@bjoernrochel/sevro 0.1.0-rc.7 → 1.0.0

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 (230) hide show
  1. package/CONTRIBUTING.md +22 -6
  2. package/README.md +5 -1
  3. package/docs/README.md +19 -15
  4. package/docs/architecture.md +138 -17
  5. package/docs/assets/README.md +5 -4
  6. package/docs/builtin-graders-v1.md +17 -3
  7. package/docs/codex-host.md +155 -49
  8. package/docs/contracts.md +2 -1
  9. package/docs/creating-cases.md +9 -2
  10. package/docs/development-cli.md +187 -19
  11. package/docs/documentation-quality.md +12 -2
  12. package/docs/evaluation-workflow.md +110 -0
  13. package/docs/extension-protocol-v1.md +26 -3
  14. package/docs/getting-started.md +23 -2
  15. package/docs/guide-evaluation.md +33 -1
  16. package/docs/identity-v1.md +25 -2
  17. package/docs/installing.md +11 -11
  18. package/docs/native-controls.md +76 -5
  19. package/docs/native-hosts.md +196 -7
  20. package/docs/reading-results.md +102 -0
  21. package/docs/recipes.md +489 -0
  22. package/docs/releases.md +31 -12
  23. package/docs/report-v1.md +29 -1
  24. package/docs/results-v1.md +141 -4
  25. package/docs/run-settings-v1.md +147 -0
  26. package/docs/runtime-v1.md +307 -18
  27. package/docs/troubleshooting.md +12 -12
  28. package/docs/typescript-quality.md +89 -26
  29. package/examples/recipes/candidate.ts +26 -0
  30. package/examples/recipes/cases/coding-change.json +21 -0
  31. package/examples/recipes/cases/host-tools.json +21 -0
  32. package/examples/recipes/cases/mise.json +21 -0
  33. package/examples/recipes/cases/node.json +21 -0
  34. package/examples/recipes/cases/python-uv.json +21 -0
  35. package/examples/recipes/cases/semantic.json +17 -0
  36. package/examples/recipes/coding-change/check.ts +23 -0
  37. package/examples/recipes/extension-artifacts.ts +65 -0
  38. package/examples/recipes/extension-methods.ts +174 -0
  39. package/examples/recipes/extension.ts +37 -0
  40. package/examples/recipes/repository-skill/.agents/skills/startup-guide/SKILL.md +7 -0
  41. package/examples/recipes/repository-skill/.claude/skills/startup-guide/SKILL.md +7 -0
  42. package/examples/recipes/repository-skill/README.md +3 -0
  43. package/examples/recipes/run-inputs.ts +188 -0
  44. package/examples/recipes/run-options.ts +131 -0
  45. package/examples/recipes/run.ts +63 -0
  46. package/examples/recipes/runtime/empty.json +1 -0
  47. package/examples/recipes/runtime/goals.json +1 -0
  48. package/examples/recipes/runtime/mise.json +6 -0
  49. package/examples/recipes/runtime/path.json +1 -0
  50. package/examples/recipes/runtime/plugins.json +4 -0
  51. package/examples/recipes/runtime/python-uv.json +10 -0
  52. package/examples/recipes/semantic-judge.ts +29 -0
  53. package/examples/templates/coding/adapter.ts +17 -0
  54. package/examples/templates/coding/case.json +15 -0
  55. package/examples/templates/repository-skill/.agents/skills/startup-guide/SKILL.md +6 -0
  56. package/examples/templates/repository-skill/extension-data.ts +64 -0
  57. package/examples/templates/repository-skill/extension.ts +31 -0
  58. package/examples/templates/response/adapter.ts +12 -0
  59. package/examples/templates/response/case.json +13 -0
  60. package/package.json +11 -5
  61. package/schemas/report-v1.schema.json +9 -0
  62. package/schemas/run-evidence-v1.schema.json +2 -1
  63. package/schemas/runtime-v1.schema.json +63 -7
  64. package/src/advisory.ts +2 -2
  65. package/src/candidate-artifacts.ts +206 -0
  66. package/src/candidate-codex-state.ts +185 -0
  67. package/src/case-validation.ts +73 -50
  68. package/src/cli/cancellation.ts +20 -0
  69. package/src/cli/command.ts +87 -0
  70. package/src/cli/completion.ts +15 -0
  71. package/src/cli/demo.ts +60 -0
  72. package/src/cli/evaluation.ts +55 -0
  73. package/src/cli/extension.ts +79 -0
  74. package/src/cli/help.ts +198 -0
  75. package/src/cli/hosts.ts +55 -0
  76. package/src/cli/init-settings.ts +30 -0
  77. package/src/cli/init.ts +111 -0
  78. package/src/{cli-invocation.ts → cli/invocation.ts} +214 -218
  79. package/src/cli/loading.ts +151 -0
  80. package/src/cli/preparation.ts +154 -0
  81. package/src/cli/presentation.ts +93 -0
  82. package/src/cli/progress.ts +6 -0
  83. package/src/cli/run-manifest.ts +128 -0
  84. package/src/cli/run-options.ts +403 -0
  85. package/src/cli/run-overrides.ts +69 -0
  86. package/src/cli/settings.ts +154 -0
  87. package/src/cli/summary.ts +43 -0
  88. package/src/cli.ts +13 -591
  89. package/src/contract-types.ts +94 -0
  90. package/src/engine.ts +13 -5
  91. package/src/evaluation/advisory.ts +150 -0
  92. package/src/{evaluation-check-plan.ts → evaluation/check-plan.ts} +103 -67
  93. package/src/evaluation/collection.ts +88 -0
  94. package/src/evaluation/context.ts +10 -0
  95. package/src/{evaluation-grading-host.ts → evaluation/grading-host.ts} +5 -5
  96. package/src/evaluation/grading.ts +235 -0
  97. package/src/{evaluation-invocation-plan.ts → evaluation/invocation-plan.ts} +147 -58
  98. package/src/{evaluation-plan.ts → evaluation/plan.ts} +208 -141
  99. package/src/evaluation/progress.ts +19 -0
  100. package/src/evaluation/resources.ts +241 -0
  101. package/src/evaluation/run.ts +206 -0
  102. package/src/{evaluation-semantic.ts → evaluation/semantic.ts} +133 -115
  103. package/src/evaluation/trial.ts +800 -0
  104. package/src/evaluation/types.ts +207 -0
  105. package/src/evidence/host-results.ts +85 -0
  106. package/src/evidence/retained-artifacts.ts +18 -0
  107. package/src/evidence/trial-writer.ts +203 -0
  108. package/src/evidence/usage.ts +17 -0
  109. package/src/extension-client.ts +34 -96
  110. package/src/extension-session.ts +24 -48
  111. package/src/{advisory-fixture.ts → fixtures/advisory.ts} +15 -76
  112. package/src/{fixture-bin.ts → fixtures/bin.ts} +1 -1
  113. package/src/{fixture-cleanup.ts → fixtures/cleanup.ts} +13 -7
  114. package/src/fixtures/create.ts +183 -0
  115. package/src/{generated-fixture.ts → fixtures/generated.ts} +10 -53
  116. package/src/{git-hooks.ts → fixtures/git-hooks.ts} +1 -1
  117. package/src/{preparation.ts → fixtures/preparation.ts} +6 -16
  118. package/src/{repository-fixture.ts → fixtures/repository.ts} +30 -64
  119. package/src/{fixture-setup.ts → fixtures/setup.ts} +33 -51
  120. package/src/{fixture-tools.ts → fixtures/tools.ts} +1 -1
  121. package/src/fixtures/workspace-state.ts +212 -0
  122. package/src/generated/cli-result.d.cts +2 -1
  123. package/src/generated/extension.d.cts +2 -1
  124. package/src/generated/report.cjs +1 -1
  125. package/src/generated/report.d.cts +2 -1
  126. package/src/generated/run-evidence.cjs +1 -1
  127. package/src/generated/run-evidence.d.cts +2 -1
  128. package/src/generated/runtime.cjs +1 -1
  129. package/src/generated/runtime.d.cts +2 -1
  130. package/src/git.ts +93 -0
  131. package/src/graders/git-head.ts +29 -84
  132. package/src/graders/shell.ts +57 -68
  133. package/src/host-diagnostic-redaction.ts +52 -0
  134. package/src/host-diagnostics.ts +331 -0
  135. package/src/hosts/claude-continuation.ts +57 -40
  136. package/src/hosts/claude-credential.ts +160 -63
  137. package/src/hosts/claude-events.ts +34 -18
  138. package/src/hosts/claude-keychain.ts +78 -21
  139. package/src/hosts/claude-login-recovery.ts +85 -0
  140. package/src/hosts/claude-login-refresh.ts +178 -0
  141. package/src/hosts/claude-login-store.ts +162 -0
  142. package/src/hosts/claude-login.ts +107 -0
  143. package/src/hosts/claude-native-goal.ts +15 -12
  144. package/src/hosts/claude-process.ts +10 -50
  145. package/src/hosts/claude-refresh-lock.ts +81 -0
  146. package/src/hosts/claude-repository-invocation.ts +9 -34
  147. package/src/hosts/claude-settings.ts +91 -14
  148. package/src/hosts/claude-skill-activations.ts +169 -0
  149. package/src/hosts/claude-state.ts +46 -16
  150. package/src/hosts/claude-tool-calls.ts +13 -10
  151. package/src/hosts/claude.ts +122 -73
  152. package/src/hosts/codex-app-server.ts +78 -52
  153. package/src/hosts/codex-configuration.ts +5 -10
  154. package/src/hosts/codex-exec-controls.ts +516 -0
  155. package/src/hosts/codex-marketplace.ts +12 -10
  156. package/src/hosts/codex-native-calls.ts +15 -122
  157. package/src/hosts/codex-process.ts +10 -54
  158. package/src/hosts/codex-profile.ts +50 -21
  159. package/src/hosts/codex-session-files.ts +1 -9
  160. package/src/hosts/codex-state.ts +141 -19
  161. package/src/hosts/codex.ts +221 -179
  162. package/src/hosts/isolation-common.ts +15 -11
  163. package/src/hosts/isolation-roots.ts +13 -20
  164. package/src/hosts/linux-sandbox.ts +27 -3
  165. package/src/hosts/mac-sandbox.ts +111 -18
  166. package/src/hosts/native-controls.ts +72 -22
  167. package/src/hosts/native-request.ts +55 -0
  168. package/src/hosts/process-lifecycle.ts +221 -45
  169. package/src/hosts/protected-worktrees.ts +1 -1
  170. package/src/hosts/skill-hook-catalog.ts +131 -0
  171. package/src/hosts/skill-hook-handler.ts +374 -0
  172. package/src/hosts/skill-hooks.ts +577 -0
  173. package/src/instrumentation.ts +7 -8
  174. package/src/native-transcript-state.ts +22 -3
  175. package/src/native-transcript-view.ts +52 -13
  176. package/src/native-transcripts.ts +34 -15
  177. package/src/paths.ts +22 -0
  178. package/src/process-lifecycle.ts +150 -0
  179. package/src/provenance.ts +27 -7
  180. package/src/prune.ts +300 -0
  181. package/src/report.ts +63 -36
  182. package/src/resolved-case.ts +1 -1
  183. package/src/results-lock.ts +42 -0
  184. package/src/results.ts +14 -18
  185. package/src/run-owner.ts +4 -4
  186. package/src/{runtime-config.ts → runtime/config.ts} +182 -44
  187. package/src/{runtime-discovery-apple.ts → runtime/discovery-apple.ts} +3 -3
  188. package/src/{runtime-discovery-providers.ts → runtime/discovery-providers.ts} +3 -9
  189. package/src/{runtime-discovery.ts → runtime/discovery.ts} +4 -4
  190. package/src/{runtime-hook-processes.ts → runtime/hook-processes.ts} +1 -1
  191. package/src/{runtime-hooks.ts → runtime/hooks.ts} +33 -22
  192. package/src/{runtime-paths.ts → runtime/paths.ts} +6 -18
  193. package/src/{runtime-query.ts → runtime/query.ts} +5 -5
  194. package/src/{runtime-seeds.ts → runtime/seeds.ts} +3 -3
  195. package/src/runtime/shared-cache.ts +130 -0
  196. package/src/{runtime-state.ts → runtime/state.ts} +28 -11
  197. package/src/schema-types.ts +73 -161
  198. package/src/shell.ts +4 -0
  199. package/src/transcript-retention.ts +146 -0
  200. package/src/value-guards.ts +22 -0
  201. package/docs/assets/sevro-logo.png +0 -0
  202. package/docs/assets/sevro-symbol.svg +0 -1
  203. package/docs/assets/sevro-wordmark.svg +0 -1
  204. package/docs/cache-ownership-validation.md +0 -49
  205. package/docs/claude-continuation-validation.md +0 -47
  206. package/docs/concurrency-validation.md +0 -86
  207. package/docs/configuration-root-validation.md +0 -70
  208. package/docs/engine-slice.md +0 -114
  209. package/docs/evidence/runtime-startup-30/README.md +0 -185
  210. package/docs/evidence/runtime-startup-30/after.json +0 -319
  211. package/docs/evidence/runtime-startup-30/baseline-timeout.jsonl +0 -2
  212. package/docs/evidence/runtime-startup-30/before.json +0 -403
  213. package/docs/evidence/runtime-startup-30/consumer-smoke.json +0 -63
  214. package/docs/evidence/runtime-startup-30/full-policy-diagnostic.jsonl +0 -3
  215. package/docs/evidence/runtime-startup-30/minimal-policy-diagnostic.jsonl +0 -3
  216. package/docs/evidence/runtime-startup-30/plugin-template-assessment.json +0 -20
  217. package/docs/evidence/runtime-startup-30/seedless-timeout.jsonl +0 -2
  218. package/docs/evidence.md +0 -17
  219. package/docs/fixture-cleanup-validation.md +0 -70
  220. package/docs/guide-verification.json +0 -239
  221. package/src/evaluation-context.ts +0 -257
  222. package/src/evaluation-fixture.ts +0 -313
  223. package/src/evaluation-run.ts +0 -198
  224. package/src/evaluation-trial.ts +0 -946
  225. package/src/evaluation-types.ts +0 -149
  226. package/src/hosts/claude-nested-skills.ts +0 -272
  227. package/src/hosts/codex-command-output.ts +0 -510
  228. package/src/hosts/codex-executor-command.ts +0 -291
  229. package/src/hosts/codex-skill-reads.ts +0 -609
  230. /package/src/{trial-scheduler.ts → evaluation/scheduler.ts} +0 -0
@@ -0,0 +1,111 @@
1
+ import { cp, mkdir, rm, writeFile } from "node:fs/promises";
2
+ import { dirname, join, resolve } from "node:path";
3
+ import { parseArgs } from "node:util";
4
+ import {
5
+ nativeReadme,
6
+ templateCommand,
7
+ templateSettings,
8
+ } from "./init-settings";
9
+
10
+ function initOptions(argv: string[]) {
11
+ const { values, positionals } = parseArgs({
12
+ args: argv,
13
+ strict: true,
14
+ allowPositionals: true,
15
+ options: {
16
+ template: { type: "string", default: "response" },
17
+ directory: { type: "string", default: "sevro-evaluation" },
18
+ },
19
+ });
20
+ if (
21
+ positionals.length ||
22
+ !["response", "coding", "repository-skill"].includes(values.template)
23
+ )
24
+ throw new Error(
25
+ "use --template response, coding or repository-skill; run sevro init --help",
26
+ );
27
+ if (!values.directory.trim())
28
+ throw new Error("--directory must name a new directory");
29
+ return { template: values.template, destination: resolve(values.directory) };
30
+ }
31
+
32
+ async function reserveDirectory(destination: string): Promise<void> {
33
+ await mkdir(dirname(destination), { recursive: true });
34
+ try {
35
+ await mkdir(destination);
36
+ } catch (cause) {
37
+ throw new Error(
38
+ `init destination already exists or cannot be created: ${destination}; choose a new --directory`,
39
+ { cause },
40
+ );
41
+ }
42
+ }
43
+
44
+ function starterReadme(template: string) {
45
+ if (template === "repository-skill") return nativeReadme();
46
+ const assertion =
47
+ template === "coding"
48
+ ? "positive, negative and zero addition assertions against edited add.sh"
49
+ : "an exact response assertion";
50
+ const prerequisites =
51
+ template === "coding"
52
+ ? "Bun and sandbox-exec on macOS, or bubblewrap/socat on Linux"
53
+ : "Bun";
54
+ return `# ${template} evaluation\n\nThis deterministic adapter exercises ${assertion}, without model credentials. Prerequisites: ${prerequisites}.\n\n\`\`\`sh\nsevro run --run-manifest-file run.json --dry\nsevro run --run-manifest-file run.json\nsevro report --latest\nSEVRO_TEMPLATE_FAIL=1 sevro run --run-manifest-file run.json\n\`\`\`\n\nThe negative control must report task=failed and exit 1. Dry preparation is not an assessment.\nEdit case.json and adapter.ts, prepare, then run again. Keep private .sevro evidence out of Git.\nFor a native model, replace the adapter route explicitly with --host, --model and --effort; Codex also requires --codex-auth-file. No model execution or native skill dispatch is established by this adapter.\n`;
55
+ }
56
+
57
+ async function initialize(template: string, destination: string) {
58
+ await reserveDirectory(destination);
59
+ try {
60
+ await cp(
61
+ join(import.meta.dir, "..", "..", "examples", "templates", template),
62
+ destination,
63
+ { recursive: true, errorOnExist: true, force: false },
64
+ );
65
+ await writeFile(
66
+ join(destination, "run.json"),
67
+ JSON.stringify(
68
+ { format: "sevro.run.v1", settings: templateSettings(template) },
69
+ null,
70
+ 2,
71
+ ) + "\n",
72
+ { flag: "wx" },
73
+ );
74
+ if (template === "repository-skill")
75
+ await writeFile(
76
+ join(destination, "command.json"),
77
+ JSON.stringify(templateCommand(destination)) + "\n",
78
+ { flag: "wx" },
79
+ );
80
+ await writeFile(join(destination, ".gitignore"), ".sevro/\n", {
81
+ flag: "wx",
82
+ });
83
+ await writeFile(join(destination, "README.md"), starterReadme(template), {
84
+ flag: "wx",
85
+ });
86
+ } catch (cause) {
87
+ await rm(destination, { recursive: true, force: true });
88
+ throw cause;
89
+ }
90
+ }
91
+
92
+ export async function initCommand(argv: string[]): Promise<number> {
93
+ try {
94
+ const { template, destination } = initOptions(argv);
95
+ await initialize(template, destination);
96
+ const quoted = `'${destination.replaceAll("'", "'\\''")}'`;
97
+ const native =
98
+ template === "repository-skill"
99
+ ? " --model YOUR_MODEL --effort YOUR_EFFORT --codex-auth-file /absolute/auth.json"
100
+ : "";
101
+ process.stdout.write(
102
+ `Created ${template} evaluation at ${destination}\nOpen README.md, then:\n cd ${quoted}\n sevro run --run-manifest-file run.json${native} --dry\n sevro run --run-manifest-file run.json${native}\n`,
103
+ );
104
+ return 0;
105
+ } catch (error) {
106
+ process.stderr.write(
107
+ `sevro: ${error instanceof Error ? error.message : "initialization failed"}\n`,
108
+ );
109
+ return 64;
110
+ }
111
+ }