crysk 0.6.1 → 0.8.0
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/README.md +28 -22
- package/dist/SKILL-1q3658n0.md +74 -0
- package/dist/SKILL-25j4q9b8.md +143 -0
- package/dist/SKILL-32cmme3j.md +143 -0
- package/dist/SKILL-4747bhne.md +143 -0
- package/dist/SKILL-4jc3sv23.md +74 -0
- package/dist/SKILL-c0492rch.md +540 -0
- package/dist/SKILL-d1nzkhhr.md +74 -0
- package/dist/SKILL-f8p7kvx0.md +143 -0
- package/dist/SKILL-gxd45g38.md +553 -0
- package/dist/SKILL-rec41991.md +66 -0
- package/dist/SKILL-zze8zj5g.md +536 -0
- package/dist/adjudicator-eckqh14e.md +78 -0
- package/dist/ci-admission-qn797a6c.md +128 -0
- package/dist/cli.js +18608 -18523
- package/dist/consolidate-4xw1pkg7.md +24 -0
- package/dist/deep-review-logic-dgr112y1.md +196 -0
- package/dist/deep-review-logic-fck5z4xe.md +197 -0
- package/dist/deep-review-logic-h1fpamfh.md +192 -0
- package/dist/deep-review-logic-m4vk5683.md +196 -0
- package/dist/deep-review-logic-z63kmgmc.md +157 -0
- package/dist/deep-review-parity-7x7s5d81.md +148 -0
- package/dist/deep-review-parity-9wxmynsc.md +115 -0
- package/dist/deep-review-parity-bwwe7cx6.md +152 -0
- package/dist/deep-review-parity-e50pw3pk.md +117 -0
- package/dist/deep-review-parity-ennmm9m1.md +151 -0
- package/{prompts/default/deep-review-parity.md → dist/deep-review-parity-h7y6y2he.md} +14 -16
- package/dist/deep-review-parity-xg6wy69c.md +151 -0
- package/dist/deep-review-reliability-6h3c4gmw.md +125 -0
- package/dist/deep-review-reliability-dcjab69x.md +122 -0
- package/dist/deep-review-reliability-ec7b5pbk.md +159 -0
- package/dist/deep-review-reliability-hvasvmt9.md +153 -0
- package/dist/deep-review-reliability-mjnayxmt.md +158 -0
- package/dist/deep-review-reliability-mm7fnayf.md +124 -0
- package/dist/deep-review-reliability-xk4xvc2w.md +158 -0
- package/dist/deep-review-resilience-0esd5qeh.md +148 -0
- package/dist/deep-review-resilience-envgcmpy.md +115 -0
- package/dist/deep-review-resilience-fgx8ythf.md +151 -0
- package/dist/deep-review-resilience-mq5qa6ad.md +152 -0
- package/dist/deep-review-resilience-pwh1gawp.md +151 -0
- package/dist/deep-review-resilience-s9n21dkx.md +117 -0
- package/dist/deep-review-resilience-vz4qskea.md +119 -0
- package/dist/deep-review-security-1vwjdc9q.md +205 -0
- package/dist/deep-review-security-x1cc1z00.md +205 -0
- package/dist/index.js +590 -485
- package/dist/opencode-ci-jc8mf4wn.md +106 -0
- package/dist/opencode-ci-qrcv7a99.md +106 -0
- package/dist/opencode-ci-rdpcc7xc.md +106 -0
- package/dist/opencode-ci-yt0k74t0.md +106 -0
- package/dist/review-refuter-7qffwtzz.md +77 -0
- package/dist/verifier-tyh8r45v.md +40 -0
- package/dist/verifier-xf4xj701.md +40 -0
- package/dist/workflow-andv2dvm.yml +158 -0
- package/package.json +1 -1
- package/prompts/consolidate.md +1 -1
- package/prompts/default/PROVENANCE.md +6 -0
- package/prompts/default/deep-review-logic.md +56 -15
- package/prompts/default/deep-review-reliability.md +64 -30
- package/prompts/default/deep-review-resilience.md +71 -37
- package/prompts/default/deep-review-security.md +205 -0
- package/prompts/default/review-refuter.md +1 -1
- package/prompts/verifier.md +40 -0
- package/skills/crysk/SKILL.md +6 -6
- package/skills/crysk-ci-setup/SKILL.md +32 -33
- package/skills/crysk-ci-setup/assets/workflow.yml +5 -10
- package/skills/crysk-ci-setup/references/ci-admission.md +5 -5
- package/skills/crysk-ci-setup/references/opencode-ci.md +6 -6
- package/skills/crysk-loop/SKILL.md +37 -54
- package/skills/crysk-loop/adjudicator.md +10 -18
- package/skills/crysk-pr/SKILL.md +1 -1
- package/src/asset-manifest.ts +6 -6
- package/src/assets.ts +4 -4
- package/src/brand.ts +20 -30
- package/src/ci/gates.ts +155 -192
- package/src/ci/outcome.ts +4 -31
- package/src/ci/platform-limits.ts +135 -0
- package/src/ci/platform-report.ts +212 -126
- package/src/ci/platform.ts +28 -26
- package/src/ci/reporter.ts +25 -26
- package/src/ci/review-admission.ts +1 -2
- package/src/ci/review-levels.ts +12 -53
- package/src/ci/setup.ts +96 -66
- package/src/ci/trace-bundle.ts +11 -27
- package/src/ci/trace-transcripts.ts +4 -6
- package/src/ci/trigger.ts +3 -5
- package/src/cli.ts +6 -5
- package/src/commands/ci-setup-questions.ts +2 -55
- package/src/commands/ci-setup.ts +2 -26
- package/src/commands/triage.ts +131 -38
- package/src/commands/usage.ts +0 -1
- package/src/compare/compare.ts +3 -3
- package/src/compare/floor-test.ts +30 -201
- package/src/compare/hunter-exam.ts +631 -0
- package/src/config/config.ts +6 -30
- package/src/diversity/accounting.ts +0 -26
- package/src/diversity/benchmark.ts +1 -17
- package/src/diversity/identity.ts +0 -14
- package/src/diversity/pipeline-integration.ts +0 -29
- package/src/doctor.ts +3 -6
- package/src/execution/claude-session-trace.ts +7 -2
- package/src/execution/contracts.ts +15 -9
- package/src/execution/harness.ts +165 -187
- package/src/execution/spend-limiter.ts +39 -131
- package/src/execution/usage-facts.ts +397 -0
- package/src/execution/usage-normalized.ts +33 -86
- package/src/harness/materiality-judge.ts +0 -5
- package/src/harness/pipeline-plan.ts +0 -3
- package/src/harness/replay-rows.ts +0 -5
- package/src/harness/route-runner.ts +2 -2
- package/src/harness/single-step.ts +3 -9
- package/src/index.ts +1 -1
- package/src/mcp/preflight.ts +2 -2
- package/src/model/catalog.ts +1 -3
- package/src/model/free-discovery.ts +11 -78
- package/src/model/model-name.ts +16 -5
- package/src/model/profiles/claude.ts +1 -2
- package/src/model/profiles/openai.ts +1 -2
- package/src/model/profiles.ts +2 -3
- package/src/model/provider-capabilities.ts +16 -55
- package/src/model/routing.ts +39 -36
- package/src/pr/admission.ts +6 -6
- package/src/pr/answered-findings.ts +8 -4
- package/src/pr/ci-publish.ts +0 -2
- package/src/pr/discovery.ts +4 -25
- package/src/pr/plan.ts +40 -121
- package/src/pr/pr-status.ts +14 -2
- package/src/pr/pr.ts +1 -1
- package/src/pr/preflight.ts +44 -15
- package/src/pr/publish-outcome.ts +1 -19
- package/src/pr/review-pr.ts +15 -47
- package/src/pr/target.ts +76 -82
- package/src/production-runtime.ts +52 -114
- package/src/progress.ts +0 -54
- package/src/rereview/prepare.ts +4 -1
- package/src/rereview/state.ts +1 -1
- package/src/rereview/verify.ts +15 -9
- package/src/review/boundary.ts +5 -8
- package/src/review/drafts.ts +3 -3
- package/src/review/findings-conformance.ts +1 -1
- package/src/review/findings.ts +0 -4
- package/src/review/pipeline.ts +194 -407
- package/src/review/preflight.ts +175 -201
- package/src/review/prompt-set.ts +1 -2
- package/src/review/report.ts +2 -162
- package/src/review/review.ts +25 -74
- package/src/review/route-preflight.ts +19 -35
- package/src/review/run.ts +37 -73
- package/src/review/size-gate.ts +3 -3
- package/src/review/spec.ts +6 -11
- package/src/review/step-runner.ts +4 -4
- package/src/runner-authority.ts +10 -9
- package/src/security/credential-broker.ts +5 -6
- package/src/server/preflight.ts +0 -1
- package/src/server/server.ts +17 -6
- package/src/store/activity.ts +1 -2
- package/src/store/backfill-preflight.ts +0 -3
- package/src/store/backfill.ts +0 -1
- package/src/store/metrics-preflight.ts +4 -14
- package/src/store/metrics.ts +5 -5
- package/src/store/preflight.ts +0 -13
- package/src/store/store.ts +11 -12
- package/src/transport-registry.ts +6 -30
- package/src/transports/claude-code-cli.ts +149 -215
- package/src/transports/opencode-client.ts +4 -24
- package/src/transports/opencode-sdk.ts +17 -112
- package/src/triage/plain-reply.ts +37 -0
- package/src/ui/activity.ts +1 -6
- package/src/ui/config-edit.ts +14 -67
- package/src/ui/config.ts +3 -8
- package/src/ui/plan.ts +11 -198
- package/src/ui/progress.ts +3 -49
- package/src/ui/result.ts +18 -77
- package/src/ui/review-menu.ts +3 -12
- package/src/ui/select.ts +9 -79
- package/src/updater.ts +1 -1
- package/src/usage.ts +1 -7
- package/prompts/scout.md +0 -140
- package/src/ci/platform-estimate.ts +0 -111
- package/src/review/scout.ts +0 -323
package/src/review/prompt-set.ts
CHANGED
|
@@ -18,7 +18,6 @@ import { wrapBlock } from "./boundary";
|
|
|
18
18
|
export const AGENT_NAMES = [
|
|
19
19
|
"deep-review-reliability",
|
|
20
20
|
"deep-review-resilience",
|
|
21
|
-
"deep-review-parity",
|
|
22
21
|
"review-refuter",
|
|
23
22
|
] as const;
|
|
24
23
|
|
|
@@ -28,9 +27,9 @@ export const AGENT_NAMES = [
|
|
|
28
27
|
// (model-profiles S4).
|
|
29
28
|
export const ENGINE_PROMPT_FILES = [
|
|
30
29
|
"consolidate.md",
|
|
31
|
-
"scout.md",
|
|
32
30
|
"summarizer.md",
|
|
33
31
|
"materiality.md",
|
|
32
|
+
"verifier.md",
|
|
34
33
|
] as const;
|
|
35
34
|
export type EnginePromptFile = (typeof ENGINE_PROMPT_FILES)[number];
|
|
36
35
|
|
package/src/review/report.ts
CHANGED
|
@@ -1,13 +1,11 @@
|
|
|
1
1
|
// The human-facing half of local mode (ROADMAP B0): a pure renderer that
|
|
2
|
-
// turns a findings document into the markdown a developer actually reads
|
|
3
|
-
// a pure cost estimator that runs BEFORE the money is spent.
|
|
2
|
+
// turns a findings document into the markdown a developer actually reads.
|
|
4
3
|
//
|
|
5
4
|
// Pure on purpose — no I/O, no git, no clock. The CLI owns every side effect,
|
|
6
5
|
// so the report can be re-rendered from an artifact on disk months later and
|
|
7
6
|
// come out byte-identical (the lab already replays old findings.json files;
|
|
8
7
|
// a renderer that reached for `new Date()` would make that a lie).
|
|
9
8
|
|
|
10
|
-
import type { PlatformEstimateBasis } from "#ci/platform-estimate";
|
|
11
9
|
import { listPaths } from "#git/refs";
|
|
12
10
|
import type { ResolvedRoutePlan, ResolvedStepRoute } from "#model/routing";
|
|
13
11
|
import type { PostMinSeverity } from "#pr/post-floor";
|
|
@@ -51,133 +49,6 @@ export interface DiffStat {
|
|
|
51
49
|
deletions: number;
|
|
52
50
|
}
|
|
53
51
|
|
|
54
|
-
export interface CostEstimate {
|
|
55
|
-
low: number;
|
|
56
|
-
high: number;
|
|
57
|
-
// Plain-language statement of how coarse this is, meant to be printed next
|
|
58
|
-
// to the numbers so nobody reads the band as a quote.
|
|
59
|
-
basis: string;
|
|
60
|
-
// Set when the range is the Crysk platform's estimate, not this band: what
|
|
61
|
-
// the platform priced it from (src/ci/platform-estimate.ts).
|
|
62
|
-
platform?: PlatformEstimateBasis;
|
|
63
|
-
}
|
|
64
|
-
|
|
65
|
-
// Per-hunter marginal cost, fitted to the measured runs (see BAND below).
|
|
66
|
-
// Deliberately three transparent coefficients instead of a regression: with
|
|
67
|
-
// this few calibration points anything fancier is false precision.
|
|
68
|
-
//
|
|
69
|
-
// The base is the load-bearing coefficient, recalibrated 2026-08-11 from
|
|
70
|
-
// 0.5 to 1.2: a hunter's floor is reading the TREE and the prompt
|
|
71
|
-
// machinery, nearly independent of a small diff. The proof was two PR-mode
|
|
72
|
-
// runs of the same 7-file / +21 −8 tree billing $4.74 and $3.92 against a
|
|
73
|
-
// $1.19–$2.39 band — both ~2x above the top, the exact under-estimate
|
|
74
|
-
// failure this band exists to prevent.
|
|
75
|
-
const USD_PER_AGENT_BASE = 1.2;
|
|
76
|
-
const USD_PER_CHANGED_LINE = 0.00021;
|
|
77
|
-
const USD_PER_FILE = 0.008;
|
|
78
|
-
|
|
79
|
-
// Skewed on purpose. Every recorded overrun was an UNDER-estimate, never
|
|
80
|
-
// over, so the upper arm is the wider one: a band that is too generous costs
|
|
81
|
-
// a second of hesitation, a band that is too tight costs real money.
|
|
82
|
-
const BAND_LOW = 0.65;
|
|
83
|
-
const BAND_HIGH = 1.4;
|
|
84
|
-
|
|
85
|
-
// WHY this function exists at all: the ROADMAP logs two cost overruns in a
|
|
86
|
-
// row — "~$21" against an actual bill, and "~$34" against an actual $48.30 —
|
|
87
|
-
// and both had the same cause. The estimate was carried over from the
|
|
88
|
-
// PREVIOUS arm instead of being computed from the tree about to be reviewed.
|
|
89
|
-
// So this estimates from the diff, and only from the diff.
|
|
90
|
-
//
|
|
91
|
-
// Calibration points, all measured on real runs of the current engine:
|
|
92
|
-
// - PR 1682's tree (7 files / +21 −8, 3 hunters + refuter) billed $4.74
|
|
93
|
-
// and $3.92 across two runs — the per-agent floor evidence;
|
|
94
|
-
// - a 45-file / +2775 −1237 tree with 5 hunters + refuter billed ~$14.78,
|
|
95
|
-
// and the same tree was recorded at ~$11/run elsewhere in the campaign;
|
|
96
|
-
// - the first B0 local run (5 files / +484, 4 hunters + refuter), $3.68.
|
|
97
|
-
// A legacy "~$2.61 small tree" point from before PR mode existed was
|
|
98
|
-
// RETIRED by the 1682 evidence: two same-engine runs of a comparable tree
|
|
99
|
-
// both landed far above it, and a calibration point the instrument itself
|
|
100
|
-
// contradicts is decoration. The ~$11 vs ~$14.78 spread on ONE tree is the
|
|
101
|
-
// whole argument for returning a band: the same diff, the same agents, a
|
|
102
|
-
// 34% swing. A point estimate here would be a fiction with a decimal point.
|
|
103
|
-
//
|
|
104
|
-
// The refuter is not a separate term: its cost rides in the coefficients,
|
|
105
|
-
// because every calibration run had the refuter leg included and a
|
|
106
|
-
// refuter's size tracks how much the hunters found, which tracks diff size.
|
|
107
|
-
export function estimateCost(
|
|
108
|
-
diffStat: DiffStat,
|
|
109
|
-
hunterCount: number,
|
|
110
|
-
summarizerEnabled = false,
|
|
111
|
-
// ROADMAP-DOORDASH M5. Defaulted OFF so every existing caller keeps its
|
|
112
|
-
// band byte-identical, and so a caller that forgets it under-quotes
|
|
113
|
-
// nothing it did not already under-quote.
|
|
114
|
-
scoutEnabled = false,
|
|
115
|
-
// Item 7 O-5a. Default 0 so every existing caller keeps its band
|
|
116
|
-
// byte-identical. Priced as its own term (not folded into `agents`) so a
|
|
117
|
-
// 40-finding re-review of a 2-file delta is not billed as 40 hunters.
|
|
118
|
-
verificationSteps = 0,
|
|
119
|
-
// The post-refuter consolidation pass. LAST and defaulted OFF so every
|
|
120
|
-
// existing call shape compiles and keeps its band; the product paths pass
|
|
121
|
-
// true. One seat, the scout's rule: a claims-only sonnet call costs far
|
|
122
|
-
// less than a hunter, and over-quoting is the cheap direction to be wrong.
|
|
123
|
-
consolidateEnabled = false,
|
|
124
|
-
): CostEstimate {
|
|
125
|
-
const files = Math.max(0, diffStat.files);
|
|
126
|
-
const lines =
|
|
127
|
-
Math.max(0, diffStat.insertions) + Math.max(0, diffStat.deletions);
|
|
128
|
-
// A validated ReviewSpec always carries at least one hunter, so a zero here
|
|
129
|
-
// means a caller asked hypothetically. Clamping keeps the band from
|
|
130
|
-
// collapsing to $0–$0, which would read as "this is free".
|
|
131
|
-
// The scout takes a full agent seat, the summarizer's precedent, and that
|
|
132
|
-
// deliberately OVER-quotes: it is diff-only, one attempt, and reads no tree,
|
|
133
|
-
// so its true cost is a fraction of a hunter's. Two reasons it gets the seat
|
|
134
|
-
// anyway. First, this band's own rule — every recorded overrun was an
|
|
135
|
-
// UNDER-estimate, so the generous direction is the cheap one to be wrong in.
|
|
136
|
-
// Second, the scout has NO calibration point of its own: M4 billed $37.00
|
|
137
|
-
// across six prompt versions, two probe defects and a reverted experiment,
|
|
138
|
-
// which is a milestone total and not a per-spawn cost. M6 records cost per
|
|
139
|
-
// arm; that is the number this coefficient gets recalibrated from, and until
|
|
140
|
-
// it exists an invented per-scout coefficient would be false precision.
|
|
141
|
-
const hunterSeats = Math.max(
|
|
142
|
-
0,
|
|
143
|
-
hunterCount +
|
|
144
|
-
(summarizerEnabled ? 1 : 0) +
|
|
145
|
-
(scoutEnabled ? 1 : 0) +
|
|
146
|
-
(consolidateEnabled ? 1 : 0),
|
|
147
|
-
);
|
|
148
|
-
const steps = Math.max(0, verificationSteps);
|
|
149
|
-
// A zero-hunter, zero-verify call still needs a band (the hypothetical
|
|
150
|
-
// floor). A skip-discovery re-review with N verify steps must NOT inherit
|
|
151
|
-
// that floor — that would bill a phantom hunter on top of verification.
|
|
152
|
-
const agents = hunterSeats === 0 && steps === 0 ? 1 : hunterSeats;
|
|
153
|
-
// Verification is its OWN term (item 7 O-5a), not folded into `agents`:
|
|
154
|
-
// a verify step reads one prior finding, not the whole tree, so multiplying
|
|
155
|
-
// it by changed-files would price a 40-finding re-review of a 2-file delta
|
|
156
|
-
// as if it were 40 hunters. The seat is the same per-agent floor — generous
|
|
157
|
-
// on purpose, same direction as the scout.
|
|
158
|
-
const mid =
|
|
159
|
-
agents *
|
|
160
|
-
(USD_PER_AGENT_BASE +
|
|
161
|
-
USD_PER_CHANGED_LINE * lines +
|
|
162
|
-
USD_PER_FILE * files) +
|
|
163
|
-
steps * USD_PER_AGENT_BASE;
|
|
164
|
-
return {
|
|
165
|
-
low: round2(mid * BAND_LOW),
|
|
166
|
-
high: round2(mid * BAND_HIGH),
|
|
167
|
-
basis:
|
|
168
|
-
"coarse band from measured runs (a 7-file / +21 −8 tree with 3 " +
|
|
169
|
-
"hunters + refuter, $3.92–$4.74 across two runs; a 45-file / " +
|
|
170
|
-
"+2775 −1237 tree with 5 hunters + refuter, ~$11–$14.78): a " +
|
|
171
|
-
"per-agent floor for hunters + refuter" +
|
|
172
|
-
`${summarizerEnabled ? " + summarizer" : ""}` +
|
|
173
|
-
`${scoutEnabled ? " + scout" : ""}` +
|
|
174
|
-
`${consolidateEnabled ? " + consolidation" : ""}` +
|
|
175
|
-
`${steps > 0 ? ` + ${steps} verification step(s)` : ""} ` +
|
|
176
|
-
"plus changed lines and files. An order-of-magnitude " +
|
|
177
|
-
"guide, not a quote — the same tree has billed 34% apart across runs.",
|
|
178
|
-
};
|
|
179
|
-
}
|
|
180
|
-
|
|
181
52
|
export interface ReportMeta {
|
|
182
53
|
repo: string;
|
|
183
54
|
base: string;
|
|
@@ -190,15 +61,6 @@ export interface ReportMeta {
|
|
|
190
61
|
// Paths the exclusion filter dropped from the reviewed diff, if any. Stated
|
|
191
62
|
// in the report because it is a mutation of the input, not a detail.
|
|
192
63
|
excludedPaths?: string[];
|
|
193
|
-
// CASH, and only cash — §8's "Budget enforcement uses cash only". A Claude
|
|
194
|
-
// subscription route reports a truthful $0.00 here.
|
|
195
|
-
costUsd: number;
|
|
196
|
-
// #173 (§8: "Artifacts and reports show cash and notional totals
|
|
197
|
-
// separately"). The list-basis figure the run was NOT charged. Absent means
|
|
198
|
-
// no segment at all, never a `$0.00` one — a runner that reported no
|
|
199
|
-
// normalized usage has no list figure, and printing one it never produced
|
|
200
|
-
// is the fabrication the usage slice exists to prevent.
|
|
201
|
-
notionalUsd?: number;
|
|
202
64
|
wallMs: number;
|
|
203
65
|
// S4: the mode the commit status is set under, which the closing note names.
|
|
204
66
|
// Absent means advisory: a local run, and any caller that never passes it,
|
|
@@ -1596,17 +1458,9 @@ export function blobUrl(
|
|
|
1596
1458
|
|
|
1597
1459
|
function runLines(doc: FindingsDocument, meta: ReportMeta): string[] {
|
|
1598
1460
|
const { files, insertions, deletions } = meta.diffStat;
|
|
1599
|
-
// #173: cash first and unqualified, notional after it and never without its
|
|
1600
|
-
// basis spelled out — §8 allows notional in a report "only when clearly
|
|
1601
|
-
// labelled", and two bare dollar figures side by side is precisely the
|
|
1602
|
-
// mixing it forbids.
|
|
1603
|
-
const notional =
|
|
1604
|
-
meta.notionalUsd === undefined
|
|
1605
|
-
? ""
|
|
1606
|
-
: ` (${usd(meta.notionalUsd)} at list price, not charged)`;
|
|
1607
1461
|
const lines = [
|
|
1608
1462
|
`${files} file${files === 1 ? "" : "s"}, +${insertions} −${deletions}` +
|
|
1609
|
-
` · run ${doc.run_status} · ${
|
|
1463
|
+
` · run ${doc.run_status} · ${duration(meta.wallMs)}`,
|
|
1610
1464
|
];
|
|
1611
1465
|
const excluded = meta.excludedPaths ?? [];
|
|
1612
1466
|
if (excluded.length > 0) {
|
|
@@ -1625,12 +1479,6 @@ function runLines(doc: FindingsDocument, meta: ReportMeta): string[] {
|
|
|
1625
1479
|
`Agents: ${rows.map((r) => `${r.key} ${r.status}`).join(" · ")}`,
|
|
1626
1480
|
);
|
|
1627
1481
|
}
|
|
1628
|
-
if (!doc.parity_hunter_fired) {
|
|
1629
|
-
lines.push("");
|
|
1630
|
-
lines.push(
|
|
1631
|
-
"The parity hunter did not fire: no changed path matched its trigger.",
|
|
1632
|
-
);
|
|
1633
|
-
}
|
|
1634
1482
|
return lines;
|
|
1635
1483
|
}
|
|
1636
1484
|
|
|
@@ -1838,10 +1686,6 @@ function oneLine(text: string): string {
|
|
|
1838
1686
|
return publishedText(text.replace(/\s+/g, " ").trim());
|
|
1839
1687
|
}
|
|
1840
1688
|
|
|
1841
|
-
function usd(amount: number): string {
|
|
1842
|
-
return `$${amount.toFixed(2)}`;
|
|
1843
|
-
}
|
|
1844
|
-
|
|
1845
1689
|
// Elapsed time as the compact `3m12s` the CLI's live progress prefixes onto
|
|
1846
1690
|
// every event line. Floor, never round: 59.9s read as "60s" would tick the
|
|
1847
1691
|
// minute early. Minutes deliberately never roll into hours — a review past
|
|
@@ -1864,10 +1708,6 @@ function duration(ms: number): string {
|
|
|
1864
1708
|
: `${minutes}m ${String(seconds).padStart(2, "0")}s`;
|
|
1865
1709
|
}
|
|
1866
1710
|
|
|
1867
|
-
function round2(value: number): number {
|
|
1868
|
-
return Math.round(value * 100) / 100;
|
|
1869
|
-
}
|
|
1870
|
-
|
|
1871
1711
|
// A step's route as the plan card shows it: the model name it resolved to,
|
|
1872
1712
|
// which already says its backend (`claude:sonnet-latest`,
|
|
1873
1713
|
// `opencode:openai/gpt-5`), and its `models` entry's effort when it sets one,
|
package/src/review/review.ts
CHANGED
|
@@ -10,7 +10,6 @@ import os from "node:os";
|
|
|
10
10
|
import path from "node:path";
|
|
11
11
|
import {
|
|
12
12
|
ingestReviewMetrics,
|
|
13
|
-
notionalCostInput,
|
|
14
13
|
persistCanonicalReview,
|
|
15
14
|
pipelineConfigInput,
|
|
16
15
|
} from "#config/config";
|
|
@@ -23,8 +22,6 @@ import {
|
|
|
23
22
|
resolveRepoRoot,
|
|
24
23
|
} from "#git/git";
|
|
25
24
|
import { engineIdentity } from "#git/identity";
|
|
26
|
-
import { countFreeRouteSteps } from "#model/free-discovery";
|
|
27
|
-
import { planRoutes } from "#model/routing";
|
|
28
25
|
import { reviewPr } from "#pr/review-pr";
|
|
29
26
|
import { reviewExitCode } from "#review/exit-code";
|
|
30
27
|
import type { Telemetry } from "#review/findings";
|
|
@@ -38,12 +35,10 @@ import {
|
|
|
38
35
|
resolveMaxConcurrentSteps,
|
|
39
36
|
resolveSharedPrefix,
|
|
40
37
|
} from "#review/preflight";
|
|
41
|
-
import { estimateCost } from "#review/report";
|
|
42
38
|
import {
|
|
43
39
|
isBundledAgentsDir,
|
|
44
40
|
pipelineConsolidateInput,
|
|
45
41
|
pipelineMaterialityInput,
|
|
46
|
-
pipelineScoutInput,
|
|
47
42
|
pipelineSummarizerInput,
|
|
48
43
|
} from "#review/route-preflight";
|
|
49
44
|
import {
|
|
@@ -54,7 +49,6 @@ import {
|
|
|
54
49
|
loadRunConfig,
|
|
55
50
|
NO_INSTRUCTIONS_NOTICE,
|
|
56
51
|
prepareRunnerForRoute,
|
|
57
|
-
resolveParityFires,
|
|
58
52
|
resolvePipelineRoute,
|
|
59
53
|
resolvePromptSet,
|
|
60
54
|
resolveReviewInstructions,
|
|
@@ -78,7 +72,6 @@ import {
|
|
|
78
72
|
import { log, styleEnabled } from "#ui/primitives";
|
|
79
73
|
import { applySizeGate, confirm, startProgressRenderer } from "#ui/progress";
|
|
80
74
|
import { renderResult } from "#ui/result";
|
|
81
|
-
import { freeBillingMode } from "#ui/select";
|
|
82
75
|
import { CliError } from "../errors";
|
|
83
76
|
import { prheroLayout } from "../home-preflight";
|
|
84
77
|
import { readLocalIgnoreRules } from "../ignore-read";
|
|
@@ -103,15 +96,13 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
103
96
|
// Two layers since C5, folded by loadEffectiveConfig into the one
|
|
104
97
|
// LocalConfig everything below already expected. The --config missing-file
|
|
105
98
|
// error and the boundary rules live there.
|
|
106
|
-
const { loaded, config, summary,
|
|
107
|
-
|
|
108
|
-
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
|
|
112
|
-
|
|
113
|
-
);
|
|
114
|
-
options = { ...options, scout, post };
|
|
99
|
+
const { loaded, config, summary, post, profile } = await loadRunConfig({
|
|
100
|
+
root: repoRoot,
|
|
101
|
+
home: os.homedir(),
|
|
102
|
+
configFlag: options.config,
|
|
103
|
+
options,
|
|
104
|
+
});
|
|
105
|
+
options = { ...options, post };
|
|
115
106
|
// S5: the Claude credential this run uses, chosen once, here, from
|
|
116
107
|
// `credentials.claude` and what is set. Every runner, binding and child below
|
|
117
108
|
// reads its env, which holds that one credential and no other.
|
|
@@ -197,8 +188,8 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
197
188
|
const { runDir, repoId } = await createRunDir(options, repoRoot, headSha);
|
|
198
189
|
const diffPath = path.join(runDir, "diff.patch");
|
|
199
190
|
// diffFromSha, never baseSha: see resolveDiffFrom. The numstat below uses
|
|
200
|
-
// the same endpoint on purpose — a
|
|
201
|
-
//
|
|
191
|
+
// the same endpoint on purpose — a stat computed over a wider range than
|
|
192
|
+
// the one being reviewed would describe a different review.
|
|
202
193
|
const diff = await git(repoRoot, ["diff", `${diffFromSha}..${headSha}`]);
|
|
203
194
|
if (!diff.ok) throw new CliError(`git diff failed: ${diff.stderr}`);
|
|
204
195
|
if (diff.stdout.trim().length === 0) {
|
|
@@ -238,23 +229,9 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
238
229
|
const codegraphAvailable = existsSync(path.join(repoRoot, ".codegraph"));
|
|
239
230
|
await writeMcpConfig(mcpConfigPath, codegraphAvailable);
|
|
240
231
|
|
|
241
|
-
// 11 — the plan.
|
|
242
|
-
|
|
243
|
-
// fire on a path no hunter was given.
|
|
244
|
-
const parityFires = resolveParityFires(
|
|
245
|
-
effectiveDiff.patch,
|
|
246
|
-
config.parity_trigger_paths,
|
|
247
|
-
);
|
|
248
|
-
const activeHunters = selectActiveHunters(spec.agents, parityFires);
|
|
232
|
+
// 11 — the plan.
|
|
233
|
+
const activeHunters = selectActiveHunters(spec.agents);
|
|
249
234
|
const hunterCount = activeHunters.length;
|
|
250
|
-
const estimate = estimateCost(
|
|
251
|
-
diffStat,
|
|
252
|
-
hunterCount,
|
|
253
|
-
summary.enabled,
|
|
254
|
-
options.scout,
|
|
255
|
-
0,
|
|
256
|
-
true,
|
|
257
|
-
);
|
|
258
235
|
const productionRoute = await resolvePipelineRoute({
|
|
259
236
|
modelsConfigured: routingConfigured(config),
|
|
260
237
|
workspaceRoot: repoRoot,
|
|
@@ -270,19 +247,6 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
270
247
|
});
|
|
271
248
|
const routePlan = productionRoute?.routePlan;
|
|
272
249
|
const productionAdmission = productionRoute?.productionAdmission;
|
|
273
|
-
// Read off the admission above, which already decided every free verdict:
|
|
274
|
-
// this adds no probe. Absent without routing, so the band stays.
|
|
275
|
-
const freeRoute =
|
|
276
|
-
routePlan === undefined || productionAdmission === undefined
|
|
277
|
-
? undefined
|
|
278
|
-
: await countFreeRouteSteps({
|
|
279
|
-
steps: planRoutes(routePlan),
|
|
280
|
-
probe: productionAdmission.freeModelProbe,
|
|
281
|
-
// The CRYSK_OPENCODE_* names admission bound each route's kind with,
|
|
282
|
-
// so the card counts free exactly the routes admission probed (U3).
|
|
283
|
-
openCodeSecretNames:
|
|
284
|
-
productionAdmission.authorityOptions.openCodeSecretNames,
|
|
285
|
-
});
|
|
286
250
|
// Named rather than inlined into renderPlan: the same context is what the
|
|
287
251
|
// confirm menu's "Show details" renders, and building it twice would risk
|
|
288
252
|
// the card and the details view disagreeing about the run they describe.
|
|
@@ -301,17 +265,13 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
301
265
|
runDir,
|
|
302
266
|
config,
|
|
303
267
|
summary,
|
|
304
|
-
parityFires,
|
|
305
268
|
codegraphAvailable,
|
|
306
|
-
estimate,
|
|
307
|
-
hunterCount,
|
|
308
269
|
// Evaluated above (step 9) and ENFORCED below at step 13, unchanged;
|
|
309
270
|
// the plan only prints the verdict, last, where the decision is made.
|
|
310
271
|
sizeGate,
|
|
311
272
|
droppedPaths: effectiveDiff.droppedPaths,
|
|
312
273
|
configProvenance: configProvenanceOf(loaded, agentsDirSource),
|
|
313
274
|
...(routePlan === undefined ? {} : { routePlan }),
|
|
314
|
-
...(freeRoute === undefined ? {} : { freeRoute }),
|
|
315
275
|
};
|
|
316
276
|
for (const line of renderPlan(planContext, styleEnabled())) log(line);
|
|
317
277
|
|
|
@@ -344,7 +304,7 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
344
304
|
productionAdmission,
|
|
345
305
|
});
|
|
346
306
|
|
|
347
|
-
// 13 — the size gate, BEFORE the
|
|
307
|
+
// 13 — the size gate, BEFORE the confirm prompt for the unattended
|
|
348
308
|
// path. The watcher spawns with --yes, so a gate that lived only inside
|
|
349
309
|
// that confirmation would never fire in the one place — unattended spend
|
|
350
310
|
// — it exists to protect. An interactive TTY is offered Continue/Cancel
|
|
@@ -356,12 +316,8 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
356
316
|
// without --pr), so the "don't post" option is not offered and the choice's
|
|
357
317
|
// `post` field carries nothing local mode could act on.
|
|
358
318
|
if (!options.yes) {
|
|
359
|
-
const choice = await confirm(
|
|
360
|
-
|
|
361
|
-
estimate.high,
|
|
362
|
-
false,
|
|
363
|
-
() => planDetails(planContext, styleEnabled()),
|
|
364
|
-
freeRoute,
|
|
319
|
+
const choice = await confirm(false, () =>
|
|
320
|
+
planDetails(planContext, styleEnabled()),
|
|
365
321
|
);
|
|
366
322
|
if (choice.kind === "cancel") {
|
|
367
323
|
log("aborted; nothing was spent.");
|
|
@@ -371,7 +327,7 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
371
327
|
|
|
372
328
|
// 15 — run, with live progress: the expectation line up front, then one
|
|
373
329
|
// stderr line per pipeline event (plus a TTY heartbeat between them).
|
|
374
|
-
log(reviewingLine(hunterCount, summary
|
|
330
|
+
log(reviewingLine(hunterCount, summary));
|
|
375
331
|
const started = performance.now();
|
|
376
332
|
const progress = startProgressRenderer(
|
|
377
333
|
started,
|
|
@@ -392,16 +348,16 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
392
348
|
// same signal and refuses to start another attempt. Two controllers would
|
|
393
349
|
// leave the ceiling unable to stop the steps it is waiting on.
|
|
394
350
|
const ceilingController = new AbortController();
|
|
395
|
-
// Local mode has no PR, so the
|
|
396
|
-
//
|
|
397
|
-
//
|
|
398
|
-
//
|
|
351
|
+
// Local mode has no PR, so the reviewed head's commit subject stands in as
|
|
352
|
+
// the title: the hunters read it as the change's stated intent (S10) and
|
|
353
|
+
// the materiality judge as its title (S9). Local mode has no description.
|
|
354
|
+
// Asked for on every run, so the hunters' request does not depend on the
|
|
355
|
+
// judge being on. A failed read leaves it absent: the hunters get no intent
|
|
356
|
+
// section and the judge's pipeline renders its fallback.
|
|
399
357
|
let prTitle: string | undefined;
|
|
400
|
-
|
|
401
|
-
|
|
402
|
-
|
|
403
|
-
prTitle = subject.stdout.trim();
|
|
404
|
-
}
|
|
358
|
+
const subject = await git(repoRoot, ["log", "-1", "--format=%s", headSha]);
|
|
359
|
+
if (subject.ok && subject.stdout.trim() !== "") {
|
|
360
|
+
prTitle = subject.stdout.trim();
|
|
405
361
|
}
|
|
406
362
|
let productionRuntime: ProductionRuntime | undefined;
|
|
407
363
|
try {
|
|
@@ -440,11 +396,10 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
440
396
|
mcpConfigPath,
|
|
441
397
|
hopBudget: options.hopBudget,
|
|
442
398
|
...(options.model ? { model: options.model } : {}),
|
|
443
|
-
parityTriggerPaths:
|
|
399
|
+
parityTriggerPaths: [],
|
|
444
400
|
suspicionPriors: config.suspicion_priors,
|
|
445
401
|
modelProfile: profile,
|
|
446
402
|
...pipelineSummarizerInput(summary, { profile }),
|
|
447
|
-
...pipelineScoutInput(options, { profile }),
|
|
448
403
|
...pipelineConsolidateInput({ profile }),
|
|
449
404
|
...pipelineMaterialityInput(config, { profile }),
|
|
450
405
|
...(prTitle === undefined ? {} : { prTitle }),
|
|
@@ -524,11 +479,7 @@ export async function review(options: CliOptions): Promise<number> {
|
|
|
524
479
|
const links = await localResultLinks(repoRoot, headSha);
|
|
525
480
|
for (const line of renderResult({
|
|
526
481
|
doc,
|
|
527
|
-
costUsd: result.usage.cost_usd_est,
|
|
528
|
-
...notionalCostInput(result),
|
|
529
|
-
...(freeRoute === undefined ? {} : { billing: freeBillingMode(freeRoute) }),
|
|
530
482
|
wallMs,
|
|
531
|
-
estimate: { low: estimate.low, high: estimate.high },
|
|
532
483
|
runDir,
|
|
533
484
|
artifacts: [path.basename(reportPath), path.basename(findingsPath)],
|
|
534
485
|
...(links === undefined ? {} : { links }),
|
|
@@ -88,12 +88,12 @@ export function enginePromptPath(
|
|
|
88
88
|
switch (file) {
|
|
89
89
|
case "consolidate.md":
|
|
90
90
|
return assets.consolidatePromptPath;
|
|
91
|
-
case "scout.md":
|
|
92
|
-
return assets.scoutPromptPath;
|
|
93
91
|
case "summarizer.md":
|
|
94
92
|
return assets.summarizerPromptPath;
|
|
95
93
|
case "materiality.md":
|
|
96
94
|
return assets.materialityPromptPath;
|
|
95
|
+
case "verifier.md":
|
|
96
|
+
return assets.verifierPromptPath;
|
|
97
97
|
}
|
|
98
98
|
}
|
|
99
99
|
|
|
@@ -119,32 +119,7 @@ export function pipelineSummarizerInput(
|
|
|
119
119
|
: {};
|
|
120
120
|
}
|
|
121
121
|
|
|
122
|
-
// The
|
|
123
|
-
// purpose and twice over (§3.7): a `review-scout.md` dropped in the agents dir
|
|
124
|
-
// without a spec entry is a hard CliError, and a new prompt-set directory
|
|
125
|
-
// holding byte-identical hunter files would be a new fingerprint — which is
|
|
126
|
-
// exactly the one-variable property M6 needs to be true by construction rather
|
|
127
|
-
// than argued. `prompts/` is the door the summarizer already walked through.
|
|
128
|
-
export function pipelineScoutInput(
|
|
129
|
-
options: Pick<CliOptions, "scout" | "scoutModel">,
|
|
130
|
-
prompts: RunPromptOptions,
|
|
131
|
-
):
|
|
132
|
-
| { scout: { promptPath: string; model?: string; bundled: true } }
|
|
133
|
-
| Record<string, never> {
|
|
134
|
-
return options.scout
|
|
135
|
-
? {
|
|
136
|
-
scout: {
|
|
137
|
-
promptPath: enginePromptPath("scout.md", prompts),
|
|
138
|
-
...(options.scoutModel === undefined
|
|
139
|
-
? {}
|
|
140
|
-
: { model: options.scoutModel }),
|
|
141
|
-
bundled: true,
|
|
142
|
-
},
|
|
143
|
-
}
|
|
144
|
-
: {};
|
|
145
|
-
}
|
|
146
|
-
|
|
147
|
-
// The consolidation pass is ENGINE-owned like the scout, and on by default in
|
|
122
|
+
// The consolidation pass is ENGINE-owned like the summarizer, and on by default in
|
|
148
123
|
// both product paths (local and PR/CI) with no flag: it was measured as a net
|
|
149
124
|
// gain on the Martian 50 (lab bench/METRICS.md 2026-10-27), and a failed pass
|
|
150
125
|
// falls back to the unconsolidated findings. The lab's direct `runPipeline`
|
|
@@ -160,6 +135,20 @@ export function pipelineConsolidateInput(prompts: RunPromptOptions): {
|
|
|
160
135
|
};
|
|
161
136
|
}
|
|
162
137
|
|
|
138
|
+
// The re-review verifier's body is ENGINE-owned like the consolidation
|
|
139
|
+
// prompt. The verifier asks whether an earlier finding's defect is still
|
|
140
|
+
// present at the current head, which is not the refuter's question, so it does
|
|
141
|
+
// not run under the refuter's body. Only the body is its own: model, effort,
|
|
142
|
+
// tools and route stay the refuter agent's (`runVerify`), so a custom
|
|
143
|
+
// `review-refuter.md` still sets those and no longer shapes what the
|
|
144
|
+
// verifier is told. Outside the agents dir on purpose: a `review-verifier.md`
|
|
145
|
+
// there would fall under the two-way rule and break every custom directory.
|
|
146
|
+
export function pipelineVerifierInput(prompts: RunPromptOptions): {
|
|
147
|
+
verifier: { promptPath: string };
|
|
148
|
+
} {
|
|
149
|
+
return { verifier: { promptPath: enginePromptPath("verifier.md", prompts) } };
|
|
150
|
+
}
|
|
151
|
+
|
|
163
152
|
// The materiality gate (odd/tasks/materiality-gate.md) is ENGINE-owned like
|
|
164
153
|
// the consolidation pass but OFF by default: it is config-only
|
|
165
154
|
// (`materiality_gate`), and absent means today's pipeline, byte for byte. The
|
|
@@ -201,10 +190,9 @@ export async function buildCliRoutePlan(params: {
|
|
|
201
190
|
models?: ModelsConfig;
|
|
202
191
|
summary: SummarySettings;
|
|
203
192
|
summarizerEnabled?: boolean;
|
|
204
|
-
scoutEnabled?: boolean;
|
|
205
193
|
// Defaults ON, like the stage itself in both product paths; a re-review
|
|
206
|
-
// that skips discovery passes false, the
|
|
207
|
-
//
|
|
194
|
+
// that skips discovery passes false, so the plan never carries a route for a
|
|
195
|
+
// step that will not run.
|
|
208
196
|
consolidateEnabled?: boolean;
|
|
209
197
|
// Defaults OFF, unlike consolidation: the gate is opt-in, and a judge key in
|
|
210
198
|
// the plan changes its fingerprint, so absent must mean today's plan.
|
|
@@ -215,7 +203,6 @@ export async function buildCliRoutePlan(params: {
|
|
|
215
203
|
profile: ModelProfile;
|
|
216
204
|
}): Promise<ResolvedRoutePlan> {
|
|
217
205
|
const summarizerEnabled = params.summarizerEnabled ?? params.summary.enabled;
|
|
218
|
-
const scoutEnabled = params.scoutEnabled ?? params.options.scout;
|
|
219
206
|
const consolidateEnabled = params.consolidateEnabled ?? true;
|
|
220
207
|
// The root prompts are the engine's own (pipeline*Input marks them), so
|
|
221
208
|
// their frontmatter is not read: the profile holds their seat (S3), the
|
|
@@ -230,9 +217,6 @@ export async function buildCliRoutePlan(params: {
|
|
|
230
217
|
...(summarizerEnabled
|
|
231
218
|
? { summarizer: { model: params.summary.model, bundled: true } }
|
|
232
219
|
: {}),
|
|
233
|
-
...(scoutEnabled
|
|
234
|
-
? { scout: { model: params.options.scoutModel, bundled: true } }
|
|
235
|
-
: {}),
|
|
236
220
|
...(consolidateEnabled ? { consolidate: { bundled: true } } : {}),
|
|
237
221
|
...(params.materialityEnabled === true
|
|
238
222
|
? { materiality: true as const }
|