orizu 0.5.21 → 0.5.22

package/README.md

@@ -9,7 +9,7 @@ Orizu is a platform that helps you build continually learning agents and other L
 There are 3 key pillars to Orizu:
 1. **Human evaluation:** We consider human evaluation to be the gold standard, particularly when the humans involved are people with deep context and understanding of the output being evaluated. This is often the cross-functional team behind a product (product managers, engineers, designers, content writers, researchers, etc) or external experts (lawyers, engineers, writers, etc). Any good evaluation system relies on the presence of such high quality judgment to compare to. We provide tools that simplify the process of collecting the feedback from this audience.
 2. **Auto evaluation:** Ultimately, human experts can only review so much. As your data scales, you need to encode their knowledge, their opinions, their taste and judgment into auto-evaluations. We automate this for you using the data we help you collect above.
-3. **Optimization:** We turn your evals into an asset for you by building an automated hill-climbing workflow. We've started with prompt optimization because multiple studies and our own experience with our customers has shown strong measurable lifts with even relatively small datasets. Over time, we'll introduce additional capabilities, including optimizing your skills and other contextual information, and even model weights if your model provider allows finetuning.
+3. **Optimization:** We turn your evals into an asset for you by building an automated hill-climbing workflow. We've started with instruction optimization because multiple studies and our own experience with our customers has shown strong measurable lifts with even relatively small datasets. Over time, we'll introduce additional capabilities, including optimizing your skills and other contextual information, and even model weights if your model provider allows finetuning.
 
 ## CLI and Agent Skills
 
@@ -109,6 +109,9 @@ Agents are first-class users of this CLI: **every command supports `--json`**
 (as a prefix, `orizu --json teams list`, or a trailing flag) and emits a single
 machine-readable JSON document instead of formatted text. Discover the full
 command surface, including global options, with `orizu capabilities --json`.
+Hosted agents may read and sync instruction sets and prepare complete manifests,
+but a human curator using the local CLI with a user token must run instruction
+create, push, shape, archive, restore, and production/default pointer mutations.
 
 ## Authentication
 
@@ -186,19 +189,19 @@ orizu logout
     <td><code>orizu projects create --name "Support QA" --team ops-eval</code></td>
   </tr>
   <tr>
-    <td colspan="2"><strong>Prompts</strong></td>
+    <td colspan="2"><strong>Instruction sets and prompt reads</strong></td>
   </tr>
   <tr>
     <td>List prompts</td>
     <td><code>orizu prompts list --project ops-eval/support-qa [--status active|archived|all]</code></td>
   </tr>
   <tr>
-    <td>Archive or restore a prompt</td>
-    <td><code>orizu prompts archive &lt;promptIdOrName&gt; --project ops-eval/support-qa</code><br><code>orizu prompts restore &lt;promptIdOrName&gt; --project ops-eval/support-qa</code></td>
+    <td>Archive or restore an instruction set (human/local CLI only)</td>
+    <td><code>orizu instructions archive &lt;slug-or-exact-name&gt; --project ops-eval/support-qa</code><br><code>orizu instructions restore &lt;slug-or-exact-name&gt; --project ops-eval/support-qa</code></td>
   </tr>
   <tr>
-    <td>Push a prompt with a report</td>
-    <td><code>orizu prompts push ./prompts/support --project ops-eval/support-qa --report-file ./prompt-report.md</code></td>
+    <td>Push an instruction-set manifest (human/local CLI only)</td>
+    <td><code>orizu instructions push ./orizu.instruction-set.json --project ops-eval/support-qa --set &lt;slug-or-exact-name&gt;</code></td>
   </tr>
   <tr>
     <td>Upload a report for an existing prompt version</td>

package/dist/gepa-engine-dispatch.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"gepa-engine-dispatch.d.ts","sourceRoot":"","sources":["../src/gepa-engine-dispatch.ts"],"names":[],"mappings":"AAEA,MAAM,MAAM,UAAU,GAAG,UAAU,GAAG,QAAQ,CAAA;AAE9C,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,UAAU,CAAA;IAClB,IAAI,EAAE,MAAM,EAAE,CAAA;IACd,WAAW,EAAE,MAAM,CAAC,UAAU,CAAA;IAC9B,MAAM,EAAE,MAAM,CAAA;CACf;AAuTD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,UAAU,GAAG,kBAAkB,CAmBtH"}
+{"version":3,"file":"gepa-engine-dispatch.d.ts","sourceRoot":"","sources":["../src/gepa-engine-dispatch.ts"],"names":[],"mappings":"AAIA,MAAM,MAAM,UAAU,GAAG,UAAU,GAAG,QAAQ,CAAA;AAE9C,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,UAAU,CAAA;IAClB,IAAI,EAAE,MAAM,EAAE,CAAA;IACd,WAAW,EAAE,MAAM,CAAC,UAAU,CAAA;IAC9B,MAAM,EAAE,MAAM,CAAA;CACf;AAyUD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,UAAU,GAAG,kBAAkB,CAmBtH"}

package/dist/gepa-engine-dispatch.js

@@ -1,4 +1,5 @@
 import { existsSync, readFileSync } from 'fs';
+import { normalizedSkilledProposerConfig } from './skilled-proposer-config.js';
 const VALUE_OPTIONS = [
     { flag: '--optimizer-version-id', environment: 'ORIZU_OPTIMIZER_VERSION_ID' },
     { flag: '--candidate-version-id', environment: 'ORIZU_PROMPT_VERSION_ID' },
@@ -148,6 +149,7 @@ const CONTROLLED_ENVIRONMENT_NAMES = new Set([
     // translation prevents an inherited environment from selecting a proposer
     // on an otherwise byte-for-byte normal official-GEPA launch.
     'ORIZU_CANDIDATE_PROPOSER',
+    'ORIZU_SKILLED_PROPOSER_CONFIG',
     'ORIZU_COMPONENT_SELECTOR',
 ]);
 function optionValueOrThrow(args, index, flag, allowEmpty = false) {
@@ -173,6 +175,7 @@ function translateOfficialOptions(args, project, environment) {
     }
     connectorEnvironment.ORIZU_PROJECT = project;
     let metadata;
+    let candidateProposerConfig;
     let hasInstructionSet = false;
     let hasCandidateVersion = false;
     const budgetFlags = [];
@@ -205,6 +208,12 @@ function translateOfficialOptions(args, project, environment) {
             index = parsed.nextIndex;
             continue;
         }
+        if (flag === '--candidate-proposer-config') {
+            const parsed = optionValueOrThrow(args, index, flag);
+            candidateProposerConfig = parsed.value;
+            index = parsed.nextIndex;
+            continue;
+        }
         const option = VALUE_OPTIONS_BY_FLAG.get(flag);
         if (option) {
             const parsed = optionValueOrThrow(args, index, flag, option.allowEmpty);
@@ -266,6 +275,16 @@ function translateOfficialOptions(args, project, environment) {
         && connectorEnvironment.ORIZU_CANDIDATE_PROPOSER !== 'skilled-proposer') {
         throw new Error('--proposal-max-calls and --proposal-max-tokens require --candidate-proposer skilled-proposer');
     }
+    if (candidateProposerConfig !== undefined) {
+        if (connectorEnvironment.ORIZU_CANDIDATE_PROPOSER !== 'skilled-proposer') {
+            throw new Error('--candidate-proposer-config requires --candidate-proposer skilled-proposer');
+        }
+        connectorEnvironment.ORIZU_SKILLED_PROPOSER_CONFIG = normalizedSkilledProposerConfig(candidateProposerConfig);
+    }
+    if (connectorEnvironment.ORIZU_CANDIDATE_PROPOSER === 'skilled-proposer'
+        && connectorEnvironment.ORIZU_REFLECTION_PROMPT_TEMPLATE !== undefined) {
+        throw new Error('--reflection-prompt-template is incompatible with --candidate-proposer skilled-proposer');
+    }
     const reflectionModel = connectorEnvironment.ORIZU_REFLECTION_MODEL ?? 'anthropic/claude-opus-4-7';
     if (!reflectionModel.startsWith('openai/') && connectorEnvironment.ORIZU_REFLECTION_MAX_TOKENS === undefined) {
         throw new Error('--reflection-max-tokens is required for Anthropic reflection models');
@@ -289,7 +308,7 @@ function validatedMetadata(raw) {
 export function dispatchGepaEngine(args, project, environment) {
     const selected = removeEngine(args);
     if (selected.engine === 'legacy') {
-        for (const flag of ['--instruction-set', '--model-config', '--component-selector', '--max-candidate-proposals', '--candidate-proposer', '--proposal-max-calls', '--proposal-max-tokens']) {
+        for (const flag of ['--instruction-set', '--model-config', '--component-selector', '--max-candidate-proposals', '--candidate-proposer', '--candidate-proposer-config', '--proposal-max-calls', '--proposal-max-tokens']) {
             if (selected.args.some(argument => argument === flag || argument.startsWith(`${flag}=`))) {
                 throw new Error(`${flag} is supported by the official GEPA engine only`);
             }

package/dist/gepa-python-paths.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"gepa-python-paths.d.ts","sourceRoot":"","sources":["../src/gepa-python-paths.ts"],"names":[],"mappings":"AAYA,wBAAgB,qBAAqB,CACnC,UAAU,EAAE,MAAM,EAAE,EACpB,UAAU,GAAE,CAAC,SAAS,EAAE,MAAM,KAAK,OAAoB,GACtD,MAAM,GAAG,IAAI,CAEf;AAED,wBAAgB,0BAA0B,IAAI,MAAM,GAAG,IAAI,CAM1D;AAED,wBAAgB,mCAAmC,IAAI,MAAM,GAAG,IAAI,CAMnE;AAED,wBAAgB,6BAA6B,IAAI,MAAM,GAAG,IAAI,CAS7D;AAED,wBAAgB,wBAAwB,CAAC,kBAAkB,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,EAAE,CASzF"}
+{"version":3,"file":"gepa-python-paths.d.ts","sourceRoot":"","sources":["../src/gepa-python-paths.ts"],"names":[],"mappings":"AAaA,wBAAgB,qBAAqB,CACnC,UAAU,EAAE,MAAM,EAAE,EACpB,UAAU,GAAE,CAAC,SAAS,EAAE,MAAM,KAAK,OAAoB,GACtD,MAAM,GAAG,IAAI,CAEf;AAED,wBAAgB,0BAA0B,IAAI,MAAM,GAAG,IAAI,CAM1D;AAED,wBAAgB,mCAAmC,IAAI,MAAM,GAAG,IAAI,CAMnE;AAED,wBAAgB,6BAA6B,IAAI,MAAM,GAAG,IAAI,CA0B7D;AAED,wBAAgB,wBAAwB,CAAC,kBAAkB,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,EAAE,CASzF"}

package/dist/gepa-python-paths.js

@@ -1,4 +1,5 @@
 import { existsSync } from 'fs';
+import { execFileSync } from 'child_process';
 import { fileURLToPath } from 'url';
 import { isOrizuSourceCheckout } from '../scripts/orizu-source-checkout.mjs';
 function prefersLiveSource() {
@@ -23,13 +24,31 @@ export function bundledOrizuGepaConnectorPythonPath() {
     return resolveGepaPythonPath(candidates);
 }
 export function bundledOfficialGepaPythonPath() {
-    const sourceArchive = fileURLToPath(new URL('../gepa-python-source.zip', import.meta.url));
     const vendor = fileURLToPath(new URL('../vendor/gepa-python/src', import.meta.url));
-    // The checked-in pinned wheel is directly importable from PYTHONPATH in a
-    // source checkout. Published packages intentionally omit that archive and
-    // must resolve only the verified tree materialized by prepack.
-    const candidates = prefersLiveSource() ? [sourceArchive, vendor] : [vendor];
-    return resolveGepaPythonPath(candidates);
+    if (prefersLiveSource()) {
+        const requiredVendorFiles = [
+            fileURLToPath(new URL('../vendor/gepa-python/manifest.json', import.meta.url)),
+            fileURLToPath(new URL('../vendor/gepa-python/src/gepa/__init__.py', import.meta.url)),
+            fileURLToPath(new URL('../vendor/gepa-python/src/gepa/adapters/optimize_anything_adapter/optimize_anything_adapter.py', import.meta.url)),
+        ];
+        if (!requiredVendorFiles.every(existsSync)) {
+            const vendoringScript = fileURLToPath(new URL('../scripts/vendor-gepa-python.mjs', import.meta.url));
+            try {
+                execFileSync(process.execPath, [vendoringScript], { stdio: 'pipe' });
+            }
+            catch (error) {
+                const details = error instanceof Error ? error.message : String(error);
+                throw new Error(`Unable to materialize bundled official GEPA: ${details}`);
+            }
+            if (!requiredVendorFiles.every(existsSync)) {
+                throw new Error('Unable to materialize bundled official GEPA: extracted tree is incomplete');
+            }
+        }
+    }
+    // zipimport cannot resolve GEPA's namespace-package adapters on every
+    // supported Python minor. Both source and published CLIs execute only the
+    // verified tree materialized from the pinned wheel.
+    return resolveGepaPythonPath([vendor]);
 }
 export function getGepaPythonPathEntries(existingPythonPath) {
     const pythonPathEntries = [

package/dist/help.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"help.d.ts","sourceRoot":"","sources":["../src/help.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB,UAAU,CAAC,EAAE,OAAO,CAAA;IACpB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;CACnB;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,EAAE,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,CAAC,EAAE,MAAM,EAAE,EAAE,CAAA;IACpB,OAAO,CAAC,EAAE,YAAY,EAAE,CAAA;IACxB,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAA;CACpB;AAoCD,eAAO,MAAM,YAAY,EAAE,aAAa,EAumCvC,CAAA;AAUD,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,aAAa,GAAG,SAAS,CAMxE;AAED,wBAAgB,qBAAqB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,aAAa,EAAE,CAWrE;AAOD,wBAAgB,cAAc,IAAI,MAAM,CA0CvC;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,CAoBtD;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,GAAG,MAAM,CA4B5D;AAED,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,CAgBxD;AAED,wBAAgB,eAAe,CAAC,OAAO,EAAE,MAAM;;;;;;;;;;;;;;;;;;;;;EAoB9C"}
+{"version":3,"file":"help.d.ts","sourceRoot":"","sources":["../src/help.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,CAAC,EAAE,OAAO,CAAA;IAClB,UAAU,CAAC,EAAE,OAAO,CAAA;IACpB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;CACnB;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,EAAE,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,CAAC,EAAE,MAAM,EAAE,EAAE,CAAA;IACpB,OAAO,CAAC,EAAE,YAAY,EAAE,CAAA;IACxB,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAA;CACpB;AAoCD,eAAO,MAAM,YAAY,EAAE,aAAa,EA6nCvC,CAAA;AAUD,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,aAAa,GAAG,SAAS,CAMxE;AAED,wBAAgB,qBAAqB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,aAAa,EAAE,CAWrE;AAOD,wBAAgB,cAAc,IAAI,MAAM,CA0CvC;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,CAoBtD;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,GAAG,MAAM,CA4B5D;AAED,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,CAgBxD;AAED,wBAAgB,eAAe,CAAC,OAAO,EAAE,MAAM;;;;;;;;;;;;;;;;;;;;;EAoB9C"}

package/dist/help.js

@@ -64,6 +64,40 @@ export const COMMAND_DOCS = [
     },
     { path: ['instructions', 'archive'], aliases: [['instruction-sets', 'archive']], usage: 'orizu instructions archive <slug-or-exact-name> --project <team/project> [--json]', summary: 'Archive an instruction set without changing resolution or sync.', group: 'Instruction sets', options: [{ name: '--project <team/project>', help: 'Project slug.', required: true }, { name: '--json', help: 'Print JSON.' }], examples: ['orizu instructions archive planner --project core/evals'] },
     { path: ['instructions', 'restore'], aliases: [['instruction-sets', 'restore']], usage: 'orizu instructions restore <slug-or-exact-name> --project <team/project> [--json]', summary: 'Restore an archived instruction set.', group: 'Instruction sets', options: [{ name: '--project <team/project>', help: 'Project slug.', required: true }, { name: '--json', help: 'Print JSON.' }], examples: ['orizu instructions restore planner --project core/evals'] },
+    {
+        path: ['instructions', 'scorers', 'set-headline'],
+        aliases: [['instruction-sets', 'scorers', 'set-headline']],
+        usage: 'orizu instructions scorers set-headline <set> --key <component-key> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]',
+        summary: 'Bind the headline scorer for one instruction-set component.',
+        group: 'Instruction sets',
+        options: [
+            { name: '--key <component-key>', help: 'Instruction-set component key.', required: true },
+            { name: '--scorer-version <id>', help: 'Scorer version to bind.', required: true },
+            { name: '--dataset-version <id>', help: 'Default dataset version.' },
+            { name: '--split-set <id>', help: 'Default dataset split set.' },
+            { name: '--split <name>', help: 'Default split name.' },
+            { name: '--project <team/project>', help: 'Project slug.' },
+            { name: '--json', help: 'Print JSON.' },
+        ],
+        examples: ['orizu instructions scorers set-headline planner --key system --scorer-version <id> --project core/evals'],
+    },
+    {
+        path: ['instructions', 'scorers', 'add'],
+        aliases: [['instruction-sets', 'scorers', 'add']],
+        usage: 'orizu instructions scorers add <set> --key <component-key> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]',
+        summary: 'Track an additional scorer for one instruction-set component.',
+        group: 'Instruction sets',
+        options: [
+            { name: '--key <component-key>', help: 'Instruction-set component key.', required: true },
+            { name: '--scorer-version <id>', help: 'Scorer version to bind.', required: true },
+            { name: '--dataset-version <id>', help: 'Default dataset version.' },
+            { name: '--split-set <id>', help: 'Default dataset split set.' },
+            { name: '--split <name>', help: 'Default split name.' },
+            { name: '--project <team/project>', help: 'Project slug.' },
+            { name: '--json', help: 'Print JSON.' },
+        ],
+        examples: ['orizu instructions scorers add planner --key system --scorer-version <id> --project core/evals'],
+    },
     {
         path: ['login'],
         usage: 'orizu login [--no-prompt-if-logged-in]',
@@ -388,18 +422,6 @@ export const COMMAND_DOCS = [
         summary: 'Move or set a non-production prompt label; use instructions profiles promote for production.',
         group: 'Prompts and judges',
     },
-    {
-        path: ['prompts', 'scorers', 'set-headline'],
-        usage: 'orizu prompts scorers set-headline <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]',
-        summary: 'Bind the headline scorer for a prompt.',
-        group: 'Scorers and runners',
-    },
-    {
-        path: ['prompts', 'scorers', 'add'],
-        usage: 'orizu prompts scorers add <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]',
-        summary: 'Track an additional scorer for a prompt.',
-        group: 'Scorers and runners',
-    },
     {
         path: ['judges', 'list'],
         usage: 'orizu judges list --project <team/project> [--status active|archived|all]',
@@ -553,8 +575,8 @@ export const COMMAND_DOCS = [
     },
     {
         path: ['optimizations', 'run-gepa'],
-        usage: 'orizu optimizations run-gepa --project <team/project> --optimizer-version-id <id> (--candidate-version-id <id> | --instruction-set <name> --model-config <identity> [--component-selector round-robin|all]) --runner-version-id <id> --candidate-runner-dir <dir> --scorer-version-id <id> --scorer-runner-version-id <id> --scorer-runner-dir <dir> [--engine official|legacy] [--scorer-input-contract gepa|flat_row] [--scorer-candidate-field <row-field>] [--allow-degenerate-seed] --dataset-version-id <id> --split-set-id <id> [--train-split train] [--val-split validation] [--budget auto|light|medium|heavy | --max-metric-calls <n> | --max-full-evals <n> | --max-iterations <n> | --max-candidate-proposals <n>] [--num-threads auto|N] [--reflection-retry-attempts N] [--reflection-http-timeout-seconds N] [--log-dir logs]',
-        summary: 'Run official GEPA locally and stream events; pass --engine legacy only for the frozen safety hatch. --instruction-set requires --model-config and cannot be combined with --candidate-version-id; round-robin updates one component-map entry per round and all updates every component. --max-candidate-proposals is official-engine-only. Validates the scorer contract on the seed before iterating; judge runners built for flat-row score runs need --scorer-input-contract flat_row (ALI-1158).',
+        usage: 'orizu optimizations run-gepa --project <team/project> --optimizer-version-id <id> (--candidate-version-id <id> | --instruction-set <name> --model-config <identity> [--component-selector round-robin|all]) --runner-version-id <id> --candidate-runner-dir <dir> --scorer-version-id <id> --scorer-runner-version-id <id> --scorer-runner-dir <dir> [--engine official|legacy] [--candidate-proposer skilled-proposer [--candidate-proposer-config <@file|path>] [--proposal-max-calls <n>] [--proposal-max-tokens <n>]] [--scorer-input-contract gepa|flat_row] [--scorer-candidate-field <row-field>] [--allow-degenerate-seed] --dataset-version-id <id> --split-set-id <id> [--train-split train] [--val-split validation] [--budget auto|light|medium|heavy | --max-metric-calls <n> | --max-full-evals <n> | --max-iterations <n> | --max-candidate-proposals <n>] [--num-threads auto|N] [--reflection-max-tokens N] [--reflection-retry-attempts N] [--reflection-http-timeout-seconds N] [--log-dir logs]',
+        summary: 'Run official GEPA locally and stream events; pass --engine legacy only for the frozen safety hatch. Skilled-proposer is official-only, its config requires that selection, and the selection cannot be combined with --reflection-prompt-template. --instruction-set requires --model-config and cannot be combined with --candidate-version-id; round-robin updates one component-map entry per round and all updates every component. --max-candidate-proposals is official-engine-only. Validates the scorer contract on the seed before iterating; judge runners built for flat-row score runs need --scorer-input-contract flat_row (ALI-1158).',
         group: 'Optimizations',
     },
     {

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AA2FgF,OAAO,EAAE,kBAAkB,EAAE,MAAM,6BAA6B,CAAA;AA8BhJ,OAAO,EAAE,iBAAiB,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAA;AAgU5E,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEvD;AAUD,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAOnD;AAyBD,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAE5D;AAcD,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,GAAG,CAqB5E;AA6BD,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAgBtD;AA+jKD,wBAAgB,kBAAkB,CAAC,SAAS,EAAE,MAAM,GAAG;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE,CA0B1G;AAED,eAAO,MAAM,iBAAiB,SAAU,CAAA;AACxC,eAAO,MAAM,uBAAuB,QAAkB,CAAA;AACtD,eAAO,MAAM,oBAAoB,aAgB/B,CAAA;AAEF,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,iBAAiB,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC,UAAU,CAaxH;AAWD,wBAAsB,wBAAwB,CAAC,eAAe,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,IAAI,CAAA;CAAE,CAAC,CA6B3H;AA0LD,wBAAsB,IAAI,CAAC,OAAO,WAAwB,iBA0gBzD"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AA2FgF,OAAO,EAAE,kBAAkB,EAAE,MAAM,6BAA6B,CAAA;AA8BhJ,OAAO,EAAE,iBAAiB,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAA;AAgU5E,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEvD;AAUD,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAOnD;AAyBD,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAE5D;AAcD,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,GAAG,CAqB5E;AA6BD,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAgBtD;AAkhKD,wBAAgB,kBAAkB,CAAC,SAAS,EAAE,MAAM,GAAG;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE,CA0B1G;AAED,eAAO,MAAM,iBAAiB,SAAU,CAAA;AACxC,eAAO,MAAM,uBAAuB,QAAkB,CAAA;AACtD,eAAO,MAAM,oBAAoB,aAgB/B,CAAA;AAEF,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,iBAAiB,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC,UAAU,CAaxH;AAWD,wBAAsB,wBAAwB,CAAC,eAAe,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,IAAI,CAAA;CAAE,CAAC,CA6B3H;AA0LD,wBAAsB,IAAI,CAAC,OAAO,WAAwB,iBAwgBzD"}

package/dist/index.js

@@ -65,7 +65,7 @@ import { dispatchGepaEngine } from './gepa-engine-dispatch.js';
 import { resolveGepaInstructionSetProfileVersion } from './instruction-set-gepa-launch.js';
 import { getGepaPythonCommand } from './gepa-python-command.js';
 import { getGepaPythonPathEntries } from './gepa-python-paths.js';
-import { prepareSkilledProposerLaunch } from './skilled-proposer-launch.js';
+import { prepareSkilledProposerLaunch, spawnSkilledProposerChild } from './skilled-proposer-launch.js';
 export { parseJsonResponse, sanitizeTerminalText } from './json-response.js';
 function getErrorMessage(error) {
     return error instanceof Error ? error.message : String(error);
@@ -1205,39 +1205,6 @@ function refusePromptMutation(replacement) {
     process.exitCode = 1;
     throw new Error(`Use: orizu instructions ${replacement}`);
 }
-async function bindPromptScorer(role) {
-    const promptId = getPositionalArg(3);
-    const project = getArg('--project') || await resolveProjectSlug(null);
-    const scorerVersionId = getArg('--scorer-version');
-    if (!promptId || !scorerVersionId) {
-        throw new Error(`Usage: orizu prompts scorers ${role === 'headline' ? 'set-headline' : 'add'} <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]`);
-    }
-    const response = await authedFetch(`/api/cli/prompts/scorers?project=${encodeURIComponent(project)}`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-            promptId,
-            scorerVersionId,
-            role,
-            datasetVersionId: getArg('--dataset-version') || undefined,
-            splitSetId: getArg('--split-set') || undefined,
-            splitName: getArg('--split') || undefined,
-        }),
-    });
-    if (!response.ok) {
-        throw new Error(`Failed to bind prompt scorer: ${await response.text()}`);
-    }
-    const data = await parseJsonResponse(response, 'Prompt scorer bind');
-    if (hasJsonFlag()) {
-        printJson(data);
-        return;
-    }
-    printLine(`${role === 'headline' ? 'Set headline' : 'Added'} scorer ${sanitizeTerminalText(scorerVersionId)} for ${sanitizeTerminalText(promptId)}`);
-    if (data.owner) {
-        printLine(`Owner: ${sanitizeTerminalText(promptId)} -> ` +
-            `${sanitizeTerminalText(data.owner.instructionSetSlug)} / ${sanitizeTerminalText(data.owner.componentKey)}`);
-    }
-}
 function getPositionalArg(index) {
     const value = cliArgs[index];
     return value && !value.startsWith('--') ? value : null;
@@ -1827,7 +1794,7 @@ async function runGepaOptimization() {
             dispatch.environment.ORIZU_INSTRUCTION_SET_PROFILE_VERSION_ID = instructionSetProfileVersionId;
         selectedEngine = dispatch.engine;
         const launch = prepareSkilledProposerLaunch(python, dispatch.engine, dispatch.environment);
-        result = spawnSync(launch.python, ['-m', dispatch.module, ...dispatch.args], { stdio: 'inherit', env: launch.environment });
+        result = await spawnSkilledProposerChild(launch.python, ['-m', dispatch.module, ...dispatch.args], dispatch.engine, launch.environment);
     }
     finally {
         verified.cleanup();
@@ -4854,12 +4821,10 @@ export async function main(rawArgs = process.argv.slice(2)) {
         return;
     }
     if (command === 'prompts' && subcommand === 'scorers' && cliArgs[2] === 'set-headline') {
-        await bindPromptScorer('headline');
-        return;
+        refusePromptMutation('scorers set-headline <set> --key <component-key> --scorer-version <id> --project <team/project>');
     }
     if (command === 'prompts' && subcommand === 'scorers' && cliArgs[2] === 'add') {
-        await bindPromptScorer('tracked');
-        return;
+        refusePromptMutation('scorers add <set> --key <component-key> --scorer-version <id> --project <team/project>');
     }
     if (command === 'judges' && subcommand === 'list') {
         await listJudges();

package/dist/instruction-set-loader.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"instruction-set-loader.d.ts","sourceRoot":"","sources":["../src/instruction-set-loader.ts"],"names":[],"mappings":"AAGA,qBAAa,yBAA0B,SAAQ,KAAK;IACtC,QAAQ,CAAC,IAAI,EAAE,MAAM;gBAAZ,IAAI,EAAE,MAAM;CAClC;AAOD,wBAAgB,kBAAkB,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,mBAAmB,EAAE,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAwBvH"}
+{"version":3,"file":"instruction-set-loader.d.ts","sourceRoot":"","sources":["../src/instruction-set-loader.ts"],"names":[],"mappings":"AAGA,qBAAa,yBAA0B,SAAQ,KAAK;IACtC,QAAQ,CAAC,IAAI,EAAE,MAAM;gBAAZ,IAAI,EAAE,MAAM;CAClC;AA2DD,wBAAgB,kBAAkB,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,mBAAmB,EAAE,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAsBvH"}

package/dist/instruction-set-loader.js

@@ -1,4 +1,4 @@
-import { existsSync, realpathSync, readFileSync, statSync } from 'node:fs';
+import { existsSync, realpathSync, readFileSync, readdirSync, statSync } from 'node:fs';
 import { resolve, sep } from 'node:path';
 export class InstructionSetLoaderError extends Error {
     code;
@@ -9,11 +9,63 @@ export class InstructionSetLoaderError extends Error {
     }
 }
 function inside(root, candidate) { return candidate === root || candidate.startsWith(`${root}${sep}`); }
-export function loadInstructionSet(directory, name, modelConfigIdentity) {
-    const unresolvedRoot = resolve(directory, name);
-    if (!existsSync(unresolvedRoot) || !statSync(unresolvedRoot).isDirectory())
+function manifestMatchesReference(candidate, reference) {
+    try {
+        const manifest = JSON.parse(readFileSync(resolve(candidate, 'manifest.json'), 'utf8'));
+        return manifest.name === reference || manifest.slug === reference;
+    }
+    catch {
+        return false;
+    }
+}
+function resolveSetRoot(directory, reference) {
+    let directoryRoot;
+    try {
+        directoryRoot = realpathSync(directory);
+    }
+    catch {
         throw new InstructionSetLoaderError('instruction_set_not_synced');
-    const setRoot = realpathSync(unresolvedRoot);
+    }
+    const direct = resolve(directoryRoot, reference);
+    try {
+        if (inside(directoryRoot, direct) && existsSync(direct) && statSync(direct).isDirectory()) {
+            const realDirect = realpathSync(direct);
+            if (inside(directoryRoot, realDirect)) {
+                try {
+                    const manifest = JSON.parse(readFileSync(resolve(realDirect, 'manifest.json'), 'utf8'));
+                    if (manifest.name === reference || manifest.slug === reference) {
+                        return realDirect;
+                    }
+                }
+                catch {
+                    // Preserve the exact-path loader's specific missing/invalid manifest error.
+                    return realDirect;
+                }
+            }
+        }
+    }
+    catch {
+        // A stale or unreadable direct-path candidate must not shadow a valid set.
+    }
+    const matches = readdirSync(directoryRoot, { withFileTypes: true }).flatMap(entry => {
+        try {
+            if (entry.name.startsWith('.') || !entry.isDirectory())
+                return [];
+            const candidate = realpathSync(resolve(directoryRoot, entry.name));
+            return inside(directoryRoot, candidate) && manifestMatchesReference(candidate, reference)
+                ? [candidate]
+                : [];
+        }
+        catch {
+            return [];
+        }
+    });
+    if (matches.length !== 1)
+        throw new InstructionSetLoaderError(matches.length > 1 ? 'instruction_set_reference_ambiguous' : 'instruction_set_not_synced');
+    return matches[0];
+}
+export function loadInstructionSet(directory, name, modelConfigIdentity) {
+    const setRoot = resolveSetRoot(directory, name);
     const manifestPath = resolve(setRoot, 'manifest.json');
     if (!existsSync(manifestPath))
         throw new InstructionSetLoaderError('instruction_set_manifest_missing');

package/dist/instruction-sets-cli.d.ts

@@ -1,4 +1,4 @@
-import { renameSync, writeFileSync } from 'node:fs';
+import { renameSync, rmSync, writeFileSync } from 'node:fs';
 export interface InstructionSetsCommandIo {
     json: boolean;
     print: (line: string) => void;
@@ -23,6 +23,8 @@ export interface SyncProfile {
     production: SyncMaterial | null;
 }
 export interface SyncSet {
+    projectId?: string;
+    instructionSetId?: string;
     name: string;
     slug?: string;
     description?: string | null;
@@ -34,7 +36,8 @@ export interface SyncSet {
 export interface SyncFileOps {
     writeFileSync: typeof writeFileSync;
     renameSync?: typeof renameSync;
+    rmSync?: typeof rmSync;
 }
-export declare function syncToDisk(out: string, set: SyncSet, fileOps?: SyncFileOps): void;
+export declare function syncToDisk(out: string, set: SyncSet, fileOps?: SyncFileOps): string;
 export declare function instructionSetsCommand(args: string[], io: InstructionSetsCommandIo): Promise<number>;
 //# sourceMappingURL=instruction-sets-cli.d.ts.map

package/dist/instruction-sets-cli.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"instruction-sets-cli.d.ts","sourceRoot":"","sources":["../src/instruction-sets-cli.ts"],"names":[],"mappings":"AAGA,OAAO,EAAuC,UAAU,EAAoB,aAAa,EAAE,MAAM,SAAS,CAAA;AAI1G,MAAM,WAAW,wBAAwB;IACvC,IAAI,EAAE,OAAO,CAAA;IACb,KAAK,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAA;IAC7B,kBAAkB,EAAE,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;CACjE;AA+BD,MAAM,WAAW,aAAa;IAAG,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAE;AAC5G,MAAM,WAAW,YAAY;IAAG,gBAAgB,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,mBAAmB,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAA;CAAE;AAC/K,MAAM,WAAW,WAAW;IAAG,mBAAmB,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,YAAY,GAAG,IAAI,CAAA;CAAE;AACnH,MAAM,WAAW,OAAO;IAAG,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,KAAK,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,YAAY,CAAC;IAAC,QAAQ,EAAE,WAAW,EAAE,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE;AAM7K,MAAM,WAAW,WAAW;IAAG,aAAa,EAAE,OAAO,aAAa,CAAC;IAAC,UAAU,CAAC,EAAE,OAAO,UAAU,CAAA;CAAE;AAoCpG,wBAAgB,UAAU,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,OAAO,GAAE,WAA+B,QAyC7F;AAED,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,wBAAwB,GAAG,OAAO,CAAC,MAAM,CAAC,CA+K1G"}
+{"version":3,"file":"instruction-sets-cli.d.ts","sourceRoot":"","sources":["../src/instruction-sets-cli.ts"],"names":[],"mappings":"AAGA,OAAO,EAAuC,UAAU,EAAE,MAAM,EAAY,aAAa,EAAE,MAAM,SAAS,CAAA;AAI1G,MAAM,WAAW,wBAAwB;IACvC,IAAI,EAAE,OAAO,CAAA;IACb,KAAK,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAA;IAC7B,kBAAkB,EAAE,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;CACjE;AA+BD,MAAM,WAAW,aAAa;IAAG,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAE;AAC5G,MAAM,WAAW,YAAY;IAAG,gBAAgB,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,mBAAmB,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAA;CAAE;AAC/K,MAAM,WAAW,WAAW;IAAG,mBAAmB,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,YAAY,GAAG,IAAI,CAAA;CAAE;AACnH,MAAM,WAAW,OAAO;IACtB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IAC3B,KAAK,EAAE,MAAM,EAAE,CAAA;IACf,OAAO,EAAE,YAAY,CAAA;IACrB,QAAQ,EAAE,WAAW,EAAE,CAAA;IACvB,UAAU,CAAC,EAAE,MAAM,EAAE,CAAA;CACtB;AASD,MAAM,WAAW,WAAW;IAC1B,aAAa,EAAE,OAAO,aAAa,CAAA;IACnC,UAAU,CAAC,EAAE,OAAO,UAAU,CAAA;IAC9B,MAAM,CAAC,EAAE,OAAO,MAAM,CAAA;CACvB;AA0FD,wBAAgB,UAAU,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,OAAO,GAAE,WAA+B,UAsF7F;AAED,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,wBAAwB,GAAG,OAAO,CAAC,MAAM,CAAC,CAqN1G"}