specrails-core 5.6.0 → 6.0.1

package/dist/agent-runtime/compact/step.d.ts

@@ -26,7 +26,6 @@ export interface CompactEnv {
 }
 /** Shorthand for the guards: `on(env, 'empty-write')`. */
 export declare function on(env: Pick<CompactEnv, 'guardrails'>, id: GuardrailId): boolean;
-export declare const COMPACT_STANCE = "You are one step of a Specrails pipeline driven by a host program. Keep replies short and concrete. Never invent files, paths or commands: everything you state must come from the inputs or from tool results. When asked for JSON, reply with exactly one JSON object and nothing else.";
 /** One host-driven `openspec_workflow` call, visible in the host log like a model tool call. */
 export declare function openspecCall(env: CompactEnv, input: Parameters<OpenSpecTools['execute']>[0]): Promise<unknown>;
 /** A tool-less structured call with one tolerant retry; the object is validated by the caller. */

package/dist/agent-runtime/compact/step.js

@@ -6,7 +6,7 @@ import { record } from './chat-client.js';
 import { runToolLoop } from './guarded-loop.js';
 /** Shorthand for the guards: `on(env, 'empty-write')`. */
 export function on(env, id) { return guardrailEnabled(env.guardrails, id); }
-export const COMPACT_STANCE = 'You are one step of a Specrails pipeline driven by a host program. Keep replies short and concrete. Never invent files, paths or commands: everything you state must come from the inputs or from tool results. When asked for JSON, reply with exactly one JSON object and nothing else.';
+const COMPACT_STANCE = 'You are one step of a Specrails pipeline driven by a host program. Keep replies short and concrete. Never invent files, paths or commands: everything you state must come from the inputs or from tool results. When asked for JSON, reply with exactly one JSON object and nothing else.';
 /** One host-driven `openspec_workflow` call, visible in the host log like a model tool call. */
 export async function openspecCall(env, input) {
     env.onEvent?.(toolEvent('openspec_workflow', input));
@@ -133,4 +133,3 @@ export function strings(value, limit, each = 1000) {
     return Array.isArray(value) ? value.filter((item) => typeof item === 'string' && item.trim().length > 0).map(item => item.trim().slice(0, each)).slice(0, limit) : [];
 }
 export function text(value, fallback = '') { return typeof value === 'string' && value.trim() ? value.trim() : fallback; }
-//# sourceMappingURL=step.js.map

package/dist/agent-runtime/compact/test-reachability.js

@@ -98,7 +98,11 @@ function tokenReaches(token, file, root, discovery) {
  * file. Unknown ⇒ true: only an enumerating command can prove absence.
  */
 export function commandReachesTest(root, command, file) {
-    const target = posix(file);
+    // Path arguments are relative to the directory the command runs in (a
+    // package inside a larger checkout), while reported files are relative to
+    // the checkout: compare both from the command's directory.
+    const base = path.resolve(root, command.cwd ?? '.');
+    const target = posix(path.relative(base, path.resolve(root, file)));
     let judged = false;
     for (const tokens of segments(expandCommandText(root, command))) {
         if (!tokens.length)
@@ -117,7 +121,7 @@ export function commandReachesTest(root, command, file) {
         if (!paths.some(token => TEST_FILE_PATTERN.test(posix(token)) || token.includes('*') || token.endsWith('...')))
             continue;
         judged = true;
-        if (paths.some(token => tokenReaches(token, target, root, discovery)))
+        if (paths.some(token => tokenReaches(token, target, base, discovery)))
             return true;
     }
     return !judged;
@@ -144,4 +148,3 @@ export function unreachedTestFiles(root, commands, reported) {
 export function unreachedTestsReason(files) {
     return `Test file${files.length === 1 ? '' : 's'} written by the developer ${files.length === 1 ? 'is' : 'are'} not executed by any verification command: ${files.join(', ')}. A test nobody runs proves nothing. Wire ${files.length === 1 ? 'it' : 'them'} into the repository's test command (for example the package.json "test" script) or propose a verificationCheck that runs ${files.length === 1 ? 'it' : 'them'}, then make ${files.length === 1 ? 'it' : 'them'} pass.`;
 }
-//# sourceMappingURL=test-reachability.js.map

package/dist/agent-runtime/config.js

@@ -290,4 +290,3 @@ export function normalizeRuntimeConfig(input, options) {
     };
     return config;
 }
-//# sourceMappingURL=config.js.map

package/dist/agent-runtime/core-host.d.ts

@@ -1,11 +1,11 @@
-import { type PipelineContext } from '../installer/runtime/pipeline-state.js';
+import { type PipelineContext } from '../pipeline/pipeline-state.js';
 import { type ExecutorRegistry } from './executors.js';
 import type { AgentEvent, RuntimeConfig, AgentEventRole } from './executor-types.js';
 import { parseAgentObject } from './graph/artifacts.js';
 import { type RuntimeIdentity } from './runtime-identity.js';
 import type { WorkflowEvent, WorkflowSpan, WorkflowState } from './workflow-types.js';
 export declare const RUNTIME_API_VERSION = 1;
-export declare const CORE_WORKFLOW_VERSION = "6";
+export declare const CORE_WORKFLOW_VERSION = "7";
 export declare const CORE_PACKAGE_VERSION: string;
 export declare function coreRuntimeIdentity(): RuntimeIdentity;
 export { parseAgentObject };

package/dist/agent-runtime/core-host.js

@@ -2,7 +2,7 @@ import { fingerprint } from './durable-store.js';
 import { prepareOpenSpec, roleOpenSpecContext } from './openspec.js';
 import { existsSync, readFileSync } from 'node:fs';
 import path from 'node:path';
-import { fingerprintCandidate, initializePipeline, inspectPipeline, pipelineStateDirectory, validatePipelineContext, validateVerificationRequest, } from '../installer/runtime/pipeline-state.js';
+import { fingerprintCandidate, initializePipeline, inspectPipeline, pipelineStateDirectory, validatePipelineContext, validateVerificationRequest, } from '../pipeline/pipeline-state.js';
 import { normalizeRuntimeConfig } from './config.js';
 import { createExecutorRegistry } from './executors.js';
 import { archive, child, journal, parseAgentObject, SLUG } from './graph/artifacts.js';
@@ -15,7 +15,7 @@ import { assertEffortSupported } from './capabilities.js';
 import { runtimePackageIntegrity } from './runtime-identity.js';
 import { readWorkflowState, runWorkflow } from './workflow.js';
 export const RUNTIME_API_VERSION = 1;
-export const CORE_WORKFLOW_VERSION = '6';
+export const CORE_WORKFLOW_VERSION = '7';
 export const CORE_PACKAGE_VERSION = JSON.parse(readFileSync(new URL('../../package.json', import.meta.url), 'utf8')).version;
 export function coreRuntimeIdentity() {
     return { packageVersion: CORE_PACKAGE_VERSION, workflowVersion: CORE_WORKFLOW_VERSION, instructionsVersion: String(ROLE_INSTRUCTIONS_VERSION), packageIntegrity: runtimePackageIntegrity(), apiVersion: 1 };
@@ -137,4 +137,3 @@ export async function runCoreWorkflow(options) {
         },
     });
 }
-//# sourceMappingURL=core-host.js.map

package/dist/agent-runtime/durable-store.d.ts

@@ -9,7 +9,6 @@ export interface WorkflowEnvelope {
     state: WorkflowState;
     graph?: SerializedGraphStore;
 }
-export declare const ENVELOPE_FORMAT = 2;
 /** Reject non-JSON data instead of silently changing the fingerprint on serialization. */
 export declare function fingerprint(value: unknown): string;
 export declare function readWorkflowEnvelope(directory: string, runId: string): Promise<WorkflowEnvelope | null>;

package/dist/agent-runtime/durable-store.js

@@ -10,7 +10,7 @@ export class WorkflowStoreError extends Error {
         this.name = 'WorkflowStoreError';
     }
 }
-export const ENVELOPE_FORMAT = 2;
+const ENVELOPE_FORMAT = 2;
 function errorCode(error) {
     return typeof error === 'object' && error !== null && 'code' in error ? String(error.code) : undefined;
 }
@@ -202,4 +202,3 @@ export async function acquireWorkflowLease(directory, runId) {
             await rm(lease, { recursive: true, force: true });
     };
 }
-//# sourceMappingURL=durable-store.js.map

package/dist/agent-runtime/efficiency-summary.d.ts

@@ -1,4 +1,4 @@
-import { type PipelineContext, type PipelineCompletion } from '../installer/runtime/pipeline-state.js';
+import { type PipelineContext, type PipelineCompletion } from '../pipeline/pipeline-state.js';
 import type { RuntimeConfig } from './executor-types.js';
 import type { WorkflowState } from './workflow-types.js';
 /** Compact derived view. Durable invocation/evidence IDs remain the accounting authority. */
@@ -21,7 +21,7 @@ export declare function efficiencySummary(state: WorkflowState, context: Pipelin
     runId: string;
     workflowVersion: string;
     technicalAcceptance: string;
-    archive: import("../installer/runtime/pipeline-state.js").PhaseStatus;
+    archive: import("../pipeline/pipeline-state.js").PhaseStatus;
     delivery: "pending" | "complete" | "pending-host";
     roles: {
         role: "architect" | "developer" | "reviewer";

package/dist/agent-runtime/efficiency-summary.js

@@ -1,4 +1,4 @@
-import { readVerificationEvidence } from '../installer/runtime/pipeline-state.js';
+import { readVerificationEvidence } from '../pipeline/pipeline-state.js';
 import { runtimeEfficiency } from './efficiency.js';
 /** Compact derived view. Durable invocation/evidence IDs remain the accounting authority. */
 export function efficiencySummary(state, context, config, completion, identity) {
@@ -49,4 +49,3 @@ export function efficiencySummary(state, context, config, completion, identity)
         metrics: runtimeEfficiency(state),
     };
 }
-//# sourceMappingURL=efficiency-summary.js.map

package/dist/agent-runtime/efficiency-types.js

@@ -5,4 +5,3 @@ export function sumCacheUsage(values) {
     return Object.fromEntries(CACHE_TOKEN_KEYS.map(key => [key, values.some(value => value[key] == null)
             ? null : values.reduce((sum, value) => sum + (value[key] ?? 0), 0)]));
 }
-//# sourceMappingURL=efficiency-types.js.map

package/dist/agent-runtime/efficiency.js

@@ -32,4 +32,3 @@ export function runtimeEfficiency(state) {
     const pending = state.history.some(attempt => attempt.pendingInvocations?.length);
     return { schemaVersion: 1, total: { ...totals(state.history), durationMs: state.usage.durationMs, inputTokens: pending ? null : state.usage.inputTokens, outputTokens: pending ? null : state.usage.outputTokens, costUsd: pending ? null : state.usage.costUsd }, phases };
 }
-//# sourceMappingURL=efficiency.js.map

package/dist/agent-runtime/evaluation-corpus.js

@@ -5,4 +5,3 @@ export const EVALUATION_CORPUS = [
     { id: 'verification-correction', verification: 'assert.strictEqual(api.clamp(11, 0, 10), 10)', description: 'Clamp values to both boundaries, including an already valid value.', repositories: ['math'], source: 'exports.clamp = value => value', solution: 'exports.clamp = (value, min, max) => Math.min(max, Math.max(min, value))', defects: ['exports.clamp = (value, min, max) => Math.max(min, value)', 'exports.clamp = (value, min, max) => Math.min(max, value)'], oracle: 'assert.strictEqual(api.clamp(-1, 0, 10), 0); assert.strictEqual(api.clamp(11, 0, 10), 10); assert.strictEqual(api.clamp(5, 0, 10), 5);', correction: 'verification' },
     { id: 'review-correction', verification: 'assert.strictEqual(typeof api.unique, "function")', description: 'Deduplicate stable values while preserving the first occurrence order.', repositories: ['collections'], source: 'exports.unique = values => values', solution: 'exports.unique = values => [...new Set(values)]', defects: ['exports.unique = values => [...new Set(values)].sort()', 'exports.unique = values => values'], oracle: 'assert.deepStrictEqual(api.unique([3, 1, 3, 2, 1]), [3, 1, 2]); assert.deepStrictEqual(api.unique([]), []);', correction: 'review' },
 ];
-//# sourceMappingURL=evaluation-corpus.js.map

package/dist/agent-runtime/evaluation.d.ts

@@ -1,5 +1,5 @@
 import type { RuntimeConfig } from './executor-types.js';
-export interface EvaluationOptions {
+interface EvaluationOptions {
     output: string;
     config?: RuntimeConfig;
     maxCostUsd?: number;
@@ -41,3 +41,4 @@ export declare function runEvaluation(options: EvaluationOptions): Promise<{
     reportedSpendUsd: number;
     observations: Record<string, unknown>[];
 }>;
+export {};

package/dist/agent-runtime/evaluation.js

@@ -8,7 +8,7 @@ import { fingerprint } from './durable-store.js';
 import { runCoreWorkflow, coreRuntimeIdentity } from './core-host.js';
 import { ExecutorRegistry } from './executors.js';
 import { OpenSpecTools } from './openspec.js';
-import { validatePipelineContext } from '../installer/runtime/pipeline-state.js';
+import { validatePipelineContext } from '../pipeline/pipeline-state.js';
 import { runtimeEfficiency } from './efficiency.js';
 function oracle(test, file) {
     return spawnSync(process.execPath, ['-e', 'const assert = require("node:assert/strict"); const api = require(process.argv[1]);' + test.oracle, file], { encoding: 'utf8', timeout: 10000 }).status === 0;
@@ -168,4 +168,3 @@ export async function runEvaluation(options) {
         rmSync(root, { recursive: true, force: true });
     }
 }
-//# sourceMappingURL=evaluation.js.map

package/dist/agent-runtime/executor-types.d.ts

@@ -1,4 +1,4 @@
-import type { VerificationCommand } from '../installer/runtime/pipeline-state.js';
+import type { VerificationCommand } from '../pipeline/pipeline-state.js';
 import type { CacheTokenUsage } from './efficiency-types.js';
 import type { GuardrailSettings } from './guardrails.js';
 export declare const DEFAULT_AGENT_TIMEOUT_MS: number;

package/dist/agent-runtime/executor-types.js

@@ -31,4 +31,3 @@ export function validateAgentRequest(request) {
     if (request.signal?.aborted)
         throw new AgentExecutionError('Agent cancelled', 'aborted');
 }
-//# sourceMappingURL=executor-types.js.map

package/dist/agent-runtime/executors.js

@@ -45,4 +45,3 @@ export function createExecutorRegistry(input, options = {}) {
         registry.register(id, executor);
     return registry;
 }
-//# sourceMappingURL=executors.js.map

package/dist/agent-runtime/gemini-policy.js

@@ -33,4 +33,3 @@ export function assertGeminiAdminPolicyAvailable(options = {}) {
     if (entries.some(file => file.endsWith('.toml')))
         throw new AgentExecutionError('Gemini system policies disable per-run admin policy enforcement. Select another provider for architect/reviewer, or ask the system administrator to configure a compatible environment: ' + directory, 'provider_capability_unsupported');
 }
-//# sourceMappingURL=gemini-policy.js.map

package/dist/agent-runtime/graph/artifacts.d.ts

@@ -1,4 +1,4 @@
-import { type PipelineContext, type PipelineState, type VerificationCommand } from '../../installer/runtime/pipeline-state.js';
+import { type PipelineContext, type PipelineState, type VerificationCommand } from '../../pipeline/pipeline-state.js';
 import type { DesignConfidence } from './state.js';
 export declare const SLUG: RegExp;
 export declare function object(value: unknown): Record<string, unknown>;

package/dist/agent-runtime/graph/artifacts.js

@@ -2,7 +2,8 @@ import { artifactPath, hash, resolveOpenSpecCli, runOpenSpec } from '../openspec
 import { randomUUID } from 'node:crypto';
 import { cpSync, existsSync, lstatSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, renameSync, rmSync, writeFileSync } from 'node:fs';
 import path from 'node:path';
-import { checkArchive, pipelineStateDirectory, transitionPipeline, validateVerificationRequest, } from '../../installer/runtime/pipeline-state.js';
+import { checkArchive, pipelineStateDirectory, transitionPipeline, validateVerificationRequest, } from '../../pipeline/pipeline-state.js';
+import { assertProposalInScope, withScopeDefault } from '../change-scope.js';
 export const SLUG = /^[a-z0-9]+(?:-[a-z0-9]+)*$/;
 export function object(value) {
     if (!value || typeof value !== 'object' || Array.isArray(value))
@@ -105,8 +106,10 @@ export function proposedVerification(context, configured, verification) {
             throw new Error('Invalid proposed verification command');
         return { repositoryId: command.repositoryId, command: command.command, args: command.args, ...(typeof command.cwd === 'string' && command.cwd ? { cwd: command.cwd } : {}) };
     }).filter(command => !covered.has(command.repositoryId) && context.repositories.some(repository => repository.id === command.repositoryId));
+    for (const command of proposed)
+        assertProposalInScope(context, command, 'proposed verification command');
     if (proposed.length)
-        validateVerificationRequest(context, { kind: 'scoped', commands: proposed });
+        validateVerificationRequest(context, { kind: 'scoped', commands: proposed.map(command => withScopeDefault(context, command)) });
     return proposed;
 }
 /** Publish an archive prepared by the real CLI. The durable write set makes a
@@ -209,4 +212,3 @@ export async function archive(context, change, signal) {
             transitionPipeline(context, 'ci', 'skipped');
     }
 }
-//# sourceMappingURL=artifacts.js.map

package/dist/agent-runtime/graph/convergence.d.ts

@@ -0,0 +1,26 @@
+import type { VerificationReceipt } from '../../pipeline/pipeline-state.js';
+import type { WorkflowState } from '../workflow-types.js';
+/** One verify visit, as the convergence guard remembers it. */
+export interface VerifyOutcome {
+    at: string;
+    candidateHash: string;
+    planHash: string;
+    passed: boolean;
+    /** Normalized identity of the first failure: the command plus its output with volatile details removed. */
+    signature?: string;
+    summary?: string;
+}
+/** Verify visits since the latest explicit resume: a human continuing a blocked run grants a fresh budget, as for attempts. */
+export declare function outcomesSinceResume(history: readonly VerifyOutcome[], checkpoint: WorkflowState): VerifyOutcome[];
+/** Identity and one-line description of a failed receipt (or of a host finding that failed a green receipt). */
+export declare function describeFailure(receipt: Pick<VerificationReceipt, 'commands' | 'reason'>, roots: readonly string[], finding?: string): {
+    signature: string;
+    summary: string;
+};
+/**
+ * Why the failure just recorded (`current`, not yet in `history`) proves the
+ * correction loop is not converging, or undefined when another round can help.
+ */
+export declare function divergence(history: readonly VerifyOutcome[], current: VerifyOutcome): string | undefined;
+/** The same test for a candidate that has not been verified again yet: an unchanged failed candidate is not worth a new run. */
+export declare function unchangedFailure(history: readonly VerifyOutcome[], candidateHash: string, planHash: string): VerifyOutcome | undefined;

package/dist/agent-runtime/graph/convergence.js

@@ -0,0 +1,62 @@
+import { fingerprint } from '../durable-store.js';
+/** Verify visits since the latest explicit resume: a human continuing a blocked run grants a fresh budget, as for attempts. */
+export function outcomesSinceResume(history, checkpoint) {
+    const resumed = [...checkpoint.events].reverse().find(event => event.type === 'workflow_resumed');
+    return history.filter(entry => !resumed || entry.at >= resumed.timestamp);
+}
+function normalized(output, roots) {
+    let text = output.replace(/\u001b\[[0-9;?]*[A-Za-z]/g, '');
+    for (const root of [...roots].sort((a, b) => b.length - a.length))
+        text = text.split(root).join('<repo>');
+    return text.split(/\r?\n/)
+        .map(line => line.replace(/\b[0-9a-f]{7,64}\b/gi, '#').replace(/\d+(?:[.,]\d+)*/g, '0').replace(/\s+/g, ' ').trim())
+        .filter(Boolean).slice(-40).join('\n');
+}
+/** The last line that says what went wrong, bounded. */
+function headline(output) {
+    const lines = output.replace(/\u001b\[[0-9;?]*[A-Za-z]/g, '').split(/\r?\n/).map(line => line.trim()).filter(Boolean);
+    const telling = [...lines].reverse().find(line => /error|fail|✕|×|not found|cannot|denied|refused|forbidden|unauthori[sz]ed/i.test(line));
+    return (telling ?? lines.at(-1) ?? '').slice(0, 300);
+}
+/** Identity and one-line description of a failed receipt (or of a host finding that failed a green receipt). */
+export function describeFailure(receipt, roots, finding) {
+    if (finding)
+        return { signature: fingerprint({ finding: normalized(finding, roots) }), summary: finding.slice(0, 300) };
+    const failed = receipt.commands.find(command => command.exitCode !== 0);
+    if (!failed)
+        return { signature: fingerprint({ reason: receipt.reason ?? 'verification failed' }), summary: receipt.reason ?? 'verification failed' };
+    const name = failed.label ?? [failed.command, ...failed.args].join(' ');
+    const detail = headline(failed.output);
+    return {
+        signature: fingerprint({ check: failed.key ?? [failed.command, ...failed.args, failed.cwd], exitCode: failed.exitCode, output: normalized(failed.output, roots) }),
+        summary: `${name} exited ${failed.exitCode}${detail ? `: ${detail}` : ''}`.slice(0, 400),
+    };
+}
+/**
+ * Why the failure just recorded (`current`, not yet in `history`) proves the
+ * correction loop is not converging, or undefined when another round can help.
+ */
+export function divergence(history, current) {
+    if (current.passed || !current.signature)
+        return undefined;
+    const cycle = history.find(entry => !entry.passed && entry.candidateHash === current.candidateHash && entry.planHash === current.planHash);
+    if (cycle)
+        return `The correction round returned the change to a state that already failed verification (${current.summary}). Running the same checks again cannot pass.`;
+    let repeats = 0;
+    for (const entry of [...history].reverse()) {
+        if (entry.passed || entry.signature !== current.signature)
+            break;
+        repeats++;
+    }
+    if (repeats >= 2)
+        return `The same verification failure survived ${repeats} correction rounds unchanged (${current.summary}). It is most likely outside what this change can fix: a pre-existing failure, another package, or the environment.`;
+    const previous = history.at(-1);
+    if (previous?.passed && history.some(entry => !entry.passed && entry.signature === current.signature)) {
+        return `A review correction brought back a verification failure that an earlier correction had fixed (${current.summary}). The review request and the checks pull in opposite directions; decide which one applies before resuming.`;
+    }
+    return undefined;
+}
+/** The same test for a candidate that has not been verified again yet: an unchanged failed candidate is not worth a new run. */
+export function unchangedFailure(history, candidateHash, planHash) {
+    return [...history].reverse().find(entry => !entry.passed && entry.candidateHash === candidateHash && entry.planHash === planHash);
+}

package/dist/agent-runtime/graph/nodes.d.ts

@@ -1,5 +1,5 @@
 import { type OpenSpecRoleContext } from '../openspec.js';
-import { type AcceptanceReport, type PipelineContext } from '../../installer/runtime/pipeline-state.js';
+import { type AcceptanceReport, type PipelineContext } from '../../pipeline/pipeline-state.js';
 import type { AgentEventRole, AgentResult, AgentRole, RuntimeConfig } from '../executor-types.js';
 import { type FrozenCriterion } from '../prompts.js';
 import type { WorkflowNode, WorkflowState } from '../workflow-types.js';