@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.
- package/CONTRIBUTING.md +22 -6
- package/README.md +5 -1
- package/docs/README.md +19 -15
- package/docs/architecture.md +138 -17
- package/docs/assets/README.md +5 -4
- package/docs/builtin-graders-v1.md +17 -3
- package/docs/codex-host.md +155 -49
- package/docs/contracts.md +2 -1
- package/docs/creating-cases.md +9 -2
- package/docs/development-cli.md +187 -19
- package/docs/documentation-quality.md +12 -2
- package/docs/evaluation-workflow.md +110 -0
- package/docs/extension-protocol-v1.md +26 -3
- package/docs/getting-started.md +23 -2
- package/docs/guide-evaluation.md +33 -1
- package/docs/identity-v1.md +25 -2
- package/docs/installing.md +11 -11
- package/docs/native-controls.md +76 -5
- package/docs/native-hosts.md +196 -7
- package/docs/reading-results.md +102 -0
- package/docs/recipes.md +489 -0
- package/docs/releases.md +31 -12
- package/docs/report-v1.md +29 -1
- package/docs/results-v1.md +141 -4
- package/docs/run-settings-v1.md +147 -0
- package/docs/runtime-v1.md +307 -18
- package/docs/troubleshooting.md +12 -12
- package/docs/typescript-quality.md +89 -26
- package/examples/recipes/candidate.ts +26 -0
- package/examples/recipes/cases/coding-change.json +21 -0
- package/examples/recipes/cases/host-tools.json +21 -0
- package/examples/recipes/cases/mise.json +21 -0
- package/examples/recipes/cases/node.json +21 -0
- package/examples/recipes/cases/python-uv.json +21 -0
- package/examples/recipes/cases/semantic.json +17 -0
- package/examples/recipes/coding-change/check.ts +23 -0
- package/examples/recipes/extension-artifacts.ts +65 -0
- package/examples/recipes/extension-methods.ts +174 -0
- package/examples/recipes/extension.ts +37 -0
- package/examples/recipes/repository-skill/.agents/skills/startup-guide/SKILL.md +7 -0
- package/examples/recipes/repository-skill/.claude/skills/startup-guide/SKILL.md +7 -0
- package/examples/recipes/repository-skill/README.md +3 -0
- package/examples/recipes/run-inputs.ts +188 -0
- package/examples/recipes/run-options.ts +131 -0
- package/examples/recipes/run.ts +63 -0
- package/examples/recipes/runtime/empty.json +1 -0
- package/examples/recipes/runtime/goals.json +1 -0
- package/examples/recipes/runtime/mise.json +6 -0
- package/examples/recipes/runtime/path.json +1 -0
- package/examples/recipes/runtime/plugins.json +4 -0
- package/examples/recipes/runtime/python-uv.json +10 -0
- package/examples/recipes/semantic-judge.ts +29 -0
- package/examples/templates/coding/adapter.ts +17 -0
- package/examples/templates/coding/case.json +15 -0
- package/examples/templates/repository-skill/.agents/skills/startup-guide/SKILL.md +6 -0
- package/examples/templates/repository-skill/extension-data.ts +64 -0
- package/examples/templates/repository-skill/extension.ts +31 -0
- package/examples/templates/response/adapter.ts +12 -0
- package/examples/templates/response/case.json +13 -0
- package/package.json +11 -5
- package/schemas/report-v1.schema.json +9 -0
- package/schemas/run-evidence-v1.schema.json +2 -1
- package/schemas/runtime-v1.schema.json +63 -7
- package/src/advisory.ts +2 -2
- package/src/candidate-artifacts.ts +206 -0
- package/src/candidate-codex-state.ts +185 -0
- package/src/case-validation.ts +73 -50
- package/src/cli/cancellation.ts +20 -0
- package/src/cli/command.ts +87 -0
- package/src/cli/completion.ts +15 -0
- package/src/cli/demo.ts +60 -0
- package/src/cli/evaluation.ts +55 -0
- package/src/cli/extension.ts +79 -0
- package/src/cli/help.ts +198 -0
- package/src/cli/hosts.ts +55 -0
- package/src/cli/init-settings.ts +30 -0
- package/src/cli/init.ts +111 -0
- package/src/{cli-invocation.ts → cli/invocation.ts} +214 -218
- package/src/cli/loading.ts +151 -0
- package/src/cli/preparation.ts +154 -0
- package/src/cli/presentation.ts +93 -0
- package/src/cli/progress.ts +6 -0
- package/src/cli/run-manifest.ts +128 -0
- package/src/cli/run-options.ts +403 -0
- package/src/cli/run-overrides.ts +69 -0
- package/src/cli/settings.ts +154 -0
- package/src/cli/summary.ts +43 -0
- package/src/cli.ts +13 -591
- package/src/contract-types.ts +94 -0
- package/src/engine.ts +13 -5
- package/src/evaluation/advisory.ts +150 -0
- package/src/{evaluation-check-plan.ts → evaluation/check-plan.ts} +103 -67
- package/src/evaluation/collection.ts +88 -0
- package/src/evaluation/context.ts +10 -0
- package/src/{evaluation-grading-host.ts → evaluation/grading-host.ts} +5 -5
- package/src/evaluation/grading.ts +235 -0
- package/src/{evaluation-invocation-plan.ts → evaluation/invocation-plan.ts} +147 -58
- package/src/{evaluation-plan.ts → evaluation/plan.ts} +208 -141
- package/src/evaluation/progress.ts +19 -0
- package/src/evaluation/resources.ts +241 -0
- package/src/evaluation/run.ts +206 -0
- package/src/{evaluation-semantic.ts → evaluation/semantic.ts} +133 -115
- package/src/evaluation/trial.ts +800 -0
- package/src/evaluation/types.ts +207 -0
- package/src/evidence/host-results.ts +85 -0
- package/src/evidence/retained-artifacts.ts +18 -0
- package/src/evidence/trial-writer.ts +203 -0
- package/src/evidence/usage.ts +17 -0
- package/src/extension-client.ts +34 -96
- package/src/extension-session.ts +24 -48
- package/src/{advisory-fixture.ts → fixtures/advisory.ts} +15 -76
- package/src/{fixture-bin.ts → fixtures/bin.ts} +1 -1
- package/src/{fixture-cleanup.ts → fixtures/cleanup.ts} +13 -7
- package/src/fixtures/create.ts +183 -0
- package/src/{generated-fixture.ts → fixtures/generated.ts} +10 -53
- package/src/{git-hooks.ts → fixtures/git-hooks.ts} +1 -1
- package/src/{preparation.ts → fixtures/preparation.ts} +6 -16
- package/src/{repository-fixture.ts → fixtures/repository.ts} +30 -64
- package/src/{fixture-setup.ts → fixtures/setup.ts} +33 -51
- package/src/{fixture-tools.ts → fixtures/tools.ts} +1 -1
- package/src/fixtures/workspace-state.ts +212 -0
- package/src/generated/cli-result.d.cts +2 -1
- package/src/generated/extension.d.cts +2 -1
- package/src/generated/report.cjs +1 -1
- package/src/generated/report.d.cts +2 -1
- package/src/generated/run-evidence.cjs +1 -1
- package/src/generated/run-evidence.d.cts +2 -1
- package/src/generated/runtime.cjs +1 -1
- package/src/generated/runtime.d.cts +2 -1
- package/src/git.ts +93 -0
- package/src/graders/git-head.ts +29 -84
- package/src/graders/shell.ts +57 -68
- package/src/host-diagnostic-redaction.ts +52 -0
- package/src/host-diagnostics.ts +331 -0
- package/src/hosts/claude-continuation.ts +57 -40
- package/src/hosts/claude-credential.ts +160 -63
- package/src/hosts/claude-events.ts +34 -18
- package/src/hosts/claude-keychain.ts +78 -21
- package/src/hosts/claude-login-recovery.ts +85 -0
- package/src/hosts/claude-login-refresh.ts +178 -0
- package/src/hosts/claude-login-store.ts +162 -0
- package/src/hosts/claude-login.ts +107 -0
- package/src/hosts/claude-native-goal.ts +15 -12
- package/src/hosts/claude-process.ts +10 -50
- package/src/hosts/claude-refresh-lock.ts +81 -0
- package/src/hosts/claude-repository-invocation.ts +9 -34
- package/src/hosts/claude-settings.ts +91 -14
- package/src/hosts/claude-skill-activations.ts +169 -0
- package/src/hosts/claude-state.ts +46 -16
- package/src/hosts/claude-tool-calls.ts +13 -10
- package/src/hosts/claude.ts +122 -73
- package/src/hosts/codex-app-server.ts +78 -52
- package/src/hosts/codex-configuration.ts +5 -10
- package/src/hosts/codex-exec-controls.ts +516 -0
- package/src/hosts/codex-marketplace.ts +12 -10
- package/src/hosts/codex-native-calls.ts +15 -122
- package/src/hosts/codex-process.ts +10 -54
- package/src/hosts/codex-profile.ts +50 -21
- package/src/hosts/codex-session-files.ts +1 -9
- package/src/hosts/codex-state.ts +141 -19
- package/src/hosts/codex.ts +221 -179
- package/src/hosts/isolation-common.ts +15 -11
- package/src/hosts/isolation-roots.ts +13 -20
- package/src/hosts/linux-sandbox.ts +27 -3
- package/src/hosts/mac-sandbox.ts +111 -18
- package/src/hosts/native-controls.ts +72 -22
- package/src/hosts/native-request.ts +55 -0
- package/src/hosts/process-lifecycle.ts +221 -45
- package/src/hosts/protected-worktrees.ts +1 -1
- package/src/hosts/skill-hook-catalog.ts +131 -0
- package/src/hosts/skill-hook-handler.ts +374 -0
- package/src/hosts/skill-hooks.ts +577 -0
- package/src/instrumentation.ts +7 -8
- package/src/native-transcript-state.ts +22 -3
- package/src/native-transcript-view.ts +52 -13
- package/src/native-transcripts.ts +34 -15
- package/src/paths.ts +22 -0
- package/src/process-lifecycle.ts +150 -0
- package/src/provenance.ts +27 -7
- package/src/prune.ts +300 -0
- package/src/report.ts +63 -36
- package/src/resolved-case.ts +1 -1
- package/src/results-lock.ts +42 -0
- package/src/results.ts +14 -18
- package/src/run-owner.ts +4 -4
- package/src/{runtime-config.ts → runtime/config.ts} +182 -44
- package/src/{runtime-discovery-apple.ts → runtime/discovery-apple.ts} +3 -3
- package/src/{runtime-discovery-providers.ts → runtime/discovery-providers.ts} +3 -9
- package/src/{runtime-discovery.ts → runtime/discovery.ts} +4 -4
- package/src/{runtime-hook-processes.ts → runtime/hook-processes.ts} +1 -1
- package/src/{runtime-hooks.ts → runtime/hooks.ts} +33 -22
- package/src/{runtime-paths.ts → runtime/paths.ts} +6 -18
- package/src/{runtime-query.ts → runtime/query.ts} +5 -5
- package/src/{runtime-seeds.ts → runtime/seeds.ts} +3 -3
- package/src/runtime/shared-cache.ts +130 -0
- package/src/{runtime-state.ts → runtime/state.ts} +28 -11
- package/src/schema-types.ts +73 -161
- package/src/shell.ts +4 -0
- package/src/transcript-retention.ts +146 -0
- package/src/value-guards.ts +22 -0
- package/docs/assets/sevro-logo.png +0 -0
- package/docs/assets/sevro-symbol.svg +0 -1
- package/docs/assets/sevro-wordmark.svg +0 -1
- package/docs/cache-ownership-validation.md +0 -49
- package/docs/claude-continuation-validation.md +0 -47
- package/docs/concurrency-validation.md +0 -86
- package/docs/configuration-root-validation.md +0 -70
- package/docs/engine-slice.md +0 -114
- package/docs/evidence/runtime-startup-30/README.md +0 -185
- package/docs/evidence/runtime-startup-30/after.json +0 -319
- package/docs/evidence/runtime-startup-30/baseline-timeout.jsonl +0 -2
- package/docs/evidence/runtime-startup-30/before.json +0 -403
- package/docs/evidence/runtime-startup-30/consumer-smoke.json +0 -63
- package/docs/evidence/runtime-startup-30/full-policy-diagnostic.jsonl +0 -3
- package/docs/evidence/runtime-startup-30/minimal-policy-diagnostic.jsonl +0 -3
- package/docs/evidence/runtime-startup-30/plugin-template-assessment.json +0 -20
- package/docs/evidence/runtime-startup-30/seedless-timeout.jsonl +0 -2
- package/docs/evidence.md +0 -17
- package/docs/fixture-cleanup-validation.md +0 -70
- package/docs/guide-verification.json +0 -239
- package/src/evaluation-context.ts +0 -257
- package/src/evaluation-fixture.ts +0 -313
- package/src/evaluation-run.ts +0 -198
- package/src/evaluation-trial.ts +0 -946
- package/src/evaluation-types.ts +0 -149
- package/src/hosts/claude-nested-skills.ts +0 -272
- package/src/hosts/codex-command-output.ts +0 -510
- package/src/hosts/codex-executor-command.ts +0 -291
- package/src/hosts/codex-skill-reads.ts +0 -609
- /package/src/{trial-scheduler.ts → evaluation/scheduler.ts} +0 -0
package/src/cli/init.ts
ADDED
|
@@ -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
|
+
}
|