crysk 0.7.0 → 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.
Files changed (133) hide show
  1. package/README.md +27 -23
  2. package/dist/SKILL-1q3658n0.md +74 -0
  3. package/dist/SKILL-25j4q9b8.md +143 -0
  4. package/dist/SKILL-4747bhne.md +143 -0
  5. package/dist/SKILL-4jc3sv23.md +74 -0
  6. package/dist/SKILL-c0492rch.md +540 -0
  7. package/dist/SKILL-d1nzkhhr.md +74 -0
  8. package/dist/SKILL-f8p7kvx0.md +143 -0
  9. package/dist/SKILL-gxd45g38.md +553 -0
  10. package/dist/SKILL-rec41991.md +66 -0
  11. package/dist/SKILL-zze8zj5g.md +536 -0
  12. package/dist/cli.js +24658 -24654
  13. package/dist/index.js +492 -463
  14. package/dist/opencode-ci-qrcv7a99.md +106 -0
  15. package/dist/opencode-ci-rdpcc7xc.md +106 -0
  16. package/dist/opencode-ci-yt0k74t0.md +106 -0
  17. package/dist/workflow-andv2dvm.yml +158 -0
  18. package/package.json +1 -1
  19. package/prompts/default/PROVENANCE.md +1 -0
  20. package/skills/crysk/SKILL.md +6 -6
  21. package/skills/crysk-ci-setup/SKILL.md +32 -33
  22. package/skills/crysk-ci-setup/assets/workflow.yml +5 -10
  23. package/skills/crysk-ci-setup/references/ci-admission.md +5 -5
  24. package/skills/crysk-ci-setup/references/opencode-ci.md +6 -6
  25. package/skills/crysk-loop/SKILL.md +37 -54
  26. package/skills/crysk-pr/SKILL.md +1 -1
  27. package/src/asset-manifest.ts +0 -6
  28. package/src/assets.ts +1 -4
  29. package/src/brand.ts +20 -30
  30. package/src/ci/gates.ts +155 -201
  31. package/src/ci/outcome.ts +4 -31
  32. package/src/ci/platform-limits.ts +135 -0
  33. package/src/ci/platform-report.ts +212 -126
  34. package/src/ci/platform.ts +28 -26
  35. package/src/ci/reporter.ts +25 -26
  36. package/src/ci/review-admission.ts +1 -2
  37. package/src/ci/review-levels.ts +12 -53
  38. package/src/ci/setup.ts +96 -66
  39. package/src/ci/trace-bundle.ts +11 -27
  40. package/src/ci/trace-transcripts.ts +4 -6
  41. package/src/ci/trigger.ts +3 -5
  42. package/src/cli.ts +6 -5
  43. package/src/commands/ci-setup-questions.ts +2 -55
  44. package/src/commands/ci-setup.ts +2 -26
  45. package/src/commands/triage.ts +131 -38
  46. package/src/commands/usage.ts +0 -1
  47. package/src/compare/compare.ts +3 -3
  48. package/src/compare/floor-test.ts +30 -201
  49. package/src/config/config.ts +6 -30
  50. package/src/diversity/accounting.ts +0 -26
  51. package/src/diversity/benchmark.ts +1 -17
  52. package/src/diversity/identity.ts +0 -14
  53. package/src/diversity/pipeline-integration.ts +0 -29
  54. package/src/doctor.ts +3 -6
  55. package/src/execution/claude-session-trace.ts +7 -2
  56. package/src/execution/contracts.ts +15 -9
  57. package/src/execution/harness.ts +165 -187
  58. package/src/execution/spend-limiter.ts +39 -131
  59. package/src/execution/usage-facts.ts +397 -0
  60. package/src/execution/usage-normalized.ts +33 -86
  61. package/src/harness/materiality-judge.ts +0 -5
  62. package/src/harness/pipeline-plan.ts +0 -3
  63. package/src/harness/replay-rows.ts +0 -5
  64. package/src/harness/route-runner.ts +2 -2
  65. package/src/harness/single-step.ts +3 -9
  66. package/src/index.ts +1 -1
  67. package/src/mcp/preflight.ts +2 -2
  68. package/src/model/catalog.ts +1 -3
  69. package/src/model/free-discovery.ts +11 -78
  70. package/src/model/model-name.ts +14 -4
  71. package/src/model/profiles/claude.ts +1 -2
  72. package/src/model/profiles/openai.ts +1 -2
  73. package/src/model/profiles.ts +1 -2
  74. package/src/model/provider-capabilities.ts +16 -55
  75. package/src/model/routing.ts +39 -36
  76. package/src/pr/admission.ts +6 -6
  77. package/src/pr/answered-findings.ts +8 -4
  78. package/src/pr/ci-publish.ts +0 -2
  79. package/src/pr/discovery.ts +4 -25
  80. package/src/pr/plan.ts +40 -121
  81. package/src/pr/pr-status.ts +14 -2
  82. package/src/pr/preflight.ts +15 -12
  83. package/src/pr/publish-outcome.ts +1 -19
  84. package/src/pr/review-pr.ts +10 -47
  85. package/src/pr/target.ts +76 -82
  86. package/src/production-runtime.ts +52 -114
  87. package/src/progress.ts +0 -54
  88. package/src/rereview/prepare.ts +4 -1
  89. package/src/rereview/state.ts +1 -1
  90. package/src/review/boundary.ts +4 -8
  91. package/src/review/drafts.ts +3 -3
  92. package/src/review/findings-conformance.ts +1 -1
  93. package/src/review/findings.ts +0 -4
  94. package/src/review/pipeline.ts +57 -395
  95. package/src/review/preflight.ts +162 -201
  96. package/src/review/prompt-set.ts +0 -2
  97. package/src/review/report.ts +2 -162
  98. package/src/review/review.ts +16 -65
  99. package/src/review/route-preflight.ts +3 -35
  100. package/src/review/run.ts +37 -73
  101. package/src/review/size-gate.ts +3 -3
  102. package/src/review/spec.ts +6 -11
  103. package/src/review/step-runner.ts +4 -4
  104. package/src/runner-authority.ts +10 -9
  105. package/src/security/credential-broker.ts +5 -6
  106. package/src/server/preflight.ts +0 -1
  107. package/src/server/server.ts +17 -6
  108. package/src/store/activity.ts +1 -2
  109. package/src/store/backfill-preflight.ts +0 -3
  110. package/src/store/backfill.ts +0 -1
  111. package/src/store/metrics-preflight.ts +4 -14
  112. package/src/store/metrics.ts +5 -5
  113. package/src/store/preflight.ts +0 -13
  114. package/src/store/store.ts +11 -12
  115. package/src/transport-registry.ts +6 -30
  116. package/src/transports/claude-code-cli.ts +149 -215
  117. package/src/transports/opencode-client.ts +4 -24
  118. package/src/transports/opencode-sdk.ts +17 -112
  119. package/src/triage/plain-reply.ts +37 -0
  120. package/src/ui/activity.ts +1 -6
  121. package/src/ui/config-edit.ts +14 -67
  122. package/src/ui/config.ts +3 -8
  123. package/src/ui/plan.ts +11 -198
  124. package/src/ui/progress.ts +3 -49
  125. package/src/ui/result.ts +18 -77
  126. package/src/ui/review-menu.ts +3 -12
  127. package/src/ui/select.ts +9 -79
  128. package/src/updater.ts +1 -1
  129. package/src/usage.ts +1 -7
  130. package/prompts/default/deep-review-parity.md +0 -152
  131. package/prompts/scout.md +0 -140
  132. package/src/ci/platform-estimate.ts +0 -111
  133. package/src/review/scout.ts +0 -323
package/src/ci/gates.ts CHANGED
@@ -1,14 +1,13 @@
1
- // Bounded unattended spend (ROADMAP Pillar 3, GitHub Actions CI): the two
2
- // gates that must clear BEFORE a CI run spawns a single agent, and the pure
1
+ // Bounded unattended spend (ROADMAP Pillar 3, GitHub Actions CI): the gates
2
+ // that must clear BEFORE a CI run spawns a single agent, and the pure
3
3
  // CI-mode skip payloads those gates produce when they don't.
4
4
  //
5
5
  // Phase 3 additions (this module's scope now covers the whole CI-mode
6
- // policy layer, not just the two spend gates): `planCiSizeSkip`/
7
- // `planCiBudgetSkip` are the ONE call reviewPr's shell makes per gate —
6
+ // policy layer, not just the spend gates): `planCiSizeSkip`/
7
+ // `planCiLimitsSkip` are the ONE call reviewPr's shell makes per gate —
8
8
  // comment + marker + step-summary markdown + $GITHUB_OUTPUT, all from the
9
- // same numbers, so the shell has no decision left to make beyond "post it /
10
- // write it". `budgetDisabledWarningMessage` covers spec 3.1's disabled-
11
- // ceiling `::warning::` requirement. The exit-code contract (spec 2.1) lives
9
+ // same inputs, so the shell has no decision left to make beyond "post it /
10
+ // write it". The exit-code contract (spec 2.1) lives
12
11
  // in src/review/exit-code.ts, shared by every mode.
13
12
  //
14
13
  // This module does NOT reimplement the size gate — that gate (deterministic
@@ -20,8 +19,8 @@
20
19
  // there is nobody to prompt AND the run must still exit 0 with a courteous
21
20
  // PR notice (assistant posture — see proposal.md's Invariants). So this
22
21
  // module adds exactly that CI-mode handling on top of an already-computed
23
- // `SizeGateVerdict`, plus the budget gate's own threshold check (there is no
24
- // pre-existing budget gate to reuse).
22
+ // `SizeGateVerdict`, plus the platform limits rule (the engine computes no
23
+ // dollar figure; see `evaluatePlatformLimits`).
25
24
  //
26
25
  // Purity (project rule 1): every function here is a total function of its
27
26
  // arguments. No file I/O, no `process.env` sniffing, no `log()`, no network,
@@ -67,6 +66,7 @@ import { commandHint, MARKER_OPEN } from "../brand";
67
66
  import { CliUsageError } from "../errors";
68
67
  import { envBillsMetered } from "../execution/usage-normalized";
69
68
  import { ciForceReviewHintLine } from "./force-comment";
69
+ import type { PlatformLimitHit, PlatformLimits } from "./platform-limits";
70
70
  import {
71
71
  type CiOutputs,
72
72
  type CiSummaryData,
@@ -83,56 +83,104 @@ import type { CiReviewPolicyMode } from "./review-policy";
83
83
  import type { DeltaRiskAssessment } from "./review-risk";
84
84
  import { OVERRIDE_COMMAND } from "./trigger";
85
85
 
86
- function usd(amount: number): string {
87
- return `$${amount.toFixed(2)}`;
88
- }
89
-
90
86
  // ---------------------------------------------------------------------------
91
- // Budget gate — spec.md 3.1: "If estimated cost exceeds --budget-usd ...".
92
- // The one comparison this module owns outright (no pre-existing gate to
93
- // reuse, unlike size).
87
+ // Platform limits — platform-owned-cost L55. The platform owns the dollar
88
+ // limits (settings, defaults, the month's spend) and sends them with its
89
+ // estimate at the token exchange, nothing when the review was forced. The
90
+ // engine keeps ONE comparison, because only it knows whether the run bills an
91
+ // API key: skip when the run does and either the month's limit is used up, or
92
+ // the platform's estimate is over the per-review limit or over what is left of
93
+ // the month.
94
94
  // ---------------------------------------------------------------------------
95
95
 
96
- export interface BudgetGateVerdict {
97
- allowed: boolean;
98
- // Present only when `allowed` is false — mirrors SizeGateVerdict's own
99
- // discriminated shape (size-gate.ts) in spirit, kept optional-on-a-flat-
100
- // object here per the task's own stated signature rather than a second
101
- // discriminated union for a single string field.
102
- reason?: string;
96
+ export type PlatformLimitsVerdict =
97
+ | { allowed: true }
98
+ | {
99
+ allowed: false;
100
+ hit: PlatformLimitHit;
101
+ // Set when the comparison used the API share of the estimate (S17), not
102
+ // the whole of it; the skip notice says so.
103
+ apiShare?: true;
104
+ };
105
+
106
+ // Which model roles the run bills on an API key (true) and which it does not
107
+ // (false), resolved once from the run's routes (`modelRoleBilling`, ci/setup.ts).
108
+ export type ModelRoleBilling = ReadonlyMap<string, boolean>;
109
+
110
+ // The part of the estimate that runs on an API key (S17): the platform's split
111
+ // summed over the roles the run bills on one, plus every key the run cannot map
112
+ // to one of its roles (an unknown role may bill, so it counts). Undefined means
113
+ // there is no usable split, and the whole estimate is compared: no split sent
114
+ // or the run's routes unresolved (a run whose roles bill no API key never gets
115
+ // here: evaluatePlatformLimits lets it run).
116
+ // Rounded to the platform's 6 decimals so a sum of its parts sits on a limit
117
+ // instead of a hair over it.
118
+ function apiShareUsd(
119
+ limits: PlatformLimits,
120
+ roleBilling: ModelRoleBilling | undefined,
121
+ ): number | undefined {
122
+ const split = limits.estimateByRoleUsd;
123
+ if (split === undefined || roleBilling === undefined) return undefined;
124
+ if (!hasApiBilledRole(roleBilling)) return undefined;
125
+ let share = 0;
126
+ for (const [role, usd] of Object.entries(split)) {
127
+ if (roleBilling.get(role) !== false) share += usd;
128
+ }
129
+ return Math.round(share * 1e6) / 1e6;
130
+ }
131
+
132
+ // True when at least one resolved role bills an API key.
133
+ function hasApiBilledRole(roleBilling: ModelRoleBilling): boolean {
134
+ return [...roleBilling.values()].some(Boolean);
103
135
  }
104
136
 
105
- // Inclusive boundary: a cost exactly AT the ceiling is allowed, matching
137
+ // Inclusive boundary: an estimate exactly AT a limit passes, matching
106
138
  // size-gate.ts's own `>` (not `>=`) comparison against its limits.
107
139
  //
108
- // A non-positive ceiling DISABLES the gate. That is not this module's
109
- // invention — size-gate.ts:20 already settled it for the sibling knobs
110
- // ("<= 0 disables the limit. Both knobs, independently."), and all three
111
- // ship side by side in one action.yml feeding one preflight, so `budget-usd:
112
- // 0` has to mean what `max-changed-lines: 0` means. The external convention
113
- // does cut the other way (a $0 spend cap elsewhere often means "spend
114
- // nothing"), which is why it is written down here rather than assumed.
140
+ // No estimate and a month with money left: the review runs. That is a stated
141
+ // limit of the design (a model the platform cannot price is not capped per
142
+ // review), not an oversight; the platform's own month ledger still counts the
143
+ // run once it reports.
115
144
  //
116
- // What settles it is that the failure modes are not symmetric. Read as "no
117
- // ceiling" when the operator meant "spend nothing", being wrong costs one
118
- // visible review, still bounded by the size gate running independently. Read
119
- // as "always skip" when the operator meant "no ceiling", being wrong makes
120
- // pr-hero go dark on every PR while each run still exits 0 green — the same
121
- // shape as a lint gate that checks nothing and passes, which this repo has
122
- // already paid for once (docs/research/scout-design.md:344). A disable is only safe
123
- // when it is loud, so the CI shell warns on a disabled ceiling.
124
- export function evaluateBudgetGate(
125
- estimatedCostUsd: number,
126
- budgetUsd: number,
127
- ): BudgetGateVerdict {
128
- if (budgetUsd <= 0) return { allowed: true };
129
- if (estimatedCostUsd <= budgetUsd) return { allowed: true };
130
- return {
131
- allowed: false,
132
- reason:
133
- `estimated cost ${usd(estimatedCostUsd)} exceeds the ` +
134
- `${usd(budgetUsd)} budget ceiling`,
135
- };
145
+ // A subscription or free route is never skipped: its runs draw on quota, and
146
+ // only API spend counts toward the limits (L54, "counts": "api_only"). In a run
147
+ // whose roles bill differently, the estimate compared is the API share when the
148
+ // platform split it per role (S17); without the split it is the whole estimate,
149
+ // the conservative side.
150
+ export function evaluatePlatformLimits(input: {
151
+ billingMode: CiBillingMode;
152
+ limits: PlatformLimits | undefined;
153
+ roleBilling?: ModelRoleBilling;
154
+ }): PlatformLimitsVerdict {
155
+ const { limits } = input;
156
+ if (input.billingMode !== "metered" || limits === undefined) {
157
+ return { allowed: true };
158
+ }
159
+ // Metered by the Claude env alone (an API key present) while no resolved role
160
+ // routes to it: the run bills no API spend, so it is never skipped (#551 R001).
161
+ // No role map (routes unresolved) keeps the conservative whole-estimate path.
162
+ if (
163
+ input.roleBilling !== undefined &&
164
+ input.roleBilling.size > 0 &&
165
+ !hasApiBilledRole(input.roleBilling)
166
+ ) {
167
+ return { allowed: true };
168
+ }
169
+ const { reviewUsd, monthLeftUsd, estimateUsd } = limits;
170
+ if (monthLeftUsd !== undefined && monthLeftUsd <= 0) {
171
+ return { allowed: false, hit: "month-used-up" };
172
+ }
173
+ if (estimateUsd === undefined) return { allowed: true };
174
+ const share = apiShareUsd(limits, input.roleBilling);
175
+ const compared = share ?? estimateUsd;
176
+ const basis = share === undefined ? {} : { apiShare: true as const };
177
+ if (reviewUsd !== undefined && compared > reviewUsd) {
178
+ return { allowed: false, hit: "review-limit", ...basis };
179
+ }
180
+ if (monthLeftUsd !== undefined && compared > monthLeftUsd) {
181
+ return { allowed: false, hit: "month-left", ...basis };
182
+ }
183
+ return { allowed: true };
136
184
  }
137
185
 
138
186
  // ---------------------------------------------------------------------------
@@ -214,27 +262,35 @@ function buildSizeSkipComment(
214
262
  return `${lines.join("\n")}\n`;
215
263
  }
216
264
 
217
- // What a platform repository can do about a budget skip (cryskai/platform#12):
218
- // it has no `budget-usd` input to raise, and what a review costs is the model
219
- // setup chosen in the dashboard.
220
- const PLATFORM_BUDGET_SKIP_HINT =
221
- "To change what reviews cost, change the model setup in the Crysk dashboard.";
265
+ // What a platform repository can do about a limits skip: the limits are
266
+ // dashboard settings (Review > Limits), not workflow inputs.
267
+ const LIMIT_SKIP_REASON: Record<PlatformLimitHit, string> = {
268
+ "month-used-up":
269
+ "⚠️ This organization's monthly review limit is used up for this calendar month.",
270
+ "review-limit":
271
+ "⚠️ The estimated cost of this review is over the per-review limit.",
272
+ "month-left":
273
+ "⚠️ The estimated cost of this review is more than what is left of this organization's monthly limit.",
274
+ };
275
+
276
+ // A mixed setup is compared on the part of the estimate that runs on an API
277
+ // key; the notice says so, because the estimate the reader may know is larger.
278
+ const API_SHARE_LINE =
279
+ "Only the part of the estimate that runs on an API key (its API share) " +
280
+ "was compared with the limit.";
222
281
 
223
282
  function buildBudgetSkipComment(data: SkipBudgetSummary): string {
224
- const platformConnected = data.platformConnected === true;
225
283
  const lines = [
226
284
  SKIP_BUDGET_COMMENT_MARKER,
227
285
  "## Crysk review skipped",
228
286
  "",
229
- `⚠️ The estimated review cost (${usd(data.estimatedCostUsd)}) exceeds ` +
230
- `the configured CI budget ceiling (${usd(data.budgetUsd)}).`,
287
+ LIMIT_SKIP_REASON[data.limit],
288
+ ...(data.apiShare === true ? ["", API_SHARE_LINE] : []),
231
289
  "",
232
- platformConnected
233
- ? PLATFORM_BUDGET_SKIP_HINT
234
- : "Crysk did not run to stay within the configured `--budget-usd` " +
235
- "ceiling.",
290
+ "Crysk did not run to stay within the limits set in the Crysk dashboard " +
291
+ "(Review > Limits). Raise the limit there to review later pushes.",
236
292
  "",
237
- ciForceReviewHintLine(data.prNumber, platformConnected),
293
+ ciForceReviewHintLine(data.prNumber, true),
238
294
  ];
239
295
  return `${lines.join("\n")}\n`;
240
296
  }
@@ -280,62 +336,34 @@ export function ciSizeGateSkip(input: CiSizeGateSkipInput): CiGateSkip | null {
280
336
  };
281
337
  }
282
338
 
283
- export interface CiBudgetGateSkipInput {
339
+ export interface CiLimitsGateSkipInput {
284
340
  isCi: boolean;
285
- estimatedCostUsd: number;
286
- budgetUsd: number;
341
+ billingMode: CiBillingMode;
342
+ limits: PlatformLimits | undefined;
343
+ // The run's roles that bill an API key, for the estimate split (S17).
344
+ roleBilling?: ModelRoleBilling;
287
345
  prNumber: number;
288
- // The run came through the Crysk platform (CRYSK_PLATFORM_CONNECTED): the
289
- // skip names only what applies to a platform repository.
290
- platformConnected?: boolean;
291
346
  }
292
347
 
293
- export function ciBudgetGateSkip(
294
- input: CiBudgetGateSkipInput,
348
+ export function ciLimitsGateSkip(
349
+ input: CiLimitsGateSkipInput,
295
350
  ): CiGateSkip | null {
296
- const verdict = evaluateBudgetGate(input.estimatedCostUsd, input.budgetUsd);
297
- if (!input.isCi || verdict.allowed) return null;
351
+ if (!input.isCi) return null;
352
+ const verdict = evaluatePlatformLimits(input);
353
+ if (verdict.allowed) return null;
298
354
  const summary: SkipBudgetSummary = {
299
355
  kind: "skipped-budget",
300
356
  prNumber: input.prNumber,
301
- estimatedCostUsd: input.estimatedCostUsd,
302
- budgetUsd: input.budgetUsd,
303
- ...(input.platformConnected === true ? { platformConnected: true } : {}),
357
+ limit: verdict.hit,
358
+ ...(verdict.apiShare === true ? { apiShare: true as const } : {}),
304
359
  };
305
360
  return { comment: buildBudgetSkipComment(summary), summary };
306
361
  }
307
362
 
308
363
  // ---------------------------------------------------------------------------
309
- // Disabled-ceiling warning — spec 3.1: "Because a silent disable is
310
- // indistinguishable from a passing gate, the CI shell MUST emit a
311
- // `::warning::` workflow command noting that the budget ceiling is
312
- // disabled." Only for an EXPLICIT <= 0 — `undefined` (the flag was never
313
- // given) means no ceiling was ever configured, so there is nothing to warn
314
- // about disabling. The shell wraps this message with
315
- // `formatWorkflowCommand("warning", ...)` (ci/reporter.ts); this module only
316
- // decides whether to and builds the text, never the annotation syntax.
317
- // ---------------------------------------------------------------------------
318
-
319
- export function budgetDisabledWarningMessage(
320
- budgetUsd: number | undefined,
321
- ): string | null {
322
- if (budgetUsd === undefined || budgetUsd > 0) return null;
323
- return (
324
- "budget-usd ceiling is disabled (<= 0 was configured); estimated cost " +
325
- "is unconstrained for this CI run. The size gate is resolved " +
326
- "separately."
327
- );
328
- }
329
-
330
- // ---------------------------------------------------------------------------
331
- // Route-aware ceiling resolution (issue #156). The budget gate compares
332
- // `estimate.high` — a TOKEN-derived figure — against a real-dollar ceiling.
333
- // On a Claude subscription route the real cash cost of a run is $0.00, so the
334
- // shipped $10 default refused to do work over an overrun that cannot happen,
335
- // and a skipped review is indistinguishable from a clean one to anyone
336
- // reading the checks (this module's own doctrine, ~110-115; ci/setup.ts:50-63).
337
- // So: no ceiling by default on a subscription route, the default ceiling on a
338
- // metered one, and an explicit operator value honoured verbatim on either.
364
+ // How the run bills (issue #156). The size gate's route-aware limits and the
365
+ // platform limits gate both read it: a subscription route has no per-token
366
+ // invoice, so the platform's dollar limits never skip it.
339
367
  // ---------------------------------------------------------------------------
340
368
 
341
369
  export type CiBillingMode = "subscription" | "metered";
@@ -425,73 +453,6 @@ export function deriveCiBillingMode(
425
453
  : "subscription";
426
454
  }
427
455
 
428
- // 12, raised from 10 when the default set gained its fourth always-on hunter.
429
- // The gate compares the estimate's UPPER bound with this ceiling, and the
430
- // estimate (review/report.ts) prices one seat per hunter. With the summarizer
431
- // on, that hunter lifted every upper bound by 20%: the smallest pull request
432
- // went from $8.53 to $10.24, so a ceiling of 10 skipped all of them. 12 admits
433
- // what 10 admitted before. With the summarizer off the rise is 25%, so a pull
434
- // request whose old upper bound sat between $9.60 and $10 is skipped now.
435
- // A workflow that WRITES `budget-usd: 10` carries an operator value and does
436
- // not move with this constant.
437
- export const CI_DEFAULT_METERED_BUDGET_USD = 12;
438
-
439
- export interface CiBudgetCeiling {
440
- // `undefined` means NO ceiling — the gate does not run at all. Distinct from
441
- // a `0` ceiling, which reaches `evaluateBudgetGate` and is allowed there by
442
- // the `<= 0` convention; the two arrive by different routes and the shell
443
- // announces them with different messages.
444
- budgetUsd: number | undefined;
445
- source: "operator" | "default-metered" | "unlimited-subscription";
446
- }
447
-
448
- // Total over the four cases. `source` exists so the caller can tell the two
449
- // no-ceiling outcomes apart without re-deriving why.
450
- //
451
- // An explicitly configured value is honoured verbatim on ANY route — that is
452
- // how an operator imposes a ceiling on a subscription (`budget-usd: 5`) or
453
- // removes one on a metered route (`budget-usd: 0`, which still reaches
454
- // `budgetDisabledWarningMessage`). The `configured !== undefined` test is what
455
- // keeps 0 and negatives on the operator branch: falling back to a truthiness
456
- // check would swallow the explicit disable and silently reimpose the default the
457
- // operator just removed.
458
- export function resolveCiBudgetCeiling(input: {
459
- configured: number | undefined;
460
- billingMode: CiBillingMode;
461
- }): CiBudgetCeiling {
462
- if (input.configured !== undefined) {
463
- return { budgetUsd: input.configured, source: "operator" };
464
- }
465
- if (input.billingMode === "metered") {
466
- return {
467
- budgetUsd: CI_DEFAULT_METERED_BUDGET_USD,
468
- source: "default-metered",
469
- };
470
- }
471
- return { budgetUsd: undefined, source: "unlimited-subscription" };
472
- }
473
-
474
- // The loudness half. `unlimited-subscription` produces no ceiling and would
475
- // otherwise be silent, which is the exact shape this module already refuses
476
- // for the `<= 0` case: a disable is only safe when it is loud.
477
- //
478
- // Kept SEPARATE from `budgetDisabledWarningMessage`, and separate in register
479
- // too — a NOTE, not a warning. The operator chose nothing wrong here; they
480
- // configured nothing and the policy resolved. Merging the two would tell
481
- // someone who never set `budget-usd` that they disabled something.
482
- export function budgetUnlimitedNoticeMessage(
483
- ceiling: CiBudgetCeiling,
484
- ): string | null {
485
- if (ceiling.source !== "unlimited-subscription") return null;
486
- return (
487
- "no budget-usd ceiling was applied: this route has no per-token invoice " +
488
- "credential (a Claude subscription or a free route), so a run draws on " +
489
- "quota rather than a per-token invoice and the estimated cost is not a " +
490
- "dollar figure to gate on. The size gate resolves from the route separately. " +
491
- "Set `budget-usd: <n>` to impose a ceiling anyway."
492
- );
493
- }
494
-
495
456
  // ---------------------------------------------------------------------------
496
457
  // Route-aware size limits — the size gate's half of the #156 policy above.
497
458
  // The size gate is a COST and predictability gate (review/size-gate.ts's own
@@ -575,7 +536,7 @@ export function sizeGateUnlimitedNoticeMessage(
575
536
 
576
537
  // ---------------------------------------------------------------------------
577
538
  // CI skip plan — the ONE call reviewPr's shell makes per gate. Composes
578
- // ciSizeGateSkip/ciBudgetGateSkip's {comment, summary} with renderStepSummary
539
+ // ciSizeGateSkip/ciLimitsGateSkip's {comment, summary} with renderStepSummary
579
540
  // (ci/reporter.ts) and the $GITHUB_OUTPUT contract into everything the shell
580
541
  // needs to publish, so the shell's own job is pure mechanical glue: post
581
542
  // `comment` under `markerPrefix` if `--post`, append `summaryMarkdown` if
@@ -596,11 +557,7 @@ export interface CiGateSkipPlan {
596
557
  }
597
558
 
598
559
  // Spec 1.1's $GITHUB_OUTPUT contract for a skip: no review ran, so every
599
- // finding count is 0 and there is no run dir. `estimatedCostUsd` is 0 for a
600
- // size skip (the gate fires before any cost estimate exists) and the
601
- // estimate that tripped the ceiling for a budget skip — the same number the
602
- // comment and summary already show (this module's own "same numbers"
603
- // doctrine, see the module header).
560
+ // finding count is 0 and there is no run dir.
604
561
  export function ciGateSkipOutputs(
605
562
  status:
606
563
  | "skipped-size"
@@ -608,14 +565,12 @@ export function ciGateSkipOutputs(
608
565
  | "skipped-coverage"
609
566
  | "skipped-excluded"
610
567
  | "manual-required",
611
- estimatedCostUsd: number,
612
568
  ): CiOutputs {
613
569
  return {
614
570
  status,
615
571
  findings_count: 0,
616
572
  blocking_count: 0,
617
573
  advisory_count: 0,
618
- cost_usd_est: estimatedCostUsd,
619
574
  run_dir: "",
620
575
  };
621
576
  }
@@ -629,7 +584,7 @@ export function planCiSizeSkip(
629
584
  comment: skip.comment,
630
585
  markerPrefix: SKIP_SIZE_COMMENT_MARKER,
631
586
  summaryMarkdown: renderStepSummary(skip.summary, input.ciMode),
632
- outputs: ciGateSkipOutputs("skipped-size", 0),
587
+ outputs: ciGateSkipOutputs("skipped-size"),
633
588
  };
634
589
  }
635
590
 
@@ -654,25 +609,25 @@ export function planCiExcludedSkip(input: {
654
609
  },
655
610
  input.ciMode,
656
611
  ),
657
- outputs: ciGateSkipOutputs("skipped-excluded", 0),
612
+ outputs: ciGateSkipOutputs("skipped-excluded"),
658
613
  };
659
614
  }
660
615
 
661
- export function planCiBudgetSkip(
662
- input: CiBudgetGateSkipInput & { force: boolean; ciMode: CiMode },
616
+ export function planCiLimitsSkip(
617
+ input: CiLimitsGateSkipInput & { force: boolean; ciMode: CiMode },
663
618
  ): CiGateSkipPlan | null {
664
619
  // CI `--force` is the `/crysk review` comment. It answers "run this one
665
- // anyway" for every CI skip, including the spend ceiling. Outside that
666
- // explicit comment the ceiling still holds. Local `--force` never reaches
667
- // here: the caller only invokes this plan when `isCi` is set.
620
+ // anyway" for every CI skip, including the platform limits. The platform
621
+ // also omits the limits for a forced dispatch. Local `--force` never
622
+ // reaches here: the caller only invokes this plan when `isCi` is set.
668
623
  if (input.force) return null;
669
- const skip = ciBudgetGateSkip(input);
624
+ const skip = ciLimitsGateSkip(input);
670
625
  if (skip === null) return null;
671
626
  return {
672
627
  comment: skip.comment,
673
628
  markerPrefix: SKIP_BUDGET_COMMENT_MARKER,
674
629
  summaryMarkdown: renderStepSummary(skip.summary, input.ciMode),
675
- outputs: ciGateSkipOutputs("skipped-budget", input.estimatedCostUsd),
630
+ outputs: ciGateSkipOutputs("skipped-budget"),
676
631
  };
677
632
  }
678
633
 
@@ -752,10 +707,12 @@ function buildManualRequiredComment(
752
707
  "",
753
708
  "What the gate does with the next pushes: no review reads them, so it never turns green on its own. " +
754
709
  "A red verdict stays red while a blocking finding is unanswered. A green verdict, or a red one whose " +
755
- "blocking findings all have a reply from someone with write access (`applied`, or a `dismissed` the " +
756
- "CI adjudicator upheld), waits as `pending` until a review verifies the head or " +
757
- `${commandHint(OVERRIDE_COMMAND)} merges it unverified. Reply before you push: the gate reads the ` +
758
- "replies when the push arrives.",
710
+ "blocking findings all have a reply from someone with write access, waits as `pending` until a review " +
711
+ "verifies the head or " +
712
+ `${commandHint(OVERRIDE_COMMAND)} merges it unverified. A reply counts when it is posted in the ` +
713
+ "finding's thread and starts with `applied` (a plain comment is enough), or when it is a `dismissed` " +
714
+ "the CI adjudicator upheld (`crysk triage reply`). Reply before you push: the gate reads the replies " +
715
+ "when the push arrives.",
759
716
  "",
760
717
  "The existing review comment still describes the last head Crysk reviewed.",
761
718
  "",
@@ -842,7 +799,7 @@ export function planCiReviewSkip(input: {
842
799
  ),
843
800
  markerPrefix: SKIP_COVERAGE_COMMENT_MARKER,
844
801
  summaryMarkdown: renderStepSummary(summary, input.ciMode),
845
- outputs: ciGateSkipOutputs("skipped-coverage", 0),
802
+ outputs: ciGateSkipOutputs("skipped-coverage"),
846
803
  };
847
804
  }
848
805
 
@@ -875,7 +832,7 @@ export function planCiReviewManualRequired(input: {
875
832
  ),
876
833
  markerPrefix: MANUAL_REQUIRED_COMMENT_MARKER,
877
834
  summaryMarkdown: renderStepSummary(summary, input.ciMode),
878
- outputs: ciGateSkipOutputs("manual-required", 0),
835
+ outputs: ciGateSkipOutputs("manual-required"),
879
836
  };
880
837
  }
881
838
 
@@ -898,7 +855,6 @@ export function planCiReview(input: {
898
855
  prNumber: number;
899
856
  headSha: string;
900
857
  findings: readonly Finding[];
901
- costUsdEst: number;
902
858
  wallMs: number;
903
859
  model: string;
904
860
  repoWebUrl?: string;
@@ -914,7 +870,6 @@ export function planCiReview(input: {
914
870
  prNumber: input.prNumber,
915
871
  headSha: input.headSha,
916
872
  findings: input.findings,
917
- costUsdEst: input.costUsdEst,
918
873
  wallMs: input.wallMs,
919
874
  model: input.model,
920
875
  ...(input.repoWebUrl === undefined ? {} : { repoWebUrl: input.repoWebUrl }),
@@ -927,7 +882,6 @@ export function planCiReview(input: {
927
882
  findings_count: input.findings.length,
928
883
  blocking_count: blockingCount,
929
884
  advisory_count: input.findings.length - blockingCount,
930
- cost_usd_est: input.costUsdEst,
931
885
  run_dir: input.runDir,
932
886
  },
933
887
  };
@@ -960,7 +914,7 @@ export function shouldWriteCiOutputs(
960
914
 
961
915
  // Spec 2.1: step-summary writing is additionally gated on the tri-state
962
916
  // `--step-summary`/`--no-step-summary` flag (`stepSummary`), unset meaning
963
- // the shell's own default of on — mirroring `summary`/`scout`'s own
917
+ // the shell's own default of on — mirroring `summary`'s own
964
918
  // unset-means-default convention (preflight.ts's CliOptions).
965
919
  export function shouldWriteStepSummary(
966
920
  isCi: boolean,