Files
PaperClipAI/server/src/__tests__/built-in-agents.test.ts
T
DottaandPaperclip a1ab55a56d fix(agents): grant configuration access by default (#15283)
## Thinking Path

> - Paperclip manages agent work and permissions for a company.
> - Agent setup can require one agent to configure another agent.
> - New standard agents could create agents, but they had no direct
configuration grant.
> - Existing agents must keep their current permissions after an
upgrade.
> - The requested new-agent defaults include 13 more direct permissions
for suggestions, skills, tools, audit, inbox, and task assignment.
> - This pull request adds the 14-grant set at creation and approval
activation, retains it through invitations, and keeps existing agents
unchanged.
> - The change keeps low trust and built-in agents on their narrower
permissions.

## Linked Issues or Issue Description

**What existing behavior does this improve?**

The agent creation and approval flows, and the agent configuration
authorization path.

**Current behavior**

A new standard agent can create an agent. It cannot make protected
changes to a peer agent unless an operator adds an `agents:configure`
grant.

**Proposed behavior**

Add direct grants for `agents:configure`, `agents:suggest-changes`,
`skills:create`, `skills:suggest-changes`, `tools:manage_connections`,
`tools:manage_profiles`, `tools:view_audit`, `audit:view_agent_actions`,
`tools:use`, `tools:manage_runtime`, `inbox:manage`, `tasks:assign`,
`tasks:assign_scope`, and `tasks:manage_active_checkouts` to new
standard agents. Scope `tasks:assign_scope` to the agent’s reporting
subtree. Keep existing agents and their grants unchanged.

**Reason and benefit**

New standard agents can complete agent setup and the requested tool,
skill, inbox, and task workflows under existing route, scope, and
approval checks.

**Breaking changes**

Existing agents keep their current permissions. There is no permission
migration.

Related PR: #12212 adds scoped grant routes. This PR changes the default
grant.

## What Changed

- Add the 14 requested direct grants in agent creation and approval
activation. Keep them when invitation approval replaces grants,
preserving explicit scopes. Remove them when an agent is deleted.
- Apply defaults only to new agents. Remove the existing-agent
permission migration, its snapshot, and its journal entry. Preserve
existing scoped grants.
- Add permission, scope, invitation, and existing-agent regression
tests. Document all default and excluded permissions.
- Prevent agent keys from creating, changing, or restoring host-executed
process or local adapter command settings, and from restoring workspace
commands through rollback.

## Verification

- Run `node_modules/.bin/vitest run
server/src/__tests__/agent-default-configure-grants.test.ts
server/src/__tests__/invite-join-grants.test.ts
packages/db/src/migration-snapshot-drift.test.ts`. All 15 tests pass.
These cover new-agent defaults, unchanged existing grants, pending
approval, invitation grants, and migration history.
- Run `node_modules/.bin/tsc -p server/tsconfig.json --noEmit`. It
passes.
- Run `packages/db/node_modules/.bin/tsx
packages/db/src/check-migration-numbering.ts` and
`packages/db/node_modules/.bin/tsx
packages/db/src/check-migration-safety.ts`. Both pass.
- Confirm that this PR has no files under `packages/db/src/migrations/`
in its final diff.
- GitHub CI at `e8173b2c41` passes build, typecheck, server suites,
browser shards, and canary verification. One unchanged OpenCode
transport test reached its five-second timeout in the first Runner shard
run. That test passes locally in 2.55 seconds. The shard passed on one
retry. All 54 checks pass, with two expected skips.
- Greptile gives this exact head 5/5. Security review passes. The PR has
no unresolved review threads or merge conflicts.

## Risks

- The 14 default permissions apply only to new standard agents. Existing
permissions stay unchanged. Low trust and managed built-in agents are
excluded.
- New grants include connection, runtime, and active-checkout
management. Existing company, responsible-user, scope, and approval
checks remain in force. The default inbox grant carries a
responsible-user-only scope so it cannot override another user's inbox
settings.
- Protected changes still need the responsible user's authority when
that check applies. Company boundaries and approval gates still apply.
- Agent-authenticated requests cannot configure process adapters or host
command settings on local adapters. Known provider credential references
remain allowed, as do narrow plain authentication overrides on new peers
when the caller uses an AI connection pool. Arbitrary environment
settings remain blocked. Board operators retain the host configuration
paths.

> I checked `ROADMAP.md`. This is a narrow fix to existing agent
configuration behavior.

## Model Used

- OpenAI Codex, GPT-6 series. This runtime did not expose its exact
hosted model ID or context window. This revision uses OpenAI GPT-6
through Codex with tool use, code execution, and tests. The runtime does
not expose the exact hosted model ID or context window.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-10-06 05:28:48 -05:00

1736 lines
69 KiB
TypeScript

import fs from "node:fs/promises";
import { fileHash } from "../services/agent-file-store.js";
import { stockHash } from "../services/managed-resource-drift.js";
import { randomUUID } from "node:crypto";
import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import path from "node:path";
import { afterAll, afterEach, beforeAll, describe, expect, it, vi } from "vitest";
import { and, eq, sql } from "drizzle-orm";
import {
activityLog,
agentConfigRevisions,
agents,
approvals,
budgetPolicies,
builtInManagedResources,
companies,
companyMemberships,
companySkillVersions,
companySkills,
createDb,
issueThreadInteractions,
issues,
principalPermissionGrants,
routines,
routineTriggers,
} from "@paperclipai/db";
import { readPaperclipSkillSyncPreference } from "@paperclipai/adapter-utils/server-utils";
import {
getEmbeddedPostgresTestSupport,
startEmbeddedPostgresTestDatabase,
} from "./helpers/embedded-postgres.js";
import { agentInstructionRevisionService } from "../services/agent-instruction-revisions.js";
import { agentInstructionsService } from "../services/agent-instructions.ts";
import { agentService } from "../services/agents.ts";
import { approvalService } from "../services/approvals.ts";
import {
builtInAgentService,
deriveBuiltInAgentStatus,
listBuiltInAgentDefinitions,
readBuiltInTextWithFallback,
reconcileBuiltInAgentsOnStartup,
validateBuiltInAgentDefinitions,
} from "../services/built-in-agents.ts";
import { readBuiltInAgentMarker, withBuiltInAgentMarker } from "../services/built-in-agent-metadata.ts";
import { issueThreadInteractionService } from "../services/issue-thread-interactions.ts";
const embeddedPostgresSupport = await getEmbeddedPostgresTestSupport();
const describeEmbeddedPostgres = embeddedPostgresSupport.supported ? describe : describe.skip;
const BUILT_IN_MARKER_UNIQUE_INDEX = "agents_company_built_in_agent_key_unique_idx";
// Mirrors migration 0192. Used to drop/restore the partial unique index when a
// test needs to simulate legacy duplicates that predate the constraint.
const BUILT_IN_MARKER_UNIQUE_INDEX_DDL = `
CREATE UNIQUE INDEX IF NOT EXISTS "${BUILT_IN_MARKER_UNIQUE_INDEX}"
ON "agents" ("company_id", ((metadata -> 'paperclipBuiltInAgent' ->> 'key')))
WHERE (metadata -> 'paperclipBuiltInAgent' ->> 'key') IS NOT NULL
AND status <> 'terminated'
`;
function issuePrefix(id: string) {
return `T${id.replace(/-/g, "").slice(0, 6).toUpperCase()}`;
}
if (!embeddedPostgresSupport.supported) {
console.warn(
`Skipping embedded Postgres built-in agent tests on this host: ${embeddedPostgresSupport.reason ?? "unsupported environment"}`,
);
}
describe("built-in agent asset loading", () => {
it("uses the first readable candidate path", () => {
const dir = mkdtempSync(path.join(tmpdir(), "paperclip-built-in-agent-"));
try {
const first = path.join(dir, "missing.md");
const second = path.join(dir, "asset.md");
writeFileSync(second, "asset text", "utf8");
expect(readBuiltInTextWithFallback(`asset:${randomUUID()}`, [first, second], "fallback text")).toBe("asset text");
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
it("falls back instead of throwing when built-in agent files are missing", () => {
const warn = vi.spyOn(console, "warn").mockImplementation(() => {});
const label = `missing:${randomUUID()}`;
try {
expect(readBuiltInTextWithFallback(label, [path.join(tmpdir(), label, "AGENTS.md")], "fallback text")).toBe(
"fallback text",
);
expect(warn).toHaveBeenCalledWith(expect.stringContaining(`Built-in agent asset ${label} was not readable`));
} finally {
warn.mockRestore();
}
});
it("warns about non-missing read errors before falling back", () => {
const warn = vi.spyOn(console, "warn").mockImplementation(() => {});
const dir = mkdtempSync(path.join(tmpdir(), "paperclip-built-in-agent-"));
const label = "unreadable:" + randomUUID();
try {
const directoryPath = path.join(dir, "asset.md");
mkdirSync(directoryPath);
expect(readBuiltInTextWithFallback(label, [directoryPath], "fallback text")).toBe("fallback text");
expect(warn).toHaveBeenCalledWith(expect.stringContaining("read error on " + directoryPath + ":"));
expect(warn).toHaveBeenCalledWith(
expect.stringContaining("Built-in agent asset " + label + " was not readable"),
);
} finally {
warn.mockRestore();
rmSync(dir, { recursive: true, force: true });
}
});
});
describeEmbeddedPostgres("built-in agents", () => {
let db!: ReturnType<typeof createDb>;
let tempDb: Awaited<ReturnType<typeof startEmbeddedPostgresTestDatabase>> | null = null;
const instructionOperator = { type: "board", userId: "local-board", source: "local_implicit" } as const;
async function writeInstructionEntry(agent: { id: string; companyId: string }, entryFile: string, content: string) {
const service = agentInstructionRevisionService(db);
const target = { companyId: agent.companyId, agentId: agent.id };
const baseline = await service.readCurrent(target, instructionOperator);
await service.commit({ ...target, entryFile, content, baseRevisionId: baseline?.revision.id ?? null, source: "board" }, instructionOperator);
const refreshed = await agentService(db).getById(agent.id);
return { adapterConfig: refreshed!.adapterConfig };
}
beforeAll(async () => {
tempDb = await startEmbeddedPostgresTestDatabase("paperclip-built-in-agents-");
db = createDb(tempDb.connectionString);
}, 20_000);
afterEach(async () => {
await db.delete(routineTriggers);
await db.delete(routines);
await db.delete(issueThreadInteractions);
await db.delete(issues);
await db.delete(builtInManagedResources);
await db.delete(companySkillVersions);
await db.delete(companySkills);
await db.delete(principalPermissionGrants);
await db.delete(companyMemberships);
await db.delete(agentConfigRevisions);
await db.delete(activityLog);
await db.delete(approvals);
await db.delete(agents);
await db.delete(budgetPolicies);
await db.delete(companies);
// Some tests drop the built-in marker unique index to simulate legacy
// (pre-migration 0192) duplicates; restore it now that all rows are gone.
await db.execute(sql.raw(BUILT_IN_MARKER_UNIQUE_INDEX_DDL));
});
afterAll(async () => {
await tempDb?.cleanup();
});
async function permissionKeysForAgent(agentId: string) {
const grants = await db
.select()
.from(principalPermissionGrants)
.where(eq(principalPermissionGrants.principalId, agentId));
return grants.map((grant) => grant.permissionKey).sort();
}
async function seedCompany(options: { requireApproval?: boolean } = {}) {
const companyId = randomUUID();
await db.insert(companies).values({
id: companyId,
name: "Paperclip",
issuePrefix: issuePrefix(companyId),
defaultResponsibleUserId: "responsible-user",
requireBoardApprovalForNewAgents: options.requireApproval ?? true,
});
return companyId;
}
it("validates the static registry and rejects invalid definitions", () => {
const definitions = listBuiltInAgentDefinitions();
expect(definitions.map((definition) => definition.key).sort()).toEqual(["briefs", "learning", "reflection-coach", "summarizer"]);
const summarizer = definitions.find((definition) => definition.key === "summarizer");
expect(summarizer).toMatchObject({
defaultAdapterType: "claude_local",
defaultAdapterConfig: { model: "claude-haiku-4-5" },
});
expect(summarizer?.defaultRuntimeConfig).toBeUndefined();
expect(() => validateBuiltInAgentDefinitions([
{
key: "briefs",
displayName: "Briefs Agent",
featureKeys: ["briefs"],
shortPurpose: "One",
defaultInstructions: "Do work",
defaultRole: "general",
},
{
key: "briefs",
displayName: "Duplicate",
featureKeys: ["duplicate"],
shortPurpose: "Two",
defaultInstructions: "Do work",
defaultRole: "general",
},
])).toThrow("Duplicate built-in agent key");
expect(() => validateBuiltInAgentDefinitions([
{
key: "Bad Key",
displayName: "Bad",
featureKeys: ["bad"],
shortPurpose: "Bad",
defaultInstructions: "Bad",
defaultRole: "general",
},
])).toThrow("Invalid built-in agent key");
expect(() => validateBuiltInAgentDefinitions([
{
key: "bad-default",
displayName: "Bad default",
featureKeys: ["bad-default"],
shortPurpose: "Bad default adapter",
defaultInstructions: "Do work",
defaultRole: "general",
allowedAdapterTypes: ["codex_local"],
defaultAdapterType: "claude_local",
},
])).toThrow("defaultAdapterType must be allowed");
});
it("lazily provisions one agent per company/key and updates the same row on setup", async () => {
const companyId = await seedCompany();
const svc = builtInAgentService(db);
const created = await svc.ensure(companyId, "briefs");
expect(created.status).toBe("needs_setup");
expect(created.agentId).toBeTruthy();
expect(created.agent).toMatchObject({
companyId,
name: "Briefs Agent",
adapterConfig: {},
status: "idle",
});
expect(readBuiltInAgentMarker(created.agent?.metadata)).toEqual({
key: "briefs",
featureKeys: ["briefs"],
});
const configured = await svc.ensure(companyId, "briefs", {
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
});
expect(configured.status).toBe("ready");
expect(configured.agentId).toBe(created.agentId);
expect(configured.agent).toMatchObject({
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
});
const reconciled = await svc.ensure(companyId, "briefs");
expect(reconciled.status).toBe("ready");
expect(reconciled.agent).toMatchObject({
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
});
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(rows).toHaveLength(1);
});
it("routes policy-gated built-in provisioning through a pending hire approval", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
const result = await builtIns.provision(companyId, "briefs", {
adapterType: "process",
adapterConfig: { command: "echo safe" },
budgetMonthlyCents: 5000,
}, { requestedByUserId: "board-user" });
expect(result.state).toMatchObject({
status: "pending_approval",
agent: {
companyId,
name: "Briefs Agent",
status: "pending_approval",
adapterType: "process",
adapterConfig: { command: "echo safe" },
budgetMonthlyCents: 5000,
},
});
expect(result.approval).toMatchObject({
companyId,
type: "hire_agent",
status: "pending",
requestedByUserId: "board-user",
requestedByAgentId: null,
payload: {
name: "Briefs Agent",
role: "general",
adapterType: "process",
adapterConfig: { command: "echo safe" },
budgetMonthlyCents: 5000,
agentId: result.state.agentId,
sourceBuiltInAgentKey: "briefs",
featureKeys: ["briefs"],
},
});
const rowsBeforeApproval = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(rowsBeforeApproval).toHaveLength(1);
expect(rowsBeforeApproval[0]).toMatchObject({ status: "pending_approval" });
await expect(builtIns.requireBuiltInAgent(companyId, "briefs")).rejects.toMatchObject({
status: 412,
details: { code: "built_in_agent_not_configured", status: "pending_approval" },
});
await expect(agentService(db).update(result.state.agentId!, {
adapterType: "process",
adapterConfig: { command: "echo tampered" },
})).rejects.toMatchObject({
status: 409,
details: {
code: "pending_approval_agent_config_frozen",
agentId: result.state.agentId,
fields: ["adapterConfig"],
},
});
await expect(builtIns.provision(companyId, "briefs", {
budgetMonthlyCents: 7500,
})).rejects.toMatchObject({
status: 409,
details: {
code: "built_in_agent_pending_approval",
key: "briefs",
agentId: result.state.agentId,
},
});
await db
.update(agents)
.set({
adapterType: "process",
adapterConfig: { command: "echo tampered" },
})
.where(eq(agents.id, result.state.agentId!));
await approvalService(db).approve(result.approval!.id, "board-user", "Approved built-in agent");
await expect(builtIns.get(companyId, "briefs")).resolves.toMatchObject({
status: "ready",
agentId: result.state.agentId,
agent: { status: "idle", adapterType: "process", adapterConfig: { command: "echo safe" }, budgetMonthlyCents: 5000 },
});
});
it("blocks policy-gated built-in reconfiguration instead of applying adapter overrides immediately", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
const ready = await builtIns.ensure(companyId, "briefs", {
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
});
await expect(builtIns.provision(companyId, "briefs", {
adapterType: "process",
adapterConfig: { command: "echo bypass" },
})).rejects.toMatchObject({
status: 409,
details: {
code: "built_in_agent_reconfiguration_requires_approval",
key: "briefs",
agentId: ready.agentId,
},
});
await expect(builtIns.get(companyId, "briefs")).resolves.toMatchObject({
status: "ready",
agentId: ready.agentId,
agent: { adapterType: "codex_local", adapterConfig: { model: "gpt-5.4" } },
});
});
it("completes first-time setup of a needs_setup built-in without a fresh board approval", async () => {
const companyId = await seedCompany({ requireApproval: true });
const builtIns = builtInAgentService(db);
// A hired-but-unconfigured built-in row: exists (its hire was already
// sanctioned) but its adapter config is still empty → `needs_setup`.
const seeded = await builtIns.ensure(companyId, "briefs");
expect(seeded.status).toBe("needs_setup");
// Configuring the adapter for the first time must apply directly instead of
// throwing "adapter changes require board approval".
const result = await builtIns.provision(companyId, "briefs", {
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
budgetMonthlyCents: 2500,
}, { requestedByUserId: "board-user" });
expect(result.approval).toBeNull();
expect(result.state).toMatchObject({
status: "ready",
agentId: seeded.agentId,
agent: {
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
budgetMonthlyCents: 2500,
},
});
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(rows).toHaveLength(1);
const noApprovals = await db.select().from(approvals).where(eq(approvals.companyId, companyId));
expect(noApprovals).toHaveLength(0);
});
it("rejects adapter types outside the built-in definition allowlist", async () => {
const companyId = await seedCompany();
await expect(builtInAgentService(db).ensure(companyId, "briefs", {
adapterType: "http",
adapterConfig: { url: "https://example.test/webhook" },
})).rejects.toMatchObject({
status: 422,
details: {
code: "built_in_agent_adapter_not_allowed",
key: "briefs",
allowedAdapterTypes: ["codex_local", "claude_local", "gemini_local", "opencode_local", "process"],
},
});
});
it("rejects unknown built-in adapter models before saving setup", async () => {
const companyId = await seedCompany();
await expect(builtInAgentService(db).ensure(companyId, "summarizer", {
adapterType: "claude_local",
adapterConfig: { model: "claude-haiku-4-6" },
})).rejects.toMatchObject({
status: 422,
details: {
code: "built_in_agent_model_unknown",
key: "summarizer",
adapterType: "claude_local",
model: "claude-haiku-4-6",
},
});
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(rows).toHaveLength(0);
});
it("recovers an orphaned marked row instead of creating a duplicate", async () => {
const companyId = await seedCompany();
const orphanId = randomUUID();
await db.insert(agents).values({
id: orphanId,
companyId,
name: "Old Briefs",
role: "general",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
metadata: withBuiltInAgentMarker({ source: "orphan" }, { key: "briefs", featureKeys: ["briefs"] }),
});
const state = await builtInAgentService(db).ensure(companyId, "briefs");
expect(state.status).toBe("ready");
expect(state.agentId).toBe(orphanId);
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(rows).toHaveLength(1);
});
it("derives not_provisioned, needs_setup, ready, and paused states", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
await expect(builtIns.get(companyId, "learning")).resolves.toMatchObject({ status: "not_provisioned" });
const needsSetup = await builtIns.ensure(companyId, "learning");
expect(needsSetup.status).toBe("needs_setup");
expect(deriveBuiltInAgentStatus(needsSetup.agent)).toBe("needs_setup");
const ready = await builtIns.ensure(companyId, "learning", {
adapterType: "claude_local",
adapterConfig: { model: "claude-sonnet-4-5" },
});
expect(ready.status).toBe("ready");
await agentService(db).pause(ready.agentId!, "manual");
await expect(builtIns.get(companyId, "learning")).resolves.toMatchObject({
status: "paused",
agentId: ready.agentId,
pauseReason: "manual",
});
});
it("requires configured built-ins with typed precondition failures and paused warnings", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
await expect(builtIns.requireBuiltInAgent(companyId, "briefs")).rejects.toMatchObject({
status: 412,
details: {
code: "built_in_agent_not_configured",
key: "briefs",
status: "not_provisioned",
agentId: null,
},
});
const needsSetup = await builtIns.ensure(companyId, "briefs");
await expect(builtIns.requireBuiltInAgent(companyId, "briefs")).rejects.toMatchObject({
status: 412,
details: {
code: "built_in_agent_not_configured",
key: "briefs",
status: "needs_setup",
agentId: needsSetup.agentId,
},
});
const ready = await builtIns.ensure(companyId, "briefs", {
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
});
await expect(builtIns.requireBuiltInAgent(companyId, "briefs")).resolves.toMatchObject({
agent: { id: ready.agentId },
warning: null,
});
await agentService(db).pause(ready.agentId!, "maintenance");
await expect(builtIns.requireBuiltInAgent(companyId, "briefs")).resolves.toMatchObject({
agent: { id: ready.agentId },
warning: {
code: "built_in_agent_paused",
key: "briefs",
agentId: ready.agentId,
pauseReason: "maintenance",
},
});
});
it("resets marked agents back to registry display defaults without replacing adapter setup", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
const ready = await builtIns.ensure(companyId, "briefs", {
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
});
await agentService(db).update(ready.agentId!, {
name: "Custom Briefs",
role: "engineer",
title: "Custom",
capabilities: "Custom purpose",
});
const reset = await builtIns.reset(companyId, "briefs", {}, instructionOperator);
expect(reset).toMatchObject({
status: "ready",
agentId: ready.agentId,
agent: {
name: "Briefs Agent",
role: "general",
title: null,
capabilities: "Prepares concise operational briefs for the board and agent company.",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
},
});
});
it("reconciles an enabled Reflection Coach bundle with skill sync and a disabled routine", async () => {
const companyId = await seedCompany({ requireApproval: false });
const root = await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4", apiKey: "do-not-copy" },
runtimeConfig: {},
permissions: {},
});
// The Reflection Coach is opt-in (not auto-created). Enabling it on demand
// materializes its managed bundle in a single pass.
const enabled = await builtInAgentService(db).ensure(companyId, "reflection-coach");
expect(enabled.agent?.adapterConfig).toMatchObject({
instructionsBundleMode: "managed",
instructionsEntryFile: "AGENTS.md",
});
expect(enabled.agent?.adapterConfig).not.toMatchObject({ model: "gpt-5.4", apiKey: "do-not-copy" });
// Startup reconcile keeps the enabled bundle tracking stock and re-grants
// the root/company default permissions.
const result = await reconcileBuiltInAgentsOnStartup(db);
expect(result.autoEnsured).toBeGreaterThanOrEqual(1);
expect(result.defaultGrantsEnsured).toBeGreaterThanOrEqual(4);
const rootGrantKeys = await permissionKeysForAgent(root.id);
expect(rootGrantKeys).toEqual(expect.arrayContaining([
"agents:configure", "agents:suggest-changes", "skills:create", "skills:suggest-changes",
]));
const state = await builtInAgentService(db).get(companyId, "reflection-coach");
expect(state).toMatchObject({
status: "paused",
agent: {
companyId,
name: "Reflection Coach",
role: "general",
title: "Reflection Coach",
icon: "eye",
adapterType: "codex_local",
permissions: {
canCreateAgents: false,
canCreateSkills: false,
},
},
});
expect(state.resources.map((resource) => [resource.resourceKind, resource.stockStatus])).toEqual([
["instructions", "stock_current"],
["skill", "stock_current"],
["routine", "stock_current"],
]);
expect(state.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
resourceId: expect.any(String),
scheduleEnabled: false,
});
const agentRows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(agentRows.filter((row) => readBuiltInAgentMarker(row.metadata)?.key === "reflection-coach")).toHaveLength(1);
const [skill] = await db
.select()
.from(companySkills)
.where(eq(companySkills.key, "paperclipai/bundled/paperclip-operations/reflection-coach"));
expect(skill).toMatchObject({
key: "paperclipai/bundled/paperclip-operations/reflection-coach",
slug: "reflection-coach",
});
expect(readPaperclipSkillSyncPreference(state.agent!.adapterConfig as Record<string, unknown>).desiredSkills).toContain(
"paperclipai/bundled/paperclip-operations/reflection-coach",
);
const [routine] = await db.select().from(routines).where(eq(routines.companyId, companyId));
expect(routine).toMatchObject({
title: "Review recent agent trajectories for coaching proposals",
status: "paused",
assigneeAgentId: state.agentId,
});
const [trigger] = await db.select().from(routineTriggers).where(eq(routineTriggers.routineId, routine!.id));
expect(trigger).toMatchObject({
kind: "schedule",
enabled: false,
cronExpression: "0 9 * * 1",
timezone: "UTC",
});
const coachGrantKeys = await permissionKeysForAgent(state.agentId!);
expect(coachGrantKeys).toEqual(expect.arrayContaining(["agents:suggest-changes", "skills:suggest-changes"]));
expect(coachGrantKeys).not.toContain("agents:configure");
expect(coachGrantKeys).not.toContain("skills:create");
});
it("recreates missing managed resource bindings idempotently during concurrent reconcile", async () => {
const companyId = await seedCompany({ requireApproval: false });
await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const builtIns = builtInAgentService(db);
await builtIns.ensure(companyId, "reflection-coach");
await db.delete(builtInManagedResources).where(eq(builtInManagedResources.companyId, companyId));
const states = await Promise.all([
builtIns.ensure(companyId, "reflection-coach"),
builtIns.ensure(companyId, "reflection-coach"),
]);
expect(states).toHaveLength(2);
for (const state of states) {
expect(state.resources.map((resource) => [resource.resourceKind, resource.stockStatus])).toEqual([
["instructions", "stock_current"],
["skill", "stock_current"],
["routine", "stock_current"],
]);
}
const bindings = await db
.select()
.from(builtInManagedResources)
.where(eq(builtInManagedResources.companyId, companyId));
expect(bindings).toHaveLength(3);
expect(new Set(bindings.map((binding) =>
`${binding.bundleKey}:${binding.resourceKind}:${binding.resourceKey}`
)).size).toBe(3);
});
it("preserves new-agent approval gates during on-demand Reflection Coach provisioning", async () => {
const companyId = await seedCompany({ requireApproval: true });
const root = await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const mutationPolicy = {
requiresDisplayedDiff: true,
requiresAcceptedTaskInteraction: true,
applyInSeparateFollowUpRun: true,
};
// The Reflection Coach is opt-in, so it's enabled on demand. With board
// approval required, provisioning it must leave a pending agent + a
// hire_agent approval rather than an active agent.
const provisioned = await builtInAgentService(db).provision(companyId, "reflection-coach");
expect(provisioned.approval).not.toBeNull();
const state = await builtInAgentService(db).get(companyId, "reflection-coach");
expect(state).toMatchObject({
status: "pending_approval",
agent: {
companyId,
name: "Reflection Coach",
status: "pending_approval",
reportsTo: root.id,
budgetMonthlyCents: 0,
permissions: {
builtInMutationPolicy: mutationPolicy,
},
},
});
expect(state.resources.map((resource) => resource.stockStatus)).toEqual(["missing", "missing", "missing"]);
const allApprovals = await db.select().from(approvals).where(eq(approvals.companyId, companyId));
const approval = allApprovals.find(
(row) => (row.payload as { agentId?: string } | null)?.agentId === state.agentId,
)!;
expect(approval).toMatchObject({
type: "hire_agent",
status: "pending",
payload: {
agentId: state.agentId,
sourceBuiltInAgentKey: "reflection-coach",
featureKeys: ["reflection-coach"],
reportsTo: root.id,
permissions: expect.objectContaining({
builtInMutationPolicy: mutationPolicy,
}),
},
});
const pendingReconcile = await reconcileBuiltInAgentsOnStartup(db);
expect(pendingReconcile.pendingApprovals).toBe(1);
const stillPending = await builtInAgentService(db).get(companyId, "reflection-coach");
expect(stillPending).toMatchObject({
status: "pending_approval",
agent: {
adapterConfig: {},
reportsTo: root.id,
status: "pending_approval",
},
});
expect(stillPending.resources.map((resource) => resource.stockStatus)).toEqual([
"missing",
"missing",
"missing",
]);
await approvalService(db).approve(approval.id, "board-user", "Approved Reflection Coach");
const approvedState = await builtInAgentService(db).get(companyId, "reflection-coach");
expect(approvedState).toMatchObject({
agent: {
reportsTo: root.id,
permissions: {
builtInMutationPolicy: mutationPolicy,
},
},
});
expect(approvedState.resources.map((resource) => resource.stockStatus)).toEqual([
"stock_current",
"stock_current",
"stock_current",
]);
await reconcileBuiltInAgentsOnStartup(db);
const agentRows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(agentRows.filter((row) => readBuiltInAgentMarker(row.metadata)?.key === "reflection-coach")).toHaveLength(1);
const approvalRows = await db.select().from(approvals).where(eq(approvals.companyId, companyId));
expect(approvalRows).toHaveLength(1);
});
it("automatically upgrades untouched stock instructions while preserving personal files", async () => {
const companyId = await seedCompany({ requireApproval: false });
const svc = builtInAgentService(db);
const created = await svc.ensure(companyId, "reflection-coach");
const agent = created.agent!;
const oldFiles = { "AGENTS.md": "# Previous stock instructions\n", "obsolete.txt": "old stock support" };
await writeInstructionEntry(agent, "AGENTS.md", oldFiles["AGENTS.md"]);
await agentInstructionsService(db).writeFile(agent, "obsolete.txt", oldFiles["obsolete.txt"]);
await agentInstructionsService(db).writeFile(agent, "personal.txt", "keep my notes");
await db.update(builtInManagedResources).set({ stockHash: stockHash(oldFiles), stockVersion: "previous",
defaultsJson: { entryFile: "AGENTS.md", files: Object.keys(oldFiles) },
}).where(and(eq(builtInManagedResources.companyId, companyId), eq(builtInManagedResources.resourceKind, "instructions")));
const updated = await svc.ensure(companyId, "reflection-coach");
expect(updated.resources.find(resource => resource.resourceKind === "instructions")).toMatchObject({ stockStatus: "stock_current" });
expect((await agentInstructionsService().readFile(updated.agent!, "AGENTS.md")).content).toContain("You are Reflection Coach");
expect((await agentInstructionsService().readFile(updated.agent!, "personal.txt")).content).toBe("keep my notes");
await expect(agentInstructionsService().readFile(updated.agent!, "obsolete.txt")).rejects.toMatchObject({ status: 404 });
const [event] = await db.select().from(activityLog).where(and(eq(activityLog.companyId, companyId), eq(activityLog.actorId, "built-in-reconcile")));
expect(event).toMatchObject({ actorType: "system", action: "agent.files_updated" });
});
it.each(["file", "database"])("retries a stock update after a %s failure without overwriting intervening edits", async (failure) => {
const companyId = await seedCompany({ requireApproval: false });
const svc = builtInAgentService(db);
const created = await svc.ensure(companyId, "reflection-coach");
const agent = created.agent!;
const oldFiles = { "AGENTS.md": "# Old stock\n", "obsolete.txt": "old supporting file" };
await writeInstructionEntry(agent, "AGENTS.md", oldFiles["AGENTS.md"]);
await agentInstructionsService(db).writeFile(agent, "obsolete.txt", oldFiles["obsolete.txt"]);
await db.update(builtInManagedResources).set({ stockHash: stockHash(oldFiles), stockVersion: "old",
defaultsJson: { entryFile: "AGENTS.md", files: Object.keys(oldFiles) },
}).where(and(eq(builtInManagedResources.companyId, companyId), eq(builtInManagedResources.resourceKind, "instructions")));
const trigger = `stock_failure_${companyId.replaceAll("-", "")}`;
const remove = fs.rm.bind(fs);
const fault = failure === "file" ? vi.spyOn(fs, "rm").mockImplementation(async (file, options) => {
if (String(file).endsWith("/obsolete.txt")) throw new Error("injected stock file failure");
return remove(file, options);
}) : null;
if (failure === "database") {
await db.execute(sql.raw(`CREATE FUNCTION ${trigger}() RETURNS trigger LANGUAGE plpgsql AS $$ BEGIN
IF NEW.company_id = '${companyId}' AND NEW.resource_kind = 'instructions' AND NEW.stock_hash <> OLD.stock_hash THEN
RAISE EXCEPTION 'injected stock binding failure'; END IF; RETURN NEW; END $$`));
await db.execute(sql.raw(`CREATE TRIGGER ${trigger} BEFORE UPDATE ON built_in_managed_resources FOR EACH ROW EXECUTE FUNCTION ${trigger}()`));
}
try { await expect(svc.ensure(companyId, "reflection-coach")).rejects.toThrow(); }
finally {
fault?.mockRestore();
if (failure === "database") {
await db.execute(sql.raw(`DROP TRIGGER ${trigger} ON built_in_managed_resources`));
await db.execute(sql.raw(`DROP FUNCTION ${trigger}()`));
}
}
const [pending] = await db.select().from(builtInManagedResources).where(and(eq(builtInManagedResources.companyId, companyId), eq(builtInManagedResources.resourceKind, "instructions")));
expect(pending.defaultsJson.pendingInstructionsUpdate).toBeTruthy();
expect(pending.stockHash).toBe(stockHash(oldFiles));
const partial = (await agentInstructionsService().readFile(agent, "AGENTS.md")).content;
expect(partial).toContain("You are Reflection Coach");
// A real operator edit after the failed attempt must stop the retry.
await writeInstructionEntry(agent, "AGENTS.md", "operator edit after failure");
await svc.ensure(companyId, "reflection-coach");
expect((await agentInstructionsService().readFile(agent, "AGENTS.md")).content).toBe("operator edit after failure");
await writeInstructionEntry(agent, "AGENTS.md", partial);
const retried = await svc.ensure(companyId, "reflection-coach");
expect(retried.resources.find(resource => resource.resourceKind === "instructions")).toMatchObject({ stockStatus: "stock_current" });
await expect(agentInstructionsService().readFile(agent, "obsolete.txt")).rejects.toMatchObject({ status: 404 });
const [complete] = await db.select().from(builtInManagedResources).where(eq(builtInManagedResources.id, pending.id));
expect(complete.defaultsJson.pendingInstructionsUpdate).toBeUndefined();
});
it.each([false, true])("supersedes an interrupted stock version, including rollback=%s", async (rollback) => {
const companyId = await seedCompany({ requireApproval: false });
const svc = builtInAgentService(db);
const created = await svc.ensure(companyId, "reflection-coach");
const agent = created.agent!;
const currentStock = (await agentInstructionsService().readFile(agent, "AGENTS.md")).content;
const oldFiles = { "AGENTS.md": rollback ? currentStock : "# Original stock" };
const interruptedFiles = { "AGENTS.md": "# Intermediate stock", "intermediate-only.txt": "intermediate support" };
await writeInstructionEntry(agent, "AGENTS.md", interruptedFiles["AGENTS.md"]);
await agentInstructionsService(db).writeFile(agent, "intermediate-only.txt", interruptedFiles["intermediate-only.txt"]);
await agentInstructionsService(db).writeFile(agent, "personal.txt", "personal notes");
await db.update(builtInManagedResources).set({ stockHash: stockHash(oldFiles), stockVersion: "old",
defaultsJson: { entryFile: "AGENTS.md", files: Object.keys(oldFiles), pendingInstructionsUpdate: {
stockHash: stockHash(interruptedFiles),
baseHashes: { "AGENTS.md": fileHash(Buffer.from(oldFiles["AGENTS.md"])), "intermediate-only.txt": null },
nextHashes: Object.fromEntries(Object.entries(interruptedFiles).map(([file, text]) => [file, fileHash(Buffer.from(text))])),
} },
}).where(and(eq(builtInManagedResources.companyId, companyId), eq(builtInManagedResources.resourceKind, "instructions")));
await writeInstructionEntry(agent, "AGENTS.md", "intervening operator edit");
await svc.ensure(companyId, "reflection-coach");
expect((await agentInstructionsService().readFile(agent, "AGENTS.md")).content).toBe("intervening operator edit");
await writeInstructionEntry(agent, "AGENTS.md", interruptedFiles["AGENTS.md"]);
const updated = await svc.ensure(companyId, "reflection-coach");
expect(updated.resources.find(resource => resource.resourceKind === "instructions")).toMatchObject({ stockStatus: "stock_current" });
expect((await agentInstructionsService().readFile(agent, "AGENTS.md")).content).toBe(currentStock);
await expect(agentInstructionsService().readFile(agent, "intermediate-only.txt")).rejects.toMatchObject({ status: 404 });
expect((await agentInstructionsService().readFile(agent, "personal.txt")).content).toBe("personal notes");
});
it("preserves Reflection Coach instruction drift on reconcile and restores it on reset", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
const created = await builtIns.ensure(companyId, "reflection-coach");
const instructions = agentInstructionsService();
await writeInstructionEntry(created.agent!, "AGENTS.md", "# Custom Reflection Coach\n\nOperator edit.\n");
const reconciled = await builtIns.ensure(companyId, "reflection-coach");
const drift = reconciled.resources.find((resource) => resource.resourceKind === "instructions");
expect(drift).toMatchObject({
stockStatus: "operator_modified",
updateAvailable: true,
resetAvailable: true,
changedFiles: ["AGENTS.md"],
});
await expect(instructions.readFile(reconciled.agent!, "AGENTS.md")).resolves.toMatchObject({
content: "# Custom Reflection Coach\n\nOperator edit.\n",
});
const reset = await builtIns.reset(companyId, "reflection-coach", {}, instructionOperator);
expect(reset.resources.find((resource) => resource.resourceKind === "instructions")).toMatchObject({
stockStatus: "stock_current",
resetAvailable: false,
});
const resetFile = await instructions.readFile(reset.agent!, "AGENTS.md");
expect(resetFile.content).toContain("Reflection Coach");
expect(resetFile.content).not.toContain("Operator edit.");
const revisions = agentInstructionRevisionService(db);
const target = { companyId, agentId: created.agent!.id, entryFile: "AGENTS.md" };
const history = await revisions.history(target, instructionOperator);
expect(history.revisions).toHaveLength(0);
const current = await revisions.readCurrent(target, instructionOperator);
expect(current?.content).toBe(resetFile.content);
});
it("blocks deleting a built-in agent", async () => {
const companyId = await seedCompany();
const state = await builtInAgentService(db).ensure(companyId, "briefs");
await expect(agentService(db).remove(state.agentId!)).rejects.toMatchObject({
status: 409,
details: {
code: "built_in_agent_undeletable",
key: "briefs",
},
});
});
it("prevents direct marker add, remove, or mutation", async () => {
const companyId = await seedCompany();
const builtIn = await builtInAgentService(db).ensure(companyId, "briefs");
const normal = await agentService(db).create(companyId, {
name: "Normal",
role: "engineer",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
await expect(agentService(db).create(companyId, {
name: "Spoof",
role: "engineer",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["briefs"] }),
})).rejects.toMatchObject({ status: 409, details: { code: "built_in_agent_marker_readonly" } });
await expect(agentService(db).update(normal.id, {
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["briefs"] }),
})).rejects.toMatchObject({ status: 409, details: { code: "built_in_agent_marker_readonly" } });
await expect(agentService(db).update(builtIn.agentId!, {
metadata: { other: "metadata" },
})).rejects.toMatchObject({ status: 409, details: { code: "built_in_agent_marker_readonly" } });
await expect(agentService(db).update(builtIn.agentId!, {
metadata: withBuiltInAgentMarker({}, { key: "learning", featureKeys: ["learning"] }),
})).rejects.toMatchObject({ status: 409, details: { code: "built_in_agent_marker_readonly" } });
await expect(agentService(db).update(builtIn.agentId!, {
metadata: withBuiltInAgentMarker({ note: "allowed" }, { key: "briefs", featureKeys: ["briefs"] }),
})).resolves.toMatchObject({
id: builtIn.agentId,
metadata: {
note: "allowed",
paperclipBuiltInAgent: { key: "briefs", featureKeys: ["briefs"] },
},
});
});
it("repairs display/default drift for marked rows during startup reconciliation", async () => {
const companyId = await seedCompany();
const agentId = randomUUID();
await db.insert(agents).values({
id: agentId,
companyId,
name: "Old Name",
role: "engineer",
title: "Old title",
capabilities: "Old purpose",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["old-briefs"] }),
});
const result = await reconcileBuiltInAgentsOnStartup(db);
expect(result).toMatchObject({ unknown: 0, duplicates: 0 });
expect(result.scanned).toBeGreaterThanOrEqual(1);
expect(result.reconciled).toBeGreaterThanOrEqual(1);
const [row] = await db.select().from(agents).where(eq(agents.id, agentId));
expect(row).toMatchObject({
name: "Briefs Agent",
role: "general",
title: null,
capabilities: "Prepares concise operational briefs for the board and agent company.",
});
expect(readBuiltInAgentMarker(row?.metadata)).toEqual({ key: "briefs", featureKeys: ["briefs"] });
});
// Reproduce a company that already carries duplicate marked rows from before
// migration 0192 by dropping the unique index, inserting two active rows, and
// pairing a pending hire_agent approval with the newer one.
async function seedLegacyDuplicateBriefs(companyId: string) {
await db.execute(sql.raw(`DROP INDEX IF EXISTS "${BUILT_IN_MARKER_UNIQUE_INDEX}"`));
const olderId = randomUUID();
const newerId = randomUUID();
await db.insert(agents).values([
{
id: olderId,
companyId,
name: "Briefs One",
role: "general",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
createdAt: new Date("2026-07-18T00:00:00.000Z"),
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["briefs"] }),
},
{
id: newerId,
companyId,
name: "Briefs Two",
role: "general",
status: "pending_approval",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
createdAt: new Date("2026-07-18T00:00:00.025Z"),
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["briefs"] }),
},
]);
const approval = await approvalService(db).create(companyId, {
type: "hire_agent",
requestedByAgentId: null,
requestedByUserId: null,
status: "pending",
payload: { agentId: newerId, sourceBuiltInAgentKey: "briefs", featureKeys: ["briefs"] },
decisionNote: null,
decidedByUserId: null,
decidedAt: null,
updatedAt: new Date(),
});
return { olderId, newerId, approvalId: approval!.id };
}
it("self-heals duplicate active instances, keeping the oldest and cancelling the newer's approval", async () => {
const companyId = await seedCompany();
const { olderId, newerId, approvalId } = await seedLegacyDuplicateBriefs(companyId);
// A plain read resolves the duplicate rather than throwing.
const state = await builtInAgentService(db).get(companyId, "briefs");
expect(state.agentId).toBe(olderId);
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
const byId = new Map(rows.map((row) => [row.id, row]));
expect(byId.get(olderId)?.status).toBe("idle");
expect(byId.get(newerId)?.status).toBe("terminated");
expect(
rows.filter((row) => row.status !== "terminated" && readBuiltInAgentMarker(row.metadata)?.key === "briefs"),
).toHaveLength(1);
const [approval] = await db.select().from(approvals).where(eq(approvals.id, approvalId));
expect(approval?.status).toBe("cancelled");
});
it("makes concurrent provisioning lose cleanly instead of creating duplicates", async () => {
const companyId = await seedCompany({ requireApproval: false });
const svc = builtInAgentService(db);
const [first, second] = await Promise.all([
svc.ensure(companyId, "briefs"),
svc.ensure(companyId, "briefs"),
]);
expect(first.agentId).toBeTruthy();
expect(second.agentId).toBe(first.agentId);
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(
rows.filter((row) => row.status !== "terminated" && readBuiltInAgentMarker(row.metadata)?.key === "briefs"),
).toHaveLength(1);
// The partial unique index now rejects any further active marked row.
await expect(
db.insert(agents).values({
id: randomUUID(),
companyId,
name: "Briefs Dupe",
role: "general",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["briefs"] }),
}),
).rejects.toMatchObject({
cause: {
code: "23505",
constraint_name: BUILT_IN_MARKER_UNIQUE_INDEX,
},
});
});
it("makes concurrent board-gated provisioning resolve to a single pending agent and approval", async () => {
const companyId = await seedCompany({ requireApproval: true });
const svc = builtInAgentService(db);
const results = await Promise.all([
svc.provision(companyId, "briefs"),
svc.provision(companyId, "briefs"),
]);
const agentIds = new Set(results.map((result) => result.state.agentId));
expect(agentIds.size).toBe(1);
const rows = await db.select().from(agents).where(eq(agents.companyId, companyId));
expect(
rows.filter((row) => row.status !== "terminated" && readBuiltInAgentMarker(row.metadata)?.key === "briefs"),
).toHaveLength(1);
const openApprovals = await db
.select()
.from(approvals)
.where(and(eq(approvals.companyId, companyId), eq(approvals.status, "pending")));
const briefsApprovals = openApprovals.filter(
(row) => (row.payload as { sourceBuiltInAgentKey?: string } | null)?.sourceBuiltInAgentKey === "briefs",
);
expect(briefsApprovals).toHaveLength(1);
// The winner returns its freshly created approval; the loser returns the
// same approval or null (if it re-resolves before that row commits), never
// a second one.
expect(results.some((result) => result.approval?.id === briefsApprovals[0]!.id)).toBe(true);
for (const result of results) {
if (result.approval) {
expect(result.approval.id).toBe(briefsApprovals[0]!.id);
}
}
});
it("self-heals duplicates during startup reconciliation without aborting later companies", async () => {
const affectedCompanyId = await seedCompany({ requireApproval: false });
const { olderId, newerId } = await seedLegacyDuplicateBriefs(affectedCompanyId);
// A second company created after the affected one — previously skipped
// entirely because the duplicate error escaped the reconciliation loop.
// Give it a drifted built-in row so we can prove reconcile still reached it.
const healthyCompanyId = await seedCompany({ requireApproval: false });
const healthyBriefsId = randomUUID();
await db.insert(agents).values({
id: healthyBriefsId,
companyId: healthyCompanyId,
name: "Stale Briefs",
role: "engineer",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
metadata: withBuiltInAgentMarker({}, { key: "briefs", featureKeys: ["briefs"] }),
});
const result = await reconcileBuiltInAgentsOnStartup(db);
expect(result.companyFailures).toBe(0);
const affectedRows = await db.select().from(agents).where(eq(agents.companyId, affectedCompanyId));
const affectedById = new Map(affectedRows.map((row) => [row.id, row]));
expect(affectedById.get(olderId)?.status).toBe("idle");
expect(affectedById.get(newerId)?.status).toBe("terminated");
expect(
affectedRows.filter(
(row) => row.status !== "terminated" && readBuiltInAgentMarker(row.metadata)?.key === "briefs",
),
).toHaveLength(1);
// The company after the affected one was still reconciled (its drifted
// built-in row was repaired to stock) rather than skipped.
const [healthyBriefs] = await db.select().from(agents).where(eq(agents.id, healthyBriefsId));
expect(healthyBriefs?.name).toBe("Briefs Agent");
});
it("automatically materializes the Reflection Coach bundle without enabling background work", async () => {
const companyId = await seedCompany();
const root = await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const state = await builtInAgentService(db).ensure(companyId, "reflection-coach");
expect(state.agent).toMatchObject({
companyId,
name: "Reflection Coach",
title: "Reflection Coach",
icon: "eye",
reportsTo: root.id,
adapterType: "codex_local",
budgetMonthlyCents: 0,
});
expect(state.status).toBe("paused");
expect(readBuiltInAgentMarker(state.agent?.metadata)).toEqual({
key: "reflection-coach",
featureKeys: ["reflection-coach"],
});
expect(state.resources.map((resource) => [resource.resourceKind, resource.stockStatus])).toEqual([
["instructions", "stock_current"],
["skill", "stock_current"],
["routine", "stock_current"],
]);
const reported = await builtInAgentService(db).get(companyId, "reflection-coach");
const reportedRoutine = reported.resources.find((resource) => resource.resourceKind === "routine");
expect(reportedRoutine).toMatchObject({
stockStatus: "stock_current",
updateAvailable: false,
resetAvailable: false,
});
expect(reportedRoutine?.currentHash).toBe(reportedRoutine?.stockHash);
const [skill] = await db
.select()
.from(companySkills)
.where(eq(companySkills.key, "paperclipai/bundled/paperclip-operations/reflection-coach"));
expect(skill).toMatchObject({
key: "paperclipai/bundled/paperclip-operations/reflection-coach",
slug: "reflection-coach",
});
expect(readPaperclipSkillSyncPreference(state.agent!.adapterConfig).desiredSkills).toContain(skill!.key);
const [routine] = await db.select().from(routines).where(eq(routines.companyId, companyId));
expect(routine).toMatchObject({
title: "Review recent agent trajectories for coaching proposals",
status: "paused",
assigneeAgentId: state.agentId,
originKind: "built_in_agent_bundle",
originId: "reflection-coach:recent-agent-reflection",
});
const [trigger] = await db.select().from(routineTriggers).where(eq(routineTriggers.routineId, routine!.id));
expect(trigger).toMatchObject({
kind: "schedule",
enabled: false,
cronExpression: "0 9 * * 1",
timezone: "UTC",
});
const grantKeys = await permissionKeysForAgent(state.agentId!);
expect(grantKeys).toEqual(expect.arrayContaining(["agents:suggest-changes", "skills:suggest-changes"]));
expect(grantKeys).not.toContain("tasks:assign");
expect(grantKeys).not.toContain("agents:configure");
expect(grantKeys).not.toContain("skills:create");
});
it("materializes the Summarizer bundle paused on Claude Haiku with a disabled routine", async () => {
const companyId = await seedCompany();
const root = await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const state = await builtInAgentService(db).ensure(companyId, "summarizer");
expect(state.agent).toMatchObject({
companyId,
name: "Summarizer",
title: "Summarizer",
icon: "sparkles",
role: "general",
reportsTo: root.id,
adapterType: "claude_local",
adapterConfig: { model: "claude-haiku-4-5" },
budgetMonthlyCents: 0,
});
expect(state.status).toBe("paused");
expect(state.pauseReason).toBe("Built-in Summarizer is disabled until explicitly configured.");
expect(readBuiltInAgentMarker(state.agent?.metadata)).toEqual({
key: "summarizer",
featureKeys: ["summarizer"],
});
expect(state.resources.map((resource) => [resource.resourceKind, resource.stockStatus])).toEqual([
["instructions", "stock_current"],
["skill", "stock_current"],
["routine", "stock_current"],
]);
const [skill] = await db
.select()
.from(companySkills)
.where(eq(companySkills.key, "paperclipai/bundled/paperclip-operations/summarize-status"));
expect(skill).toMatchObject({
key: "paperclipai/bundled/paperclip-operations/summarize-status",
slug: "summarize-status",
});
expect(readPaperclipSkillSyncPreference(state.agent!.adapterConfig).desiredSkills).toContain(skill!.key);
const [routine] = await db
.select()
.from(routines)
.where(and(eq(routines.companyId, companyId), eq(routines.assigneeAgentId, state.agentId!)));
expect(routine).toMatchObject({
title: "Refresh stale summary slots",
status: "paused",
assigneeAgentId: state.agentId,
originKind: "built_in_agent_bundle",
originId: "summarizer:refresh-stale-summaries",
});
const [trigger] = await db.select().from(routineTriggers).where(eq(routineTriggers.routineId, routine!.id));
expect(trigger).toMatchObject({
kind: "schedule",
enabled: false,
cronExpression: "0 8 * * *",
timezone: "UTC",
});
});
it("keeps the Summarizer not-configured until an adapter model is set", async () => {
const companyId = await seedCompany();
await expect(builtInAgentService(db).requireBuiltInAgent(companyId, "summarizer")).rejects.toMatchObject({
status: 412,
details: {
code: "built_in_agent_not_configured",
key: "summarizer",
},
});
// Provisioning without a model leaves it paused (default), still not runnable as "ready".
const paused = await builtInAgentService(db).ensure(companyId, "summarizer");
expect(paused.status).toBe("paused");
await expect(builtInAgentService(db).requireBuiltInAgent(companyId, "summarizer")).resolves.toMatchObject({
warning: { code: "built_in_agent_paused", key: "summarizer" },
});
});
it("restores Summarizer instruction drift on reset", async () => {
const companyId = await seedCompany();
const builtIns = builtInAgentService(db);
const created = await builtIns.ensure(companyId, "summarizer");
const instructions = agentInstructionsService();
await writeInstructionEntry(created.agent!, "AGENTS.md", "# Custom Summarizer\n\nOperator edit.\n");
const reconciled = await builtIns.ensure(companyId, "summarizer");
expect(reconciled.resources.find((resource) => resource.resourceKind === "instructions")).toMatchObject({
stockStatus: "operator_modified",
resetAvailable: true,
changedFiles: ["AGENTS.md"],
});
const reset = await builtIns.reset(companyId, "summarizer", {}, instructionOperator);
expect(reset.resources.find((resource) => resource.resourceKind === "instructions")).toMatchObject({
stockStatus: "stock_current",
resetAvailable: false,
});
const resetFile = await instructions.readFile(reset.agent!, "AGENTS.md");
expect(resetFile.content).toContain("Summarizer");
expect(resetFile.content).toContain("<<<SUMMARY-DRAFT>>>");
expect(resetFile.content).toContain("<<<END-SUMMARY-DRAFT>>>");
expect(resetFile.content).not.toContain("Operator edit.");
});
it("controls the Reflection Coach routine schedule without enabling it by default", async () => {
const companyId = await seedCompany();
await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const builtIns = builtInAgentService(db);
const created = await builtIns.ensure(companyId, "reflection-coach");
expect(created.status).toBe("paused");
expect(created.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
stockStatus: "stock_current",
scheduleEnabled: false,
});
const enabled = await builtIns.enableRoutineSchedule(
companyId,
"reflection-coach",
"recent-agent-reflection",
{ userId: "board-user" },
);
expect(enabled.status).toBe("needs_setup");
expect(enabled.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
stockStatus: "stock_current",
scheduleEnabled: true,
});
const [enabledRoutine] = await db.select().from(routines).where(eq(routines.companyId, companyId));
const [enabledTrigger] = await db.select().from(routineTriggers).where(eq(routineTriggers.routineId, enabledRoutine!.id));
expect(enabledRoutine).toMatchObject({ status: "active" });
expect(enabledTrigger).toMatchObject({ enabled: true });
const disabled = await builtIns.disableRoutineSchedule(
companyId,
"reflection-coach",
"recent-agent-reflection",
{ userId: "board-user" },
);
expect(disabled.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
stockStatus: "stock_current",
scheduleEnabled: false,
});
const [disabledRoutine] = await db.select().from(routines).where(eq(routines.id, enabledRoutine!.id));
const [disabledTrigger] = await db.select().from(routineTriggers).where(eq(routineTriggers.id, enabledTrigger!.id));
expect(disabledRoutine).toMatchObject({ status: "paused" });
expect(disabledTrigger).toMatchObject({ enabled: false });
});
it("surfaces pending Reflection Coach proposal interactions on the routine resource", async () => {
const companyId = await seedCompany();
await agentService(db).create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const created = await builtInAgentService(db).ensure(companyId, "reflection-coach");
const proposalIssueId = randomUUID();
await db.insert(issues).values({
id: proposalIssueId,
companyId,
title: "Review Reflection Coach proposal",
status: "in_review",
priority: "medium",
identifier: `${issuePrefix(companyId)}-42`,
issueNumber: 42,
assigneeAgentId: created.agentId,
createdByAgentId: created.agentId,
});
const interactionId = randomUUID();
await db.insert(issueThreadInteractions).values({
id: interactionId,
companyId,
issueId: proposalIssueId,
kind: "request_confirmation",
status: "pending",
continuationPolicy: "wake_assignee",
title: "Review proposed coaching change",
summary: "Accept or reject the proposed update.",
createdByAgentId: created.agentId,
payload: {
version: 1,
prompt: "Accept the proposed coaching change?",
acceptLabel: "Accept",
rejectLabel: "Reject",
},
});
const state = await builtInAgentService(db).get(companyId, "reflection-coach");
expect(state.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
pendingUpdateInteractionId: interactionId,
pendingUpdateIssueId: proposalIssueId,
pendingUpdateIssueIdentifier: `${issuePrefix(companyId)}-42`,
});
});
it("gates Reflection Coach proposal mutations until an accepted follow-up apply step", async () => {
const companyId = await seedCompany();
const agentsSvc = agentService(db);
await agentsSvc.create(companyId, {
name: "CEO",
role: "ceo",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const target = await agentsSvc.create(companyId, {
name: "Target Coder",
role: "engineer",
status: "idle",
adapterType: "codex_local",
adapterConfig: { model: "gpt-5.4" },
runtimeConfig: {},
permissions: {},
});
const created = await builtInAgentService(db).ensure(companyId, "reflection-coach");
const coach = created.agent!;
const instructionsSvc = agentInstructionsService();
const originalInstructions = "# Target Coder\n\nWork from the assigned issue.\n";
const prepared = await writeInstructionEntry(target, "AGENTS.md", originalInstructions);
let persistedTarget = (await agentsSvc.update(target.id, { adapterConfig: prepared.adapterConfig }))!;
const interactionsSvc = issueThreadInteractionService(db);
const applyAcceptedProposalFollowUp = async (input: {
interactionId: string;
nextInstructions: string;
}) => {
const interaction = await interactionsSvc.getById(input.interactionId);
if (interaction?.kind !== "request_confirmation" || interaction.status !== "accepted") {
return false;
}
const written = await writeInstructionEntry(persistedTarget, "AGENTS.md", input.nextInstructions);
persistedTarget = (await agentsSvc.update(persistedTarget.id, { adapterConfig: written.adapterConfig }))!;
return true;
};
const readTargetInstructions = async () =>
(await instructionsSvc.readFile(persistedTarget, "AGENTS.md")).content;
const proposalIssueId = randomUUID();
await db.insert(issues).values({
id: proposalIssueId,
companyId,
title: "Review Reflection Coach proposal",
status: "in_review",
priority: "medium",
identifier: `${issuePrefix(companyId)}-43`,
issueNumber: 43,
assigneeUserId: "board-user",
createdByAgentId: coach.id,
});
const acceptedInstructions = `${originalInstructions}\nWhen finishing, name the exact verification command.\n`;
const acceptedProposal = await interactionsSvc.create({
id: proposalIssueId,
companyId,
}, {
kind: "request_confirmation",
continuationPolicy: "wake_assignee_on_accept",
title: "Review proposed coaching change",
summary: "Accept or reject the proposed instruction diff.",
payload: {
version: 1,
prompt: "Apply this Reflection Coach instruction diff in a follow-up run?",
acceptLabel: "Accept",
rejectLabel: "Reject",
detailsMarkdown: [
"```diff",
" # Target Coder",
"",
" Work from the assigned issue.",
"+When finishing, name the exact verification command.",
"```",
].join("\n"),
target: {
type: "custom",
key: `agent:${target.id}:instructions`,
revisionId: "proposal-v1",
label: "Target Coder AGENTS.md diff",
},
},
}, {
agentId: coach.id,
});
const accepted = await interactionsSvc.acceptInteraction(
{ id: proposalIssueId, companyId, goalId: null, projectId: null },
acceptedProposal.id,
{},
{ userId: "board-user" },
);
expect(accepted.interaction).toMatchObject({
id: acceptedProposal.id,
kind: "request_confirmation",
status: "accepted",
});
expect(accepted.continuationIssue).toMatchObject({
id: proposalIssueId,
assigneeAgentId: coach.id,
assigneeUserId: null,
status: "todo",
});
expect(await readTargetInstructions()).toBe(originalInstructions);
await expect(applyAcceptedProposalFollowUp({
interactionId: acceptedProposal.id,
nextInstructions: acceptedInstructions,
})).resolves.toBe(true);
expect(await readTargetInstructions()).toBe(acceptedInstructions);
const rejectedProposal = await interactionsSvc.create({
id: proposalIssueId,
companyId,
}, {
kind: "request_confirmation",
continuationPolicy: "wake_assignee_on_accept",
idempotencyKey: "reflection-coach:proposal-v2",
title: "Review rejected coaching change",
summary: "Rejecting this diff must not mutate the target instructions.",
payload: {
version: 1,
prompt: "Apply this rejected Reflection Coach instruction diff?",
acceptLabel: "Accept",
rejectLabel: "Reject",
detailsMarkdown: [
"```diff",
"+This rejected line must not be applied.",
"```",
].join("\n"),
target: {
type: "custom",
key: `agent:${target.id}:instructions`,
revisionId: "proposal-v2",
label: "Target Coder AGENTS.md rejected diff",
},
},
}, {
agentId: coach.id,
});
const rejected = await interactionsSvc.rejectInteraction(
{ id: proposalIssueId, companyId },
rejectedProposal.id,
{ reason: "Not the right rule." },
{ userId: "board-user" },
);
expect(rejected).toMatchObject({
id: rejectedProposal.id,
kind: "request_confirmation",
status: "rejected",
result: expect.objectContaining({
outcome: "rejected",
reason: "Not the right rule.",
}),
});
await expect(applyAcceptedProposalFollowUp({
interactionId: rejectedProposal.id,
nextInstructions: `${acceptedInstructions}\nThis rejected line must not be applied.\n`,
})).resolves.toBe(false);
expect(await readTargetInstructions()).toBe(acceptedInstructions);
});
it("preserves Reflection Coach stock drift until explicit reset", async () => {
const companyId = await seedCompany();
const created = await builtInAgentService(db).ensure(companyId, "reflection-coach");
const agent = created.agent!;
const instructionsSvc = agentInstructionsService();
await writeInstructionEntry(agent, "AGENTS.md", "# Custom Reflection Coach\n\nDo not overwrite me.\n");
await db
.update(companySkills)
.set({ markdown: "---\nname: reflection-coach\n---\n\n# Custom skill\n" })
.where(eq(companySkills.companyId, companyId));
await db
.update(routines)
.set({ title: "DRIFTED BY TEST - do not clobber" })
.where(eq(routines.companyId, companyId));
const drifted = await builtInAgentService(db).ensure(companyId, "reflection-coach");
expect(drifted.resources.find((resource) => resource.resourceKind === "instructions")).toMatchObject({
stockStatus: "operator_modified",
resetAvailable: true,
});
expect(drifted.resources.find((resource) => resource.resourceKind === "skill")).toMatchObject({
stockStatus: "operator_modified",
resetAvailable: true,
});
expect(drifted.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
stockStatus: "operator_modified",
resetAvailable: true,
});
expect((await instructionsSvc.readFile(drifted.agent!, "AGENTS.md")).content).toContain("Do not overwrite me.");
const [preservedRoutine] = await db.select().from(routines).where(eq(routines.companyId, companyId));
expect(preservedRoutine?.title).toBe("DRIFTED BY TEST - do not clobber");
const reset = await builtInAgentService(db).reset(companyId, "reflection-coach", {
resources: ["instructions", "routine"],
}, instructionOperator);
expect(reset.resources.find((resource) => resource.resourceKind === "instructions")).toMatchObject({
stockStatus: "stock_current",
resetAvailable: false,
});
expect(reset.resources.find((resource) => resource.resourceKind === "skill")).toMatchObject({
stockStatus: "operator_modified",
resetAvailable: true,
});
expect(reset.resources.find((resource) => resource.resourceKind === "routine")).toMatchObject({
stockStatus: "stock_current",
resetAvailable: false,
});
expect((await instructionsSvc.readFile(reset.agent!, "AGENTS.md")).content).toContain("You are Reflection Coach");
const [resetRoutine] = await db.select().from(routines).where(eq(routines.companyId, companyId));
expect(resetRoutine?.title).toBe("Review recent agent trajectories for coaching proposals");
});
});