real contrary evidence, never manufacture one from absence.
|
|
172
|
+
*
|
|
173
|
+
* @param {string} tier
|
|
174
|
+
* @param {{status: string, reason?: string}|null} pocLeg
|
|
175
|
+
* @returns {{ ok: boolean, violations: string[] }}
|
|
176
|
+
*/
|
|
177
|
+
function checkMechanicalTierEvidence(tier, pocLeg) {
|
|
178
|
+
if (!pocLeg || typeof pocLeg !== 'object') return { ok: true, violations: [] };
|
|
179
|
+
if (tier === 'FULL' && pocLeg.status === 'still-exploitable') {
|
|
180
|
+
return {
|
|
181
|
+
ok: false,
|
|
182
|
+
violations: [`tier 'FULL' is refuted by mechanical evidence: the proof-of-concept still demonstrates the vulnerability against the patch${pocLeg.reason ? ` (${pocLeg.reason})` : ''}`],
|
|
183
|
+
};
|
|
184
|
+
}
|
|
185
|
+
return { ok: true, violations: [] };
|
|
186
|
+
}
|
|
187
|
+
|
|
188
|
+
/**
|
|
189
|
+
* Compose the four gates for a single fix's output.
|
|
190
|
+
*
|
|
191
|
+
* ok = residual-honesty ok AND evidence-citation ok AND mechanical-tier-
|
|
192
|
+
* evidence ok, further constrained by the tier/residual consistency
|
|
193
|
+
* invariant:
|
|
194
|
+
* - a FULL tier must NOT carry a residual (a full fix has nothing left);
|
|
195
|
+
* - a non-FULL tier MUST document a residual (say what's still open).
|
|
196
|
+
*
|
|
197
|
+
* @param {{ residual?: string, verdict?: string, evidence?: any, signals?: object }} input
|
|
198
|
+
* @param {{ pocLeg?: object|null }} [mechanical] - FR-308: optional real
|
|
199
|
+
* execution evidence (fix-verify.js's pocLeg) to cross-check the
|
|
200
|
+
* self-reported tier against. Omitted entirely by any caller that has no
|
|
201
|
+
* PoC leg to offer — this parameter never REQUIRES mechanical evidence,
|
|
202
|
+
* it only USES it when present.
|
|
203
|
+
* @returns {{ ok: boolean, tier: string, violations: string[] }}
|
|
204
|
+
*/
|
|
205
|
+
function gateFixOutput({ residual, verdict, evidence, signals } = {}, { pocLeg = null } = {}) {
|
|
206
|
+
const tier = computeFixTier(signals);
|
|
207
|
+
const residualCheck = checkResidualHonesty(residual);
|
|
208
|
+
const evidenceCheck = requireCitedEvidence(verdict, evidence);
|
|
209
|
+
const mechanicalCheck = checkMechanicalTierEvidence(tier, pocLeg);
|
|
210
|
+
|
|
211
|
+
const violations = [...residualCheck.violations, ...evidenceCheck.violations, ...mechanicalCheck.violations];
|
|
212
|
+
let ok = residualCheck.ok && evidenceCheck.ok && mechanicalCheck.ok;
|
|
213
|
+
|
|
214
|
+
const residualEmpty = typeof residual !== 'string' || residual.trim() === '';
|
|
215
|
+
if (tier === 'FULL' && !residualEmpty) {
|
|
216
|
+
violations.push('FULL tier cannot carry a residual');
|
|
217
|
+
ok = false;
|
|
218
|
+
}
|
|
219
|
+
if (tier !== 'FULL' && residualEmpty) {
|
|
220
|
+
violations.push('non-FULL tier must document a residual');
|
|
221
|
+
ok = false;
|
|
222
|
+
}
|
|
223
|
+
|
|
224
|
+
return { ok, tier, violations };
|
|
225
|
+
}
|
|
226
|
+
|
|
227
|
+
const _internals = Object.freeze({ BANNED_RESIDUAL_PHRASES, CITATION_RE, FP_VERDICTS });
|
|
228
|
+
|
|
229
|
+
|
|
230
|
+
/***/ }),
|
|
231
|
+
|
|
232
|
+
/***/ 32238:
|
|
6
233
|
/***/ ((__unused_webpack___webpack_module__, __webpack_exports__, __webpack_require__) => {
|
|
7
234
|
|
|
8
235
|
/* harmony export */ __webpack_require__.d(__webpack_exports__, {
|
|
@@ -12,9 +239,9 @@ export const modules = {
|
|
|
12
239
|
/* harmony export */ sU: () => (/* binding */ loadFixAttempts)
|
|
13
240
|
/* harmony export */ });
|
|
14
241
|
/* unused harmony exports FIX_STAGES, bucketOf, summarizeFixDurations, _internals, summarizeFixAxes, renderFixAxes */
|
|
15
|
-
/* harmony import */ var node_fs__WEBPACK_IMPORTED_MODULE_0__ = __webpack_require__(
|
|
16
|
-
/* harmony import */ var node_path__WEBPACK_IMPORTED_MODULE_1__ = __webpack_require__(
|
|
17
|
-
/* harmony import */ var _state_dir_js__WEBPACK_IMPORTED_MODULE_2__ = __webpack_require__(
|
|
242
|
+
/* harmony import */ var node_fs__WEBPACK_IMPORTED_MODULE_0__ = __webpack_require__(73024);
|
|
243
|
+
/* harmony import */ var node_path__WEBPACK_IMPORTED_MODULE_1__ = __webpack_require__(76760);
|
|
244
|
+
/* harmony import */ var _state_dir_js__WEBPACK_IMPORTED_MODULE_2__ = __webpack_require__(31174);
|
|
18
245
|
// Time-to-validated-fix (R5, the reporting half).
|
|
19
246
|
//
|
|
20
247
|
// `verifyFix` already RUNS the stages and `test-runner.js` already times the
|
|
@@ -292,17 +519,20 @@ function renderFixAxes(sum) {
|
|
|
292
519
|
|
|
293
520
|
/***/ }),
|
|
294
521
|
|
|
295
|
-
/***/
|
|
296
|
-
/***/ ((
|
|
522
|
+
/***/ 64113:
|
|
523
|
+
/***/ ((__webpack_module__, __webpack_exports__, __webpack_require__) => {
|
|
297
524
|
|
|
525
|
+
__webpack_require__.a(__webpack_module__, async (__webpack_handle_async_dependencies__, __webpack_async_result__) => { try {
|
|
298
526
|
/* harmony export */ __webpack_require__.d(__webpack_exports__, {
|
|
299
527
|
/* harmony export */ verifyFixWithTests: () => (/* binding */ verifyFixWithTests)
|
|
300
528
|
/* harmony export */ });
|
|
301
529
|
/* unused harmony export runProjectTests */
|
|
302
|
-
/* harmony import */ var node_child_process__WEBPACK_IMPORTED_MODULE_0__ = __webpack_require__(
|
|
303
|
-
/* harmony import */ var node_fs__WEBPACK_IMPORTED_MODULE_1__ = __webpack_require__(
|
|
304
|
-
/* harmony import */ var node_path__WEBPACK_IMPORTED_MODULE_2__ = __webpack_require__(
|
|
305
|
-
/* harmony import */ var _fix_verify_js__WEBPACK_IMPORTED_MODULE_3__ = __webpack_require__(
|
|
530
|
+
/* harmony import */ var node_child_process__WEBPACK_IMPORTED_MODULE_0__ = __webpack_require__(31421);
|
|
531
|
+
/* harmony import */ var node_fs__WEBPACK_IMPORTED_MODULE_1__ = __webpack_require__(73024);
|
|
532
|
+
/* harmony import */ var node_path__WEBPACK_IMPORTED_MODULE_2__ = __webpack_require__(76760);
|
|
533
|
+
/* harmony import */ var _fix_verify_js__WEBPACK_IMPORTED_MODULE_3__ = __webpack_require__(27838);
|
|
534
|
+
var __webpack_async_dependencies__ = __webpack_handle_async_dependencies__([_fix_verify_js__WEBPACK_IMPORTED_MODULE_3__]);
|
|
535
|
+
_fix_verify_js__WEBPACK_IMPORTED_MODULE_3__ = (__webpack_async_dependencies__.then ? (await __webpack_async_dependencies__)() : __webpack_async_dependencies__)[0];
|
|
306
536
|
// Closed-loop fix verification (v0.68).
|
|
307
537
|
//
|
|
308
538
|
// Existing `fix-verify.js` does scan + lint. This module adds the third
|
|
@@ -470,412 +700,42 @@ function _summarize(legs, verdict) {
|
|
|
470
700
|
return `${verdict} (${bits.join(' · ')})`;
|
|
471
701
|
}
|
|
472
702
|
|
|
703
|
+
__webpack_async_result__();
|
|
704
|
+
} catch(e) { __webpack_async_result__(e); } });
|
|
473
705
|
|
|
474
706
|
/***/ }),
|
|
475
707
|
|
|
476
|
-
/***/
|
|
477
|
-
/***/ ((
|
|
708
|
+
/***/ 27838:
|
|
709
|
+
/***/ ((__webpack_module__, __webpack_exports__, __webpack_require__) => {
|
|
478
710
|
|
|
479
|
-
|
|
711
|
+
__webpack_require__.a(__webpack_module__, async (__webpack_handle_async_dependencies__, __webpack_async_result__) => { try {
|
|
480
712
|
__webpack_require__.r(__webpack_exports__);
|
|
481
|
-
|
|
482
|
-
|
|
483
|
-
|
|
484
|
-
|
|
485
|
-
|
|
486
|
-
|
|
487
|
-
|
|
488
|
-
|
|
489
|
-
|
|
490
|
-
var
|
|
491
|
-
|
|
492
|
-
var
|
|
493
|
-
|
|
494
|
-
|
|
495
|
-
//
|
|
496
|
-
var engine = __webpack_require__(9753);
|
|
497
|
-
;// CONCATENATED MODULE: ./src/posture/fix-honesty-gate.js
|
|
498
|
-
// Deterministic honesty gates on fix / finding output (#7).
|
|
499
|
-
//
|
|
500
|
-
// The project's verification discipline (scanner/CLAUDE.md) exists because
|
|
501
|
-
// several releases shipped broken or false because work was reported as done
|
|
502
|
-
// without confirming the artifact changed. Two of those failure modes are
|
|
503
|
-
// *textual* — they live in the prose an agent emits alongside a fix — and can
|
|
504
|
-
// be caught deterministically, with no LLM and no network:
|
|
713
|
+
/* harmony export */ __webpack_require__.d(__webpack_exports__, {
|
|
714
|
+
/* harmony export */ runProjectLinter: () => (/* binding */ runProjectLinter),
|
|
715
|
+
/* harmony export */ verifyFix: () => (/* binding */ verifyFix),
|
|
716
|
+
/* harmony export */ verifyPatch: () => (/* binding */ verifyPatch)
|
|
717
|
+
/* harmony export */ });
|
|
718
|
+
/* harmony import */ var node_child_process__WEBPACK_IMPORTED_MODULE_0__ = __webpack_require__(31421);
|
|
719
|
+
/* harmony import */ var node_fs__WEBPACK_IMPORTED_MODULE_1__ = __webpack_require__(73024);
|
|
720
|
+
/* harmony import */ var node_path__WEBPACK_IMPORTED_MODULE_2__ = __webpack_require__(76760);
|
|
721
|
+
/* harmony import */ var _engine_js__WEBPACK_IMPORTED_MODULE_3__ = __webpack_require__(67198);
|
|
722
|
+
/* harmony import */ var _fix_honesty_gate_js__WEBPACK_IMPORTED_MODULE_4__ = __webpack_require__(27785);
|
|
723
|
+
/* harmony import */ var _test_runner_js__WEBPACK_IMPORTED_MODULE_5__ = __webpack_require__(35136);
|
|
724
|
+
/* harmony import */ var _fix_metrics_js__WEBPACK_IMPORTED_MODULE_6__ = __webpack_require__(32238);
|
|
725
|
+
var __webpack_async_dependencies__ = __webpack_handle_async_dependencies__([_engine_js__WEBPACK_IMPORTED_MODULE_3__]);
|
|
726
|
+
_engine_js__WEBPACK_IMPORTED_MODULE_3__ = (__webpack_async_dependencies__.then ? (await __webpack_async_dependencies__)() : __webpack_async_dependencies__)[0];
|
|
727
|
+
// Closed-loop /fix verification (Sentinel-parity FR-L4-4, FR-L4-5).
|
|
505
728
|
//
|
|
506
|
-
//
|
|
507
|
-
//
|
|
508
|
-
// without naming a concrete remaining vector. A residual you can't name
|
|
509
|
-
// is a residual you're guessing about; reject the guess.
|
|
729
|
+
// Given a candidate patch (the new file content + the finding stableId being
|
|
730
|
+
// fixed), verify it:
|
|
510
731
|
//
|
|
511
|
-
//
|
|
512
|
-
//
|
|
513
|
-
//
|
|
514
|
-
//
|
|
515
|
-
//
|
|
516
|
-
//
|
|
517
|
-
//
|
|
518
|
-
// reject) is WORKAROUND; anything short of (sink signature changed + all callers
|
|
519
|
-
// routed + a discriminating test) is at most MITIGATION; only the full set with
|
|
520
|
-
// no partial-sanitization caveat earns FULL.
|
|
521
|
-
//
|
|
522
|
-
// Pure functions, no side effects, no throwing — safe to call from a command,
|
|
523
|
-
// a hook, or the MCP verify_fix path.
|
|
524
|
-
|
|
525
|
-
// Vague-assurance phrases that a real residual must never hide behind. Matched
|
|
526
|
-
// case-insensitively with word boundaries so "later" doesn't trip on
|
|
527
|
-
// "collateral" and "tbd" doesn't trip on a longer token.
|
|
528
|
-
const BANNED_RESIDUAL_PHRASES = Object.freeze([
|
|
529
|
-
'adequately handled',
|
|
530
|
-
'adequately handles',
|
|
531
|
-
'properly validated',
|
|
532
|
-
'properly handled',
|
|
533
|
-
'handled properly',
|
|
534
|
-
'handled safely',
|
|
535
|
-
'future work',
|
|
536
|
-
'more work needed',
|
|
537
|
-
'to be done',
|
|
538
|
-
'tbd',
|
|
539
|
-
'later',
|
|
540
|
-
]);
|
|
541
|
-
|
|
542
|
-
// A citation shaped like `file:line` — one or more non-space, non-colon chars,
|
|
543
|
-
// a colon, then digits. Unanchored: it need only appear somewhere in the item.
|
|
544
|
-
const CITATION_RE = /[^\s:]+:\d+/;
|
|
545
|
-
|
|
546
|
-
// Verdicts that assert the finding is not real and therefore demand a citation.
|
|
547
|
-
// Compared after normalizing separators (`_`/space → `-`) and lowercasing, so
|
|
548
|
-
// FALSE_POSITIVE, false-positive, and "provably safe" all land here.
|
|
549
|
-
const FP_VERDICTS = Object.freeze(new Set(['false-positive', 'provably-safe', 'safe']));
|
|
550
|
-
|
|
551
|
-
function _escapeRe(s) {
|
|
552
|
-
return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
|
|
553
|
-
}
|
|
554
|
-
|
|
555
|
-
/**
|
|
556
|
-
* Reject vague-assurance / hand-wave residual-risk prose.
|
|
557
|
-
*
|
|
558
|
-
* An empty or whitespace-only residual is ok — there is no residual to lie
|
|
559
|
-
* about. A non-empty residual is rejected when it contains any banned phrase;
|
|
560
|
-
* each match yields one violation naming the offending phrase.
|
|
561
|
-
*
|
|
562
|
-
* @param {string} residualText
|
|
563
|
-
* @returns {{ ok: boolean, violations: string[] }}
|
|
564
|
-
*/
|
|
565
|
-
function checkResidualHonesty(residualText) {
|
|
566
|
-
const text = typeof residualText === 'string' ? residualText : '';
|
|
567
|
-
if (text.trim() === '') return { ok: true, violations: [] };
|
|
568
|
-
|
|
569
|
-
const violations = [];
|
|
570
|
-
for (const phrase of BANNED_RESIDUAL_PHRASES) {
|
|
571
|
-
const re = new RegExp(`\\b${_escapeRe(phrase)}\\b`, 'i');
|
|
572
|
-
if (re.test(text)) {
|
|
573
|
-
violations.push(`vague-assurance phrase: "${phrase}"`);
|
|
574
|
-
}
|
|
575
|
-
}
|
|
576
|
-
return { ok: violations.length === 0, violations };
|
|
577
|
-
}
|
|
578
|
-
|
|
579
|
-
function _isCitation(item) {
|
|
580
|
-
if (typeof item === 'string') return CITATION_RE.test(item);
|
|
581
|
-
if (item && typeof item === 'object' && typeof item.location === 'string') {
|
|
582
|
-
return CITATION_RE.test(item.location);
|
|
583
|
-
}
|
|
584
|
-
return false;
|
|
585
|
-
}
|
|
586
|
-
|
|
587
|
-
function _normalizeVerdict(verdict) {
|
|
588
|
-
return String(verdict).trim().toLowerCase().replace(/[_\s]+/g, '-');
|
|
589
|
-
}
|
|
590
|
-
|
|
591
|
-
/**
|
|
592
|
-
* Require a file:line citation behind a "this is not real" verdict.
|
|
593
|
-
*
|
|
594
|
-
* For a false-positive / provably-safe / safe verdict (case-insensitive; also
|
|
595
|
-
* accepts FALSE_POSITIVE), at least one evidence item must be a `file:line`
|
|
596
|
-
* citation — either a string matching /[^\s:]+:\d+/ or an object
|
|
597
|
-
* `{ location: "file:line" }`. Any other verdict passes unconditionally.
|
|
598
|
-
*
|
|
599
|
-
* @param {string} verdict
|
|
600
|
-
* @param {Array|string|object} evidence
|
|
601
|
-
* @returns {{ ok: boolean, violations: string[] }}
|
|
602
|
-
*/
|
|
603
|
-
function requireCitedEvidence(verdict, evidence) {
|
|
604
|
-
if (typeof verdict !== 'string' || !FP_VERDICTS.has(_normalizeVerdict(verdict))) {
|
|
605
|
-
return { ok: true, violations: [] };
|
|
606
|
-
}
|
|
607
|
-
const items = Array.isArray(evidence)
|
|
608
|
-
? evidence
|
|
609
|
-
: evidence == null
|
|
610
|
-
? []
|
|
611
|
-
: [evidence];
|
|
612
|
-
if (items.some(_isCitation)) return { ok: true, violations: [] };
|
|
613
|
-
return {
|
|
614
|
-
ok: false,
|
|
615
|
-
violations: ['false-positive/safe verdict requires a file:line citation'],
|
|
616
|
-
};
|
|
617
|
-
}
|
|
618
|
-
|
|
619
|
-
/**
|
|
620
|
-
* Classify a fix into FULL | MITIGATION | WORKAROUND, conservative-first.
|
|
621
|
-
*
|
|
622
|
-
* @param {object} signals
|
|
623
|
-
* @param {boolean} signals.sinkSignatureChanged
|
|
624
|
-
* @param {boolean} signals.allCallersRouted
|
|
625
|
-
* @param {boolean} signals.testDiscriminates - a test that fails pre-fix, passes post-fix
|
|
626
|
-
* @param {boolean} [signals.rateLimitOnly]
|
|
627
|
-
* @param {boolean} [signals.docsOnly]
|
|
628
|
-
* @param {boolean} [signals.logOnlyNoReject]
|
|
629
|
-
* @param {boolean} [signals.partialSanitization]
|
|
630
|
-
* @returns {'FULL'|'MITIGATION'|'WORKAROUND'}
|
|
631
|
-
*/
|
|
632
|
-
function computeFixTier(signals) {
|
|
633
|
-
const s = signals && typeof signals === 'object' ? signals : {};
|
|
634
|
-
if (s.rateLimitOnly || s.docsOnly || s.logOnlyNoReject) return 'WORKAROUND';
|
|
635
|
-
const complete = s.sinkSignatureChanged && s.allCallersRouted && s.testDiscriminates;
|
|
636
|
-
if (s.partialSanitization || !complete) return 'MITIGATION';
|
|
637
|
-
return 'FULL';
|
|
638
|
-
}
|
|
639
|
-
|
|
640
|
-
/**
|
|
641
|
-
* FR-308: cross-check a fix-completeness TIER against MECHANICAL evidence,
|
|
642
|
-
* when any is available. `signals` (computeFixTier's input) is agent-
|
|
643
|
-
* self-reported — this module cannot compute sinkSignatureChanged /
|
|
644
|
-
* allCallersRouted / testDiscriminates itself (see the header above: "the
|
|
645
|
-
* gate can only run against claims the AGENT self-reports... nothing here
|
|
646
|
-
* is server-computable"). `pocLeg` is different: fix-verify.js's PoC leg is
|
|
647
|
-
* a REAL execution result (posture/CLAUDE.md's execution-proof tiers), not
|
|
648
|
-
* a claim. When it is available and shows the proof-of-concept STILL
|
|
649
|
-
* demonstrates the vulnerability against the patch, a self-reported FULL
|
|
650
|
-
* tier is not merely internally inconsistent — it is REFUTED by fact. This
|
|
651
|
-
* is the literal "a mitigation or workaround cannot be represented as a
|
|
652
|
-
* full fix" acceptance criterion, now backed by mechanical evidence where
|
|
653
|
-
* it exists rather than by self-report consistency alone.
|
|
654
|
-
*
|
|
655
|
-
* A `pocLeg` of `not-requested` or `inconclusive` carries no mechanical
|
|
656
|
-
* signal either way and is a no-op here — this check can only ever ADD a
|
|
657
|
-
* violation on real contrary evidence, never manufacture one from absence.
|
|
658
|
-
*
|
|
659
|
-
* @param {string} tier
|
|
660
|
-
* @param {{status: string, reason?: string}|null} pocLeg
|
|
661
|
-
* @returns {{ ok: boolean, violations: string[] }}
|
|
662
|
-
*/
|
|
663
|
-
function checkMechanicalTierEvidence(tier, pocLeg) {
|
|
664
|
-
if (!pocLeg || typeof pocLeg !== 'object') return { ok: true, violations: [] };
|
|
665
|
-
if (tier === 'FULL' && pocLeg.status === 'still-exploitable') {
|
|
666
|
-
return {
|
|
667
|
-
ok: false,
|
|
668
|
-
violations: [`tier 'FULL' is refuted by mechanical evidence: the proof-of-concept still demonstrates the vulnerability against the patch${pocLeg.reason ? ` (${pocLeg.reason})` : ''}`],
|
|
669
|
-
};
|
|
670
|
-
}
|
|
671
|
-
return { ok: true, violations: [] };
|
|
672
|
-
}
|
|
673
|
-
|
|
674
|
-
/**
|
|
675
|
-
* Compose the four gates for a single fix's output.
|
|
676
|
-
*
|
|
677
|
-
* ok = residual-honesty ok AND evidence-citation ok AND mechanical-tier-
|
|
678
|
-
* evidence ok, further constrained by the tier/residual consistency
|
|
679
|
-
* invariant:
|
|
680
|
-
* - a FULL tier must NOT carry a residual (a full fix has nothing left);
|
|
681
|
-
* - a non-FULL tier MUST document a residual (say what's still open).
|
|
682
|
-
*
|
|
683
|
-
* @param {{ residual?: string, verdict?: string, evidence?: any, signals?: object }} input
|
|
684
|
-
* @param {{ pocLeg?: object|null }} [mechanical] - FR-308: optional real
|
|
685
|
-
* execution evidence (fix-verify.js's pocLeg) to cross-check the
|
|
686
|
-
* self-reported tier against. Omitted entirely by any caller that has no
|
|
687
|
-
* PoC leg to offer — this parameter never REQUIRES mechanical evidence,
|
|
688
|
-
* it only USES it when present.
|
|
689
|
-
* @returns {{ ok: boolean, tier: string, violations: string[] }}
|
|
690
|
-
*/
|
|
691
|
-
function gateFixOutput({ residual, verdict, evidence, signals } = {}, { pocLeg = null } = {}) {
|
|
692
|
-
const tier = computeFixTier(signals);
|
|
693
|
-
const residualCheck = checkResidualHonesty(residual);
|
|
694
|
-
const evidenceCheck = requireCitedEvidence(verdict, evidence);
|
|
695
|
-
const mechanicalCheck = checkMechanicalTierEvidence(tier, pocLeg);
|
|
696
|
-
|
|
697
|
-
const violations = [...residualCheck.violations, ...evidenceCheck.violations, ...mechanicalCheck.violations];
|
|
698
|
-
let ok = residualCheck.ok && evidenceCheck.ok && mechanicalCheck.ok;
|
|
699
|
-
|
|
700
|
-
const residualEmpty = typeof residual !== 'string' || residual.trim() === '';
|
|
701
|
-
if (tier === 'FULL' && !residualEmpty) {
|
|
702
|
-
violations.push('FULL tier cannot carry a residual');
|
|
703
|
-
ok = false;
|
|
704
|
-
}
|
|
705
|
-
if (tier !== 'FULL' && residualEmpty) {
|
|
706
|
-
violations.push('non-FULL tier must document a residual');
|
|
707
|
-
ok = false;
|
|
708
|
-
}
|
|
709
|
-
|
|
710
|
-
return { ok, tier, violations };
|
|
711
|
-
}
|
|
712
|
-
|
|
713
|
-
const _internals = Object.freeze({ BANNED_RESIDUAL_PHRASES, CITATION_RE, FP_VERDICTS });
|
|
714
|
-
|
|
715
|
-
;// CONCATENATED MODULE: ./src/posture/test-runner.js
|
|
716
|
-
// R5 (partial, roadmap) — the project's own test suite as a verification
|
|
717
|
-
// stage for `verifyFix()` (see `fix-verify.js`). Closes the gap where a
|
|
718
|
-
// "verified" fix only proved a finding's stableId stopped firing — a patch
|
|
719
|
-
// that deletes the feature entirely would satisfy that just as well as a
|
|
720
|
-
// real fix. Running the project's own tests is the cheapest available check
|
|
721
|
-
// that the application still works.
|
|
722
|
-
//
|
|
723
|
-
// Execution-safety note: this spawns the TARGET PROJECT's own test command
|
|
724
|
-
// in the target project's own directory. That is deliberately NOT routed
|
|
725
|
-
// through the R1 confinement sandbox (`../sandbox/`). That sandbox exists to
|
|
726
|
-
// contain untrusted proof-of-concept exploit code the scanner itself
|
|
727
|
-
// synthesizes — code nobody has vetted, being run for the first time. A
|
|
728
|
-
// project's pre-existing test suite is the opposite case: it is the
|
|
729
|
-
// project's own trusted source, already sitting on disk, and running it is
|
|
730
|
-
// exactly what a human developer does by hand before trusting a fix.
|
|
731
|
-
// Wrapping "npm test" / "pytest" / "go test" in the PoC sandbox's
|
|
732
|
-
// syscall/network/filesystem restrictions would break the large majority of
|
|
733
|
-
// real test suites (they bind local ports, spawn child processes, write temp
|
|
734
|
-
// fixtures, etc.) for no corresponding security benefit.
|
|
735
|
-
|
|
736
|
-
|
|
737
|
-
|
|
738
|
-
|
|
739
|
-
|
|
740
|
-
const DEFAULT_TIMEOUT_MS = 300_000;
|
|
741
|
-
const NPM_PLACEHOLDER = /Error: no test specified/i;
|
|
742
|
-
|
|
743
|
-
function _exists(scanRoot, rel) {
|
|
744
|
-
try { return external_node_fs_.existsSync(external_node_path_.join(scanRoot, rel)); } catch { return false; }
|
|
745
|
-
}
|
|
746
|
-
|
|
747
|
-
function _isDir(scanRoot, rel) {
|
|
748
|
-
try { return external_node_fs_.statSync(external_node_path_.join(scanRoot, rel)).isDirectory(); } catch { return false; }
|
|
749
|
-
}
|
|
750
|
-
|
|
751
|
-
function _binaryAvailable(cmd) {
|
|
752
|
-
try {
|
|
753
|
-
const r = (0,external_node_child_process_.spawnSync)(cmd, ['--version'], { timeout: 5_000, stdio: 'ignore' });
|
|
754
|
-
return !(r.error && r.error.code === 'ENOENT');
|
|
755
|
-
} catch {
|
|
756
|
-
return false;
|
|
757
|
-
}
|
|
758
|
-
}
|
|
759
|
-
|
|
760
|
-
// Detect the project's test command. Read-only — never spawns the actual
|
|
761
|
-
// test run, only (optionally) a cheap `--version` probe to confirm a tool
|
|
762
|
-
// like `pytest` is actually installed before committing to it. Returns
|
|
763
|
-
// `null` when nothing detectable is found — most scanned repos will hit
|
|
764
|
-
// this path, and that must not be treated as a failure by callers.
|
|
765
|
-
function detectTestCommand(scanRoot) {
|
|
766
|
-
if (!scanRoot) return null;
|
|
767
|
-
|
|
768
|
-
// JS/TS — package.json with a real (non-placeholder) `scripts.test`.
|
|
769
|
-
let pkg = null;
|
|
770
|
-
try { pkg = JSON.parse(external_node_fs_.readFileSync(external_node_path_.join(scanRoot, 'package.json'), 'utf8')); } catch { pkg = null; }
|
|
771
|
-
const testScript = pkg && pkg.scripts && pkg.scripts.test;
|
|
772
|
-
if (testScript && !NPM_PLACEHOLDER.test(String(testScript))) {
|
|
773
|
-
if (_exists(scanRoot, 'pnpm-lock.yaml')) return { cmd: 'pnpm', args: ['test'], kind: 'pnpm' };
|
|
774
|
-
if (_exists(scanRoot, 'yarn.lock')) return { cmd: 'yarn', args: ['test'], kind: 'yarn' };
|
|
775
|
-
if (_exists(scanRoot, 'bun.lockb') || _exists(scanRoot, 'bun.lock')) return { cmd: 'bun', args: ['test'], kind: 'bun' };
|
|
776
|
-
return { cmd: 'npm', args: ['test', '--silent'], kind: 'npm' };
|
|
777
|
-
}
|
|
778
|
-
|
|
779
|
-
// Python — pytest.ini / pyproject.toml / tox.ini / a tests/ dir, and the
|
|
780
|
-
// `pytest` binary actually available. If pytest isn't installed we do NOT
|
|
781
|
-
// report a python test command — falling through lets a later language
|
|
782
|
-
// marker (e.g. go.mod in a polyglot repo) still be detected.
|
|
783
|
-
const pyMarker = _exists(scanRoot, 'pytest.ini') || _exists(scanRoot, 'pyproject.toml') ||
|
|
784
|
-
_exists(scanRoot, 'tox.ini') || _isDir(scanRoot, 'tests');
|
|
785
|
-
if (pyMarker && _binaryAvailable('pytest')) {
|
|
786
|
-
return { cmd: 'pytest', args: ['-q'], kind: 'pytest' };
|
|
787
|
-
}
|
|
788
|
-
|
|
789
|
-
// Go
|
|
790
|
-
if (_exists(scanRoot, 'go.mod')) {
|
|
791
|
-
return { cmd: 'go', args: ['test', './...'], kind: 'go' };
|
|
792
|
-
}
|
|
793
|
-
|
|
794
|
-
return null;
|
|
795
|
-
}
|
|
796
|
-
|
|
797
|
-
// Run the detected test command with a walltime budget. Always returns a
|
|
798
|
-
// result object — never throws. Distinguishes four outcomes:
|
|
799
|
-
// - no detectable/runnable command -> status: 'skipped' (does NOT fail)
|
|
800
|
-
// - ran and exited 0 -> status: 'passed'
|
|
801
|
-
// - ran and exited non-zero -> status: 'failed'
|
|
802
|
-
// - ran past the timeout budget -> status: 'failed', timedOut: true
|
|
803
|
-
function runProjectTests(scanRoot, { timeoutMs = DEFAULT_TIMEOUT_MS } = {}) {
|
|
804
|
-
const startedAt = Date.now();
|
|
805
|
-
const command = detectTestCommand(scanRoot);
|
|
806
|
-
if (!command) {
|
|
807
|
-
return {
|
|
808
|
-
status: 'skipped', passed: null, skipped: true,
|
|
809
|
-
reason: 'no-test-command-detected', exitCode: null, timedOut: false,
|
|
810
|
-
durationMs: Date.now() - startedAt,
|
|
811
|
-
};
|
|
812
|
-
}
|
|
813
|
-
|
|
814
|
-
let r;
|
|
815
|
-
try {
|
|
816
|
-
r = (0,external_node_child_process_.spawnSync)(command.cmd, command.args, {
|
|
817
|
-
cwd: scanRoot,
|
|
818
|
-
encoding: 'utf8',
|
|
819
|
-
timeout: timeoutMs,
|
|
820
|
-
env: { ...process.env, CI: '1' },
|
|
821
|
-
});
|
|
822
|
-
} catch (e) {
|
|
823
|
-
// The spawn call itself threw (rare — e.g. cwd vanished). Treat as
|
|
824
|
-
// "could not run", not "ran and failed".
|
|
825
|
-
return {
|
|
826
|
-
status: 'skipped', passed: null, skipped: true,
|
|
827
|
-
reason: `spawn-error: ${e.message}`, exitCode: null, timedOut: false,
|
|
828
|
-
durationMs: Date.now() - startedAt,
|
|
829
|
-
};
|
|
830
|
-
}
|
|
831
|
-
const durationMs = Date.now() - startedAt;
|
|
832
|
-
|
|
833
|
-
if (r.error && r.error.code === 'ENOENT') {
|
|
834
|
-
// The detected tool isn't actually installed on this machine. Not a
|
|
835
|
-
// test failure — the suite never ran.
|
|
836
|
-
return {
|
|
837
|
-
status: 'skipped', passed: null, skipped: true,
|
|
838
|
-
reason: `${command.kind}-not-installed`, exitCode: null, timedOut: false, durationMs,
|
|
839
|
-
};
|
|
840
|
-
}
|
|
841
|
-
|
|
842
|
-
if (r.status === null) {
|
|
843
|
-
// spawnSync sets status:null both on timeout-kill and on being killed by
|
|
844
|
-
// another signal; either way the run did not complete, which is a
|
|
845
|
-
// verification failure, never a skip — we asked for a result and the
|
|
846
|
-
// process was terminated before producing one.
|
|
847
|
-
return {
|
|
848
|
-
status: 'failed', passed: false, skipped: false,
|
|
849
|
-
reason: 'timed-out', exitCode: null, timedOut: true, durationMs,
|
|
850
|
-
};
|
|
851
|
-
}
|
|
852
|
-
|
|
853
|
-
return {
|
|
854
|
-
status: r.status === 0 ? 'passed' : 'failed',
|
|
855
|
-
passed: r.status === 0,
|
|
856
|
-
skipped: false,
|
|
857
|
-
reason: r.status === 0 ? null : 'test-failures',
|
|
858
|
-
exitCode: r.status,
|
|
859
|
-
timedOut: false,
|
|
860
|
-
durationMs,
|
|
861
|
-
};
|
|
862
|
-
}
|
|
863
|
-
|
|
864
|
-
// EXTERNAL MODULE: ./src/posture/fix-metrics.js
|
|
865
|
-
var fix_metrics = __webpack_require__(2238);
|
|
866
|
-
;// CONCATENATED MODULE: ./src/posture/fix-verify.js
|
|
867
|
-
// Closed-loop /fix verification (Sentinel-parity FR-L4-4, FR-L4-5).
|
|
868
|
-
//
|
|
869
|
-
// Given a candidate patch (the new file content + the finding stableId being
|
|
870
|
-
// fixed), verify it:
|
|
871
|
-
//
|
|
872
|
-
// 1. The original finding's stableId no longer fires on the patched file.
|
|
873
|
-
// 2. No new findings at severity ≥ medium were introduced by the patch.
|
|
874
|
-
// 3. The project's existing linter (when present) passes on the patched file.
|
|
875
|
-
// 4. The project's own test suite (when detectable) still passes. This is
|
|
876
|
-
// the R5 gap-closer: a patch that silently deletes the feature would
|
|
877
|
-
// satisfy (1) and (2) just as well as a real fix — only running the
|
|
878
|
-
// tests catches that. See `test-runner.js` for detection + execution.
|
|
732
|
+
// 1. The original finding's stableId no longer fires on the patched file.
|
|
733
|
+
// 2. No new findings at severity ≥ medium were introduced by the patch.
|
|
734
|
+
// 3. The project's existing linter (when present) passes on the patched file.
|
|
735
|
+
// 4. The project's own test suite (when detectable) still passes. This is
|
|
736
|
+
// the R5 gap-closer: a patch that silently deletes the feature would
|
|
737
|
+
// satisfy (1) and (2) just as well as a real fix — only running the
|
|
738
|
+
// tests catches that. See `test-runner.js` for detection + execution.
|
|
879
739
|
//
|
|
880
740
|
// If any of those fail, the caller is expected to NOT apply the patch and
|
|
881
741
|
// instead surface a "fix plan" — a numbered list of steps the engineer can
|
|
@@ -912,7 +772,7 @@ async function verifyPatch({
|
|
|
912
772
|
// would mass-mark the rest of the project as remediated, then
|
|
913
773
|
// `reintroduced` on the next real scan. The patched content is also not
|
|
914
774
|
// committed, so there is no history to resolve provenance against anyway.
|
|
915
|
-
scan = await (0,
|
|
775
|
+
scan = await (0,_engine_js__WEBPACK_IMPORTED_MODULE_3__/* .runFullScan */ .wW)({ fileContents, depFileContents, scanRoot, provenance: false }, () => {});
|
|
916
776
|
} catch (e) {
|
|
917
777
|
return { ok: false, reason: 'rescan-failed', error: e.message };
|
|
918
778
|
}
|
|
@@ -943,7 +803,7 @@ function runProjectLinter(scanRoot, filePaths) {
|
|
|
943
803
|
if (!scanRoot || !Array.isArray(filePaths) || filePaths.length === 0) {
|
|
944
804
|
return { ok: true, runner: 'none' };
|
|
945
805
|
}
|
|
946
|
-
const has = (p) => { try { return
|
|
806
|
+
const has = (p) => { try { return node_fs__WEBPACK_IMPORTED_MODULE_1__.existsSync(node_path__WEBPACK_IMPORTED_MODULE_2__.join(scanRoot, p)); } catch { return false; } };
|
|
947
807
|
// Pick the linter by config file present in the repo root.
|
|
948
808
|
const jsFiles = filePaths.filter(f => /\.(?:js|jsx|ts|tsx|mjs|cjs)$/i.test(f));
|
|
949
809
|
const pyFiles = filePaths.filter(f => /\.py$/i.test(f));
|
|
@@ -971,7 +831,7 @@ function runProjectLinter(scanRoot, filePaths) {
|
|
|
971
831
|
function runLinter(cwd, cmd, args) {
|
|
972
832
|
let r;
|
|
973
833
|
try {
|
|
974
|
-
r = (0,
|
|
834
|
+
r = (0,node_child_process__WEBPACK_IMPORTED_MODULE_0__.spawnSync)(cmd, args, { cwd, encoding: 'utf8', timeout: 60_000 });
|
|
975
835
|
} catch (e) {
|
|
976
836
|
return { ok: true, runner: cmd, skipped: true, reason: 'binary-missing', error: e.message };
|
|
977
837
|
}
|
|
@@ -1023,8 +883,8 @@ function _candidateDiffersFromDisk(scanRoot, files) {
|
|
|
1023
883
|
for (const [rel, content] of Object.entries(files)) {
|
|
1024
884
|
if (typeof content !== 'string') continue;
|
|
1025
885
|
try {
|
|
1026
|
-
const abs =
|
|
1027
|
-
if (
|
|
886
|
+
const abs = node_path__WEBPACK_IMPORTED_MODULE_2__.resolve(scanRoot, rel);
|
|
887
|
+
if (node_fs__WEBPACK_IMPORTED_MODULE_1__.readFileSync(abs, 'utf8') !== content) return true;
|
|
1028
888
|
} catch {
|
|
1029
889
|
return true; // candidate file absent on disk -> definitely not applied
|
|
1030
890
|
}
|
|
@@ -1054,7 +914,7 @@ async function verifyFix({
|
|
|
1054
914
|
_lap('rescan');
|
|
1055
915
|
const lint = runProjectLinter(scanRoot, Object.keys(files || {}));
|
|
1056
916
|
_lap('lint');
|
|
1057
|
-
const tests = runProjectTests(scanRoot, testTimeoutMs != null ? { timeoutMs: testTimeoutMs } : {});
|
|
917
|
+
const tests = (0,_test_runner_js__WEBPACK_IMPORTED_MODULE_5__/* .runProjectTests */ .I)(scanRoot, testTimeoutMs != null ? { timeoutMs: testTimeoutMs } : {});
|
|
1058
918
|
_lap('tests');
|
|
1059
919
|
// True when a candidate patch was supplied but has not been written, so the
|
|
1060
920
|
// suite necessarily ran against the pre-patch tree. Surfaced in the summary
|
|
@@ -1119,7 +979,7 @@ async function verifyFix({
|
|
|
1119
979
|
fixMeta.evidence !== undefined || fixMeta.signals !== undefined);
|
|
1120
980
|
let honesty = null;
|
|
1121
981
|
if (hasHonestyClaim) {
|
|
1122
|
-
try { honesty = gateFixOutput(fixMeta, { pocLeg }); } catch { honesty = null; }
|
|
982
|
+
try { honesty = (0,_fix_honesty_gate_js__WEBPACK_IMPORTED_MODULE_4__/* .gateFixOutput */ ._u)(fixMeta, { pocLeg }); } catch { honesty = null; }
|
|
1123
983
|
}
|
|
1124
984
|
_lap('honesty');
|
|
1125
985
|
|
|
@@ -1174,7 +1034,7 @@ async function verifyFix({
|
|
|
1174
1034
|
// suite ran against the pre-patch tree, so its timing is real while its
|
|
1175
1035
|
// verdict is about a different tree.
|
|
1176
1036
|
if (recordMetrics && scanRoot) {
|
|
1177
|
-
(0,
|
|
1037
|
+
(0,_fix_metrics_js__WEBPACK_IMPORTED_MODULE_6__/* .recordFixAttempt */ .I3)(scanRoot, {
|
|
1178
1038
|
at: new Date().toISOString(),
|
|
1179
1039
|
stableId: originalFindingStableId || null,
|
|
1180
1040
|
ok,
|
|
@@ -1194,6 +1054,169 @@ async function verifyFix({
|
|
|
1194
1054
|
return { ok, verifiedFull, degradedLegs, rescan, lint, tests, testedPrePatch: _testedPrePatch, honesty, poc: pocLeg, durations, summary };
|
|
1195
1055
|
}
|
|
1196
1056
|
|
|
1057
|
+
__webpack_async_result__();
|
|
1058
|
+
} catch(e) { __webpack_async_result__(e); } });
|
|
1059
|
+
|
|
1060
|
+
/***/ }),
|
|
1061
|
+
|
|
1062
|
+
/***/ 35136:
|
|
1063
|
+
/***/ ((__unused_webpack___webpack_module__, __webpack_exports__, __webpack_require__) => {
|
|
1064
|
+
|
|
1065
|
+
/* harmony export */ __webpack_require__.d(__webpack_exports__, {
|
|
1066
|
+
/* harmony export */ I: () => (/* binding */ runProjectTests)
|
|
1067
|
+
/* harmony export */ });
|
|
1068
|
+
/* unused harmony export detectTestCommand */
|
|
1069
|
+
/* harmony import */ var node_child_process__WEBPACK_IMPORTED_MODULE_0__ = __webpack_require__(31421);
|
|
1070
|
+
/* harmony import */ var node_fs__WEBPACK_IMPORTED_MODULE_1__ = __webpack_require__(73024);
|
|
1071
|
+
/* harmony import */ var node_path__WEBPACK_IMPORTED_MODULE_2__ = __webpack_require__(76760);
|
|
1072
|
+
// R5 (partial, roadmap) — the project's own test suite as a verification
|
|
1073
|
+
// stage for `verifyFix()` (see `fix-verify.js`). Closes the gap where a
|
|
1074
|
+
// "verified" fix only proved a finding's stableId stopped firing — a patch
|
|
1075
|
+
// that deletes the feature entirely would satisfy that just as well as a
|
|
1076
|
+
// real fix. Running the project's own tests is the cheapest available check
|
|
1077
|
+
// that the application still works.
|
|
1078
|
+
//
|
|
1079
|
+
// Execution-safety note: this spawns the TARGET PROJECT's own test command
|
|
1080
|
+
// in the target project's own directory. That is deliberately NOT routed
|
|
1081
|
+
// through the R1 confinement sandbox (`../sandbox/`). That sandbox exists to
|
|
1082
|
+
// contain untrusted proof-of-concept exploit code the scanner itself
|
|
1083
|
+
// synthesizes — code nobody has vetted, being run for the first time. A
|
|
1084
|
+
// project's pre-existing test suite is the opposite case: it is the
|
|
1085
|
+
// project's own trusted source, already sitting on disk, and running it is
|
|
1086
|
+
// exactly what a human developer does by hand before trusting a fix.
|
|
1087
|
+
// Wrapping "npm test" / "pytest" / "go test" in the PoC sandbox's
|
|
1088
|
+
// syscall/network/filesystem restrictions would break the large majority of
|
|
1089
|
+
// real test suites (they bind local ports, spawn child processes, write temp
|
|
1090
|
+
// fixtures, etc.) for no corresponding security benefit.
|
|
1091
|
+
|
|
1092
|
+
|
|
1093
|
+
|
|
1094
|
+
|
|
1095
|
+
|
|
1096
|
+
const DEFAULT_TIMEOUT_MS = 300_000;
|
|
1097
|
+
const NPM_PLACEHOLDER = /Error: no test specified/i;
|
|
1098
|
+
|
|
1099
|
+
function _exists(scanRoot, rel) {
|
|
1100
|
+
try { return node_fs__WEBPACK_IMPORTED_MODULE_1__.existsSync(node_path__WEBPACK_IMPORTED_MODULE_2__.join(scanRoot, rel)); } catch { return false; }
|
|
1101
|
+
}
|
|
1102
|
+
|
|
1103
|
+
function _isDir(scanRoot, rel) {
|
|
1104
|
+
try { return node_fs__WEBPACK_IMPORTED_MODULE_1__.statSync(node_path__WEBPACK_IMPORTED_MODULE_2__.join(scanRoot, rel)).isDirectory(); } catch { return false; }
|
|
1105
|
+
}
|
|
1106
|
+
|
|
1107
|
+
function _binaryAvailable(cmd) {
|
|
1108
|
+
try {
|
|
1109
|
+
const r = (0,node_child_process__WEBPACK_IMPORTED_MODULE_0__.spawnSync)(cmd, ['--version'], { timeout: 5_000, stdio: 'ignore' });
|
|
1110
|
+
return !(r.error && r.error.code === 'ENOENT');
|
|
1111
|
+
} catch {
|
|
1112
|
+
return false;
|
|
1113
|
+
}
|
|
1114
|
+
}
|
|
1115
|
+
|
|
1116
|
+
// Detect the project's test command. Read-only — never spawns the actual
|
|
1117
|
+
// test run, only (optionally) a cheap `--version` probe to confirm a tool
|
|
1118
|
+
// like `pytest` is actually installed before committing to it. Returns
|
|
1119
|
+
// `null` when nothing detectable is found — most scanned repos will hit
|
|
1120
|
+
// this path, and that must not be treated as a failure by callers.
|
|
1121
|
+
function detectTestCommand(scanRoot) {
|
|
1122
|
+
if (!scanRoot) return null;
|
|
1123
|
+
|
|
1124
|
+
// JS/TS — package.json with a real (non-placeholder) `scripts.test`.
|
|
1125
|
+
let pkg = null;
|
|
1126
|
+
try { pkg = JSON.parse(node_fs__WEBPACK_IMPORTED_MODULE_1__.readFileSync(node_path__WEBPACK_IMPORTED_MODULE_2__.join(scanRoot, 'package.json'), 'utf8')); } catch { pkg = null; }
|
|
1127
|
+
const testScript = pkg && pkg.scripts && pkg.scripts.test;
|
|
1128
|
+
if (testScript && !NPM_PLACEHOLDER.test(String(testScript))) {
|
|
1129
|
+
if (_exists(scanRoot, 'pnpm-lock.yaml')) return { cmd: 'pnpm', args: ['test'], kind: 'pnpm' };
|
|
1130
|
+
if (_exists(scanRoot, 'yarn.lock')) return { cmd: 'yarn', args: ['test'], kind: 'yarn' };
|
|
1131
|
+
if (_exists(scanRoot, 'bun.lockb') || _exists(scanRoot, 'bun.lock')) return { cmd: 'bun', args: ['test'], kind: 'bun' };
|
|
1132
|
+
return { cmd: 'npm', args: ['test', '--silent'], kind: 'npm' };
|
|
1133
|
+
}
|
|
1134
|
+
|
|
1135
|
+
// Python — pytest.ini / pyproject.toml / tox.ini / a tests/ dir, and the
|
|
1136
|
+
// `pytest` binary actually available. If pytest isn't installed we do NOT
|
|
1137
|
+
// report a python test command — falling through lets a later language
|
|
1138
|
+
// marker (e.g. go.mod in a polyglot repo) still be detected.
|
|
1139
|
+
const pyMarker = _exists(scanRoot, 'pytest.ini') || _exists(scanRoot, 'pyproject.toml') ||
|
|
1140
|
+
_exists(scanRoot, 'tox.ini') || _isDir(scanRoot, 'tests');
|
|
1141
|
+
if (pyMarker && _binaryAvailable('pytest')) {
|
|
1142
|
+
return { cmd: 'pytest', args: ['-q'], kind: 'pytest' };
|
|
1143
|
+
}
|
|
1144
|
+
|
|
1145
|
+
// Go
|
|
1146
|
+
if (_exists(scanRoot, 'go.mod')) {
|
|
1147
|
+
return { cmd: 'go', args: ['test', './...'], kind: 'go' };
|
|
1148
|
+
}
|
|
1149
|
+
|
|
1150
|
+
return null;
|
|
1151
|
+
}
|
|
1152
|
+
|
|
1153
|
+
// Run the detected test command with a walltime budget. Always returns a
|
|
1154
|
+
// result object — never throws. Distinguishes four outcomes:
|
|
1155
|
+
// - no detectable/runnable command -> status: 'skipped' (does NOT fail)
|
|
1156
|
+
// - ran and exited 0 -> status: 'passed'
|
|
1157
|
+
// - ran and exited non-zero -> status: 'failed'
|
|
1158
|
+
// - ran past the timeout budget -> status: 'failed', timedOut: true
|
|
1159
|
+
function runProjectTests(scanRoot, { timeoutMs = DEFAULT_TIMEOUT_MS } = {}) {
|
|
1160
|
+
const startedAt = Date.now();
|
|
1161
|
+
const command = detectTestCommand(scanRoot);
|
|
1162
|
+
if (!command) {
|
|
1163
|
+
return {
|
|
1164
|
+
status: 'skipped', passed: null, skipped: true,
|
|
1165
|
+
reason: 'no-test-command-detected', exitCode: null, timedOut: false,
|
|
1166
|
+
durationMs: Date.now() - startedAt,
|
|
1167
|
+
};
|
|
1168
|
+
}
|
|
1169
|
+
|
|
1170
|
+
let r;
|
|
1171
|
+
try {
|
|
1172
|
+
r = (0,node_child_process__WEBPACK_IMPORTED_MODULE_0__.spawnSync)(command.cmd, command.args, {
|
|
1173
|
+
cwd: scanRoot,
|
|
1174
|
+
encoding: 'utf8',
|
|
1175
|
+
timeout: timeoutMs,
|
|
1176
|
+
env: { ...process.env, CI: '1' },
|
|
1177
|
+
});
|
|
1178
|
+
} catch (e) {
|
|
1179
|
+
// The spawn call itself threw (rare — e.g. cwd vanished). Treat as
|
|
1180
|
+
// "could not run", not "ran and failed".
|
|
1181
|
+
return {
|
|
1182
|
+
status: 'skipped', passed: null, skipped: true,
|
|
1183
|
+
reason: `spawn-error: ${e.message}`, exitCode: null, timedOut: false,
|
|
1184
|
+
durationMs: Date.now() - startedAt,
|
|
1185
|
+
};
|
|
1186
|
+
}
|
|
1187
|
+
const durationMs = Date.now() - startedAt;
|
|
1188
|
+
|
|
1189
|
+
if (r.error && r.error.code === 'ENOENT') {
|
|
1190
|
+
// The detected tool isn't actually installed on this machine. Not a
|
|
1191
|
+
// test failure — the suite never ran.
|
|
1192
|
+
return {
|
|
1193
|
+
status: 'skipped', passed: null, skipped: true,
|
|
1194
|
+
reason: `${command.kind}-not-installed`, exitCode: null, timedOut: false, durationMs,
|
|
1195
|
+
};
|
|
1196
|
+
}
|
|
1197
|
+
|
|
1198
|
+
if (r.status === null) {
|
|
1199
|
+
// spawnSync sets status:null both on timeout-kill and on being killed by
|
|
1200
|
+
// another signal; either way the run did not complete, which is a
|
|
1201
|
+
// verification failure, never a skip — we asked for a result and the
|
|
1202
|
+
// process was terminated before producing one.
|
|
1203
|
+
return {
|
|
1204
|
+
status: 'failed', passed: false, skipped: false,
|
|
1205
|
+
reason: 'timed-out', exitCode: null, timedOut: true, durationMs,
|
|
1206
|
+
};
|
|
1207
|
+
}
|
|
1208
|
+
|
|
1209
|
+
return {
|
|
1210
|
+
status: r.status === 0 ? 'passed' : 'failed',
|
|
1211
|
+
passed: r.status === 0,
|
|
1212
|
+
skipped: false,
|
|
1213
|
+
reason: r.status === 0 ? null : 'test-failures',
|
|
1214
|
+
exitCode: r.status,
|
|
1215
|
+
timedOut: false,
|
|
1216
|
+
durationMs,
|
|
1217
|
+
};
|
|
1218
|
+
}
|
|
1219
|
+
|
|
1197
1220
|
|
|
1198
1221
|
/***/ })
|
|
1199
1222
|
|