@tea-agent/loop-agent 0.16.0 → 0.16.1-beta.0

package/dist/infrastructure/evaluation/proposer-store.js

@@ -1,78 +0,0 @@
-import { access, mkdir, mkdtemp, readdir, readFile, rename, rm } from "node:fs/promises";
-import path from "node:path";
-import { proposerRunIdSchema, proposerRunSchema, } from "../../application/evaluation/types.js";
-import { sha256Hex } from "../../application/evaluation/candidate-hash.js";
-import { writeJsonAtomic, writeTextAtomic } from "../harness/atomic-write.js";
-import { EVALUATION_ROOT } from "./store.js";
-const SAFE_ID = /^[A-Za-z0-9][A-Za-z0-9._-]*$/;
-const MANIFEST_INTEGRITY_FILE = "manifest.sha256";
-export function assertProposerRunId(value) {
-    proposerRunIdSchema.parse(value);
-}
-export function proposerRunDir(repoRoot, proposerRunId) {
-    assertProposerRunId(proposerRunId);
-    return path.join(repoRoot, EVALUATION_ROOT, "proposer-runs", proposerRunId);
-}
-export function proposerRunManifestPath(repoRoot, proposerRunId) {
-    return path.join(proposerRunDir(repoRoot, proposerRunId), "manifest.json");
-}
-async function pathExists(filePath) {
-    try {
-        await access(filePath);
-        return true;
-    }
-    catch {
-        return false;
-    }
-}
-function serialize(record) {
-    return `${JSON.stringify(record, null, 2)}\n`;
-}
-export async function writeProposerRun(input) {
-    const parsed = proposerRunSchema.parse(input.run);
-    const root = path.dirname(proposerRunDir(input.repoRoot, parsed.proposerRunId));
-    await mkdir(root, { recursive: true });
-    const stagingDir = await mkdtemp(path.join(root, `.${parsed.proposerRunId}.register-`));
-    const manifestPath = proposerRunManifestPath(input.repoRoot, parsed.proposerRunId);
-    try {
-        if (await pathExists(manifestPath)) {
-            throw new Error(`proposer run already exists: ${parsed.proposerRunId}`);
-        }
-        await writeJsonAtomic(path.join(stagingDir, "manifest.json"), parsed, {
-            repoRoot: input.repoRoot,
-        });
-        await writeTextAtomic(path.join(stagingDir, MANIFEST_INTEGRITY_FILE), `${sha256Hex(serialize(parsed))}\n`, { repoRoot: input.repoRoot });
-        await rename(stagingDir, proposerRunDir(input.repoRoot, parsed.proposerRunId));
-    }
-    catch (error) {
-        await rm(stagingDir, { recursive: true, force: true });
-        throw error;
-    }
-    return manifestPath;
-}
-export async function readProposerRun(repoRoot, proposerRunId) {
-    assertProposerRunId(proposerRunId);
-    const manifestPath = proposerRunManifestPath(repoRoot, proposerRunId);
-    const rawText = await readFile(manifestPath, "utf-8");
-    const expected = (await readFile(path.join(proposerRunDir(repoRoot, proposerRunId), MANIFEST_INTEGRITY_FILE), "utf-8")).trim();
-    if (sha256Hex(rawText) !== expected) {
-        throw new Error(`proposer run integrity mismatch for ${proposerRunId}`);
-    }
-    return proposerRunSchema.parse(JSON.parse(rawText));
-}
-export async function listProposerRunIds(repoRoot) {
-    const root = path.join(repoRoot, EVALUATION_ROOT, "proposer-runs");
-    try {
-        const entries = await readdir(root, { withFileTypes: true });
-        return entries
-            .filter((entry) => entry.isDirectory() && SAFE_ID.test(entry.name))
-            .map((entry) => entry.name)
-            .sort();
-    }
-    catch (error) {
-        const code = error.code;
-        if (code === "ENOENT")
-            return [];
-        throw error;
-    }
-}

package/dist/worker/outcomes/evidence-tokens.js

@@ -1,29 +0,0 @@
-/**
- * Shared evidence / artifact identity helpers for Outcome gates, Feature review,
- * and Feature Verification Bundle projection.
- */
-/** Canonical required-output / required_evidence token for shell exit-zero. */
-export const SHELL_VERIFICATION_TOKEN = "shell_verification";
-/**
- * Accept the canonical underscore form and the legacy hyphen form used in early
- * fullstack packets so dual-coverage and bundle evidence stay aligned.
- */
-export function isShellVerificationToken(token) {
-    return token === SHELL_VERIFICATION_TOKEN || token === "shell-verification";
-}
-/** Match an evidence token against artifact `kind` or `schemaId`. */
-export function artifactMatchesEvidenceToken(artifact, token) {
-    return artifact.kind === token || artifact.schemaId === token;
-}
-/**
- * Derive stable `kind` + versioned `schemaId` from a DAG structured-artifact
- * schema id (e.g. `backend-test-result-v1` → kind `backend-test-result`).
- * Unversioned ids keep kind === schemaId.
- */
-export function splitStructuredArtifactIdentity(schemaId) {
-    const match = /^(.*)-v\d+$/.exec(schemaId);
-    if (match?.[1]) {
-        return { kind: match[1], schemaId };
-    }
-    return { kind: schemaId, schemaId };
-}

package/dist/workflows/dag/budget-enforcement.js

@@ -1,67 +0,0 @@
-import { writeFile } from "node:fs/promises";
-import path from "node:path";
-import { checkBudgetPreNode, createBudgetLedger, formatBudgetReportMarkdown, isHardBudgetBreached, recordNodeBudgetSample, skipPendingNodesForBudgetBreach, } from "../../application/evaluation/budget.js";
-export function initRunBudgetLedger(state, budget) {
-    if (!budget)
-        return;
-    state.budget = structuredClone(budget);
-    state.budgetLedger = createBudgetLedger(budget);
-}
-function wallTimeMs(state, now = Date.now()) {
-    return Math.max(0, now - new Date(state.startedAt).getTime());
-}
-function repairPasses(state) {
-    return state.convergence?.currentPass ?? 0;
-}
-export function applyHardBudgetBreach(state, breach) {
-    if (!state.budgetLedger || state.budgetLedger.mode !== "hard")
-        return;
-    skipPendingNodesForBudgetBreach(state.nodes, breach);
-    state.failureCategory = "budget_breach";
-}
-export function preflightBudgetOrBreach(state) {
-    const ledger = state.budgetLedger;
-    if (!ledger)
-        return undefined;
-    const breach = checkBudgetPreNode(ledger, {
-        wallTimeMs: wallTimeMs(state),
-        repairPasses: repairPasses(state),
-    });
-    if (breach && isHardBudgetBreached(ledger)) {
-        applyHardBudgetBreach(state, breach);
-    }
-    return breach;
-}
-export function recordFinishedNodeBudget(state, node, contextChars) {
-    const ledger = state.budgetLedger;
-    if (!ledger)
-        return undefined;
-    const breach = recordNodeBudgetSample(ledger, {
-        nodeId: node.id,
-        tokensUsed: node.tokensUsed,
-        contextChars,
-        wallTimeMs: wallTimeMs(state),
-        repairPasses: repairPasses(state),
-    });
-    if (breach && isHardBudgetBreached(ledger)) {
-        applyHardBudgetBreach(state, breach);
-    }
-    return breach;
-}
-export async function writeBudgetLedgerArtifacts(runDir, ledger) {
-    if (!ledger)
-        return;
-    const jsonPath = path.join(runDir, "budget-ledger.json");
-    const mdPath = path.join(runDir, "budget-report.md");
-    await writeFile(jsonPath, `${JSON.stringify(ledger, null, 2)}\n`, "utf8");
-    await writeFile(mdPath, formatBudgetReportMarkdown(ledger), "utf8");
-}
-export function assertFrozenBudget(specBudget, stateBudget, runId) {
-    if (!specBudget && !stateBudget)
-        return;
-    if (!specBudget ||
-        !stateBudget ||
-        JSON.stringify(specBudget) !== JSON.stringify(stateBudget)) {
-        throw new Error(`dag run ${runId} budget drifted between run.json and state.json; refuse to continue`);
-    }
-}

package/dist/workflows/dag/context-policy.js

@@ -1,137 +0,0 @@
-import { DEFAULT_SKILL_INSTRUCTION_MAX_CHARS, DEFAULT_SKILL_INSTRUCTION_TOTAL_MAX_CHARS, } from "./skill-instructions.js";
-import { MAX_UPSTREAM_CHARS } from "./prompt.js";
-import { resolveDagNodeSkills } from "./skills.js";
-export const CONTEXT_POLICY_IDS = [
-    "baseline-v1",
-    "role-specialized-v1",
-];
-export const DEFAULT_CONTEXT_POLICY_ID = "baseline-v1";
-function roleOrUndefined(task) {
-    return task.role;
-}
-function pickByRole(table, role) {
-    if (role && table[role] !== undefined)
-        return table[role];
-    return table.default;
-}
-class BaselineContextPolicy {
-    id = "baseline-v1";
-    description = "Current DAG context assembly: shared upstream char budget, role skill defaults, learned patterns only for implementer.";
-    resolveSkills(spec, task) {
-        return resolveDagNodeSkills(spec, task);
-    }
-    resolveSkillInstructionBudget(task) {
-        return {
-            includeLearnedPatterns: task.role === "implementer",
-            perSkillMaxChars: DEFAULT_SKILL_INSTRUCTION_MAX_CHARS,
-            totalMaxChars: DEFAULT_SKILL_INSTRUCTION_TOTAL_MAX_CHARS,
-        };
-    }
-    resolveMaxUpstreamChars(_task) {
-        return MAX_UPSTREAM_CHARS;
-    }
-    toManifest() {
-        return {
-            schemaVersion: 1,
-            policyId: this.id,
-            description: this.description,
-            knobs: {
-                maxUpstreamCharsByRole: { default: MAX_UPSTREAM_CHARS },
-                includeLearnedPatternsRoles: ["implementer"],
-                perSkillMaxCharsByRole: {
-                    default: DEFAULT_SKILL_INSTRUCTION_MAX_CHARS,
-                },
-                totalMaxCharsByRole: {
-                    default: DEFAULT_SKILL_INSTRUCTION_TOTAL_MAX_CHARS,
-                },
-            },
-        };
-    }
-}
-/**
- * First A/B challenger: keep skill name resolution identical to baseline, but
- * specialize upstream / skill-instruction budgets by role so scouts see less
- * noise and implementers retain more upstream + learned patterns surface.
- */
-class RoleSpecializedContextPolicy {
-    id = "role-specialized-v1";
-    description = "Role-specialized upstream and skill-instruction budgets; skill names still resolve via baseline merge order.";
-    maxUpstreamCharsByRole = {
-        default: MAX_UPSTREAM_CHARS,
-        scout: 1_200,
-        reviewer: 1_200,
-        implementer: 3_000,
-        verifier: 1_600,
-        closeout: 1_600,
-        planner: MAX_UPSTREAM_CHARS,
-        supervisor: MAX_UPSTREAM_CHARS,
-    };
-    perSkillMaxCharsByRole = {
-        default: DEFAULT_SKILL_INSTRUCTION_MAX_CHARS,
-        scout: 2_500,
-        implementer: 3_500,
-    };
-    totalMaxCharsByRole = {
-        default: DEFAULT_SKILL_INSTRUCTION_TOTAL_MAX_CHARS,
-        scout: 10_000,
-        implementer: 14_000,
-    };
-    learnedPatternRoles = new Set([
-        "implementer",
-        "closeout",
-    ]);
-    resolveSkills(spec, task) {
-        return resolveDagNodeSkills(spec, task);
-    }
-    resolveSkillInstructionBudget(task) {
-        const role = roleOrUndefined(task);
-        return {
-            includeLearnedPatterns: role
-                ? this.learnedPatternRoles.has(role)
-                : false,
-            perSkillMaxChars: pickByRole(this.perSkillMaxCharsByRole, role),
-            totalMaxChars: pickByRole(this.totalMaxCharsByRole, role),
-        };
-    }
-    resolveMaxUpstreamChars(task) {
-        return pickByRole(this.maxUpstreamCharsByRole, roleOrUndefined(task));
-    }
-    toManifest() {
-        return {
-            schemaVersion: 1,
-            policyId: this.id,
-            description: this.description,
-            knobs: {
-                maxUpstreamCharsByRole: this.maxUpstreamCharsByRole,
-                includeLearnedPatternsRoles: [...this.learnedPatternRoles],
-                perSkillMaxCharsByRole: this.perSkillMaxCharsByRole,
-                totalMaxCharsByRole: this.totalMaxCharsByRole,
-            },
-        };
-    }
-}
-const POLICIES = {
-    "baseline-v1": new BaselineContextPolicy(),
-    "role-specialized-v1": new RoleSpecializedContextPolicy(),
-};
-export function isContextPolicyId(value) {
-    return CONTEXT_POLICY_IDS.includes(value);
-}
-export function getContextPolicy(policyId) {
-    return POLICIES[policyId];
-}
-export function listContextPolicies() {
-    return CONTEXT_POLICY_IDS.map((id) => POLICIES[id]);
-}
-export function resolveContextPolicyId(spec) {
-    const raw = spec.defaults?.contextPolicyId;
-    if (!raw)
-        return DEFAULT_CONTEXT_POLICY_ID;
-    if (!isContextPolicyId(raw)) {
-        throw new Error(`unknown contextPolicyId "${raw}"; expected one of ${CONTEXT_POLICY_IDS.join(", ")}`);
-    }
-    return raw;
-}
-export function resolveContextPolicy(spec) {
-    return getContextPolicy(resolveContextPolicyId(spec));
-}

package/docs/templates/evaluation/campaign-budget-v1.json

@@ -1,12 +0,0 @@
-{
-	"schemaVersion": 1,
-	"mode": "hard",
-	"limits": {
-		"maxTokens": 2000000,
-		"maxWallTimeMs": 86400000,
-		"maxExecutorCalls": 500,
-		"maxRepairPasses": 50,
-		"maxConcurrency": 2,
-		"maxContextChars": 120000
-	}
-}

package/docs/templates/evaluation/campaign-dogfood-v0.json

@@ -1,24 +0,0 @@
-{
-  "schemaVersion": 1,
-  "campaignId": "camp-dogfood-context-policy-ab",
-  "createdAt": "2026-07-19T00:00:00.000Z",
-  "status": "prepared",
-  "policyVersion": "promotion-policy-v1",
-  "incumbentCandidateId": "REPLACE_INCUMBENT_CANDIDATE_ID",
-  "challengerCandidateIds": ["REPLACE_CHALLENGER_CANDIDATE_ID"],
-  "corpusId": "corpus-dogfood-v0",
-  "corpusHash": "REPLACE_AFTER_CORPUS_REGISTER",
-  "privateVerifierId": "pv-dogfood-v0",
-  "privateVerifierManifestHash": "REPLACE_AFTER_PRIVATE_VERIFIER_REGISTER",
-  "budget": {
-    "schemaVersion": 1,
-    "mode": "hard",
-    "limits": {
-      "maxExecutorCalls": 500,
-      "maxWallTimeMs": 86400000,
-      "maxRepairPasses": 50,
-      "maxConcurrency": 2
-    }
-  },
-  "notes": "M4 dogfood: single-variable Context Policy A/B. Replace candidate/corpus/privateVerifier hashes after register. held_out cells are promotion_gate_only."
-}

package/docs/templates/evaluation/campaign-evidence-v1.json

@@ -1,44 +0,0 @@
-{
-  "schemaVersion": 1,
-  "campaignId": "campaign-dogfood-v0",
-  "challengerCandidateId": "cand-challenger-example",
-  "cells": [
-    {
-      "candidateId": "cand-baseline-example",
-      "taskRef": "synthetic/private-1",
-      "seed": 1,
-      "split": "private",
-      "verifyPassed": false
-    },
-    {
-      "candidateId": "cand-challenger-example",
-      "taskRef": "synthetic/private-1",
-      "seed": 1,
-      "split": "private",
-      "verifyPassed": true
-    },
-    {
-      "candidateId": "cand-baseline-example",
-      "taskRef": "synthetic/heldout-1",
-      "seed": 1,
-      "split": "held_out",
-      "verifyPassed": true
-    },
-    {
-      "candidateId": "cand-challenger-example",
-      "taskRef": "synthetic/heldout-1",
-      "seed": 1,
-      "split": "held_out",
-      "verifyPassed": true
-    }
-  ],
-  "diffs": [],
-  "verifyCommands": {
-    "incumbent": ["npm test", "npm run typecheck"],
-    "challenger": ["npm test", "npm run typecheck"]
-  },
-  "budgetStatus": {
-    "incumbent": "ok",
-    "challenger": "ok"
-  }
-}

package/docs/templates/evaluation/context-policy-baseline-v1.json

@@ -1,17 +0,0 @@
-{
-  "schemaVersion": 1,
-  "policyId": "baseline-v1",
-  "description": "Current DAG context assembly: shared upstream char budget, role skill defaults, learned patterns only for implementer.",
-  "knobs": {
-    "maxUpstreamCharsByRole": {
-      "default": 2000
-    },
-    "includeLearnedPatternsRoles": ["implementer"],
-    "perSkillMaxCharsByRole": {
-      "default": 3000
-    },
-    "totalMaxCharsByRole": {
-      "default": 12000
-    }
-  }
-}

package/docs/templates/evaluation/context-policy-role-specialized-v1.json

@@ -1,28 +0,0 @@
-{
-  "schemaVersion": 1,
-  "policyId": "role-specialized-v1",
-  "description": "Role-specialized upstream and skill-instruction budgets; skill names still resolve via baseline merge order.",
-  "knobs": {
-    "maxUpstreamCharsByRole": {
-      "default": 2000,
-      "scout": 1200,
-      "reviewer": 1200,
-      "implementer": 3000,
-      "verifier": 1600,
-      "closeout": 1600,
-      "planner": 2000,
-      "supervisor": 2000
-    },
-    "includeLearnedPatternsRoles": ["implementer", "closeout"],
-    "perSkillMaxCharsByRole": {
-      "default": 3000,
-      "scout": 2500,
-      "implementer": 3500
-    },
-    "totalMaxCharsByRole": {
-      "default": 12000,
-      "scout": 10000,
-      "implementer": 14000
-    }
-  }
-}

package/docs/templates/evaluation/corpus-dogfood-v0.manifest.json

@@ -1,118 +0,0 @@
-{
-  "schemaVersion": 1,
-  "corpusId": "corpus-dogfood-v0",
-  "createdAt": "2026-07-19T00:00:00.000Z",
-  "description": "W1.5 initial dogfood corpus contract: 15 heterogeneous tasks with public/private/held_out splits. Contract-only; not live campaign evidence.",
-  "tasks": [
-    {
-      "taskRef": "features/F-2026-001/login-flow",
-      "class": "fullstack",
-      "split": "public",
-      "seeds": [1, 2, 3],
-      "featureId": "F-2026-001",
-      "description": "Dogfood login Feature packet (public feedback allowed)."
-    },
-    {
-      "taskRef": "features/F-2026-002/profile-filter",
-      "class": "implement",
-      "split": "public",
-      "seeds": [1, 2],
-      "featureId": "F-2026-002",
-      "description": "Profile filter implementation path."
-    },
-    {
-      "taskRef": "features/F-2026-005/welcome-banner",
-      "class": "fullstack",
-      "split": "public",
-      "seeds": [1, 2, 3],
-      "featureId": "F-2026-005",
-      "description": "fullstack-v1 welcome sample (public)."
-    },
-    {
-      "taskRef": "synthetic/docs-agents-converge",
-      "class": "docs",
-      "split": "public",
-      "seeds": [1, 2],
-      "description": "Docs/AGENTS converge style task (synthetic contract slot)."
-    },
-    {
-      "taskRef": "synthetic/docs-changelog-summary",
-      "class": "docs",
-      "split": "public",
-      "seeds": [1, 2, 3],
-      "description": "Changelog-style documentation task."
-    },
-    {
-      "taskRef": "synthetic/frontend-welcome-render",
-      "class": "frontend",
-      "split": "public",
-      "seeds": [1, 2],
-      "description": "Frontend render/accessibility contract slot."
-    },
-    {
-      "taskRef": "features/F-2026-003/empty-profile-regression",
-      "class": "bugfix",
-      "split": "private",
-      "seeds": [1, 2, 3],
-      "featureId": "F-2026-003",
-      "description": "Empty profile id/displayName regression (private selection)."
-    },
-    {
-      "taskRef": "features/F-2026-004/versioned-bootstrap",
-      "class": "bootstrap",
-      "split": "private",
-      "seeds": [1, 2],
-      "featureId": "F-2026-004",
-      "description": "Versioned self-host bootstrap canary (private)."
-    },
-    {
-      "taskRef": "synthetic/backend-test-welcome-api",
-      "class": "backend-test",
-      "split": "private",
-      "seeds": [1, 2, 3],
-      "description": "Backend API test workflow contract slot."
-    },
-    {
-      "taskRef": "synthetic/refactor-outcome-tokens",
-      "class": "refactor",
-      "split": "private",
-      "seeds": [1, 2],
-      "description": "Bounded refactor around evidence token naming."
-    },
-    {
-      "taskRef": "synthetic/implement-ready-planner-gate",
-      "class": "implement",
-      "split": "private",
-      "seeds": [1, 2, 3],
-      "description": "Ready Planner artifact-gate implementation slot."
-    },
-    {
-      "taskRef": "synthetic/heldout-docs-handoff",
-      "class": "docs",
-      "split": "held_out",
-      "seeds": [1, 2],
-      "description": "Held-out docs/handoff generalization slot; never used for candidate generation."
-    },
-    {
-      "taskRef": "synthetic/heldout-bugfix-gate",
-      "class": "bugfix",
-      "split": "held_out",
-      "seeds": [1, 2, 3],
-      "description": "Held-out bugfix gate; promotion-only signal."
-    },
-    {
-      "taskRef": "synthetic/heldout-fullstack-smoke",
-      "class": "fullstack",
-      "split": "held_out",
-      "seeds": [1, 2],
-      "description": "Held-out fullstack smoke; not for public feedback."
-    },
-    {
-      "taskRef": "synthetic/heldout-bootstrap-slot",
-      "class": "bootstrap",
-      "split": "held_out",
-      "seeds": [1, 2, 3],
-      "description": "Held-out bootstrap identity/slot discipline."
-    }
-  ]
-}

package/docs/templates/evaluation/matrix-dag-dry-run-v1.json

@@ -1,21 +0,0 @@
-{
-  "version": 3,
-  "title": "eval-matrix-cell-dry-run",
-  "runtimeContract": {
-    "schemaVersion": 1,
-    "agentRuntime": "pi-only",
-    "repairWriterProtocol": "explicit-node-v1"
-  },
-  "tasks": [
-    {
-      "id": "matrix-cell-static",
-      "depends_on": [],
-      "complexity": "LOW",
-      "executor": "static",
-      "subtask_prompt": "Matrix cell placeholder; evaluation binding is injected by campaign matrix.",
-      "static": {
-        "resultMarkdown": "matrix cell dry-run placeholder"
-      }
-    }
-  ]
-}

package/docs/templates/evaluation/matrix-fixture-v1.json

@@ -1,10 +0,0 @@
-{
-  "schemaVersion": 1,
-  "defaultVerifyPassed": false,
-  "results": [
-    {
-      "split": "public",
-      "verifyPassed": true
-    }
-  ]
-}

package/docs/templates/evaluation/private-verifier-dogfood-v0.json

@@ -1,16 +0,0 @@
-{
-  "schemaVersion": 1,
-  "privateVerifierId": "pv-dogfood-v0",
-  "createdAt": "2026-07-19T00:00:00.000Z",
-  "description": "M4 dogfood private verifier. Controller-only; never enter candidate contentRefs/prompts/skill snapshots.",
-  "checks": [
-    {
-      "checkId": "node-version-ok",
-      "kind": "shell",
-      "command": ["node", "-e", "process.exit(process.versions.node ? 0 : 1)"],
-      "cwd": ".",
-      "timeoutMs": 15000,
-      "expectedExitCode": 0
-    }
-  ]
-}

package/docs/templates/product-line/feature.yaml

@@ -1,11 +0,0 @@
-schema_version: 1
-feature_id: F-YYYY-NNN
-# generic = legacy/simple packet (default when this file is absent)
-# fullstack-v1 = enable structural gates for explicit workflows + dual-coverage ACs
-profile: generic
-# Uncomment and set true flags when profile is fullstack-v1:
-# scope:
-#   backend: true
-#   frontend: true
-#   backendVerification: true
-#   frontendVerification: true