add support for -w,--writable-root to add more writable roots for sandbox (#263)

This adds support for a new flag, `-w,--writable-root`, that can be
specified multiple times to _amend_ the list of folders that should be
configured as "writable roots" by the sandbox used in `full-auto` mode.
Values that are passed as relative paths will be resolved to absolute
paths.

Incidentally, this required updating a number of the `agent*.test.ts`
files: it feels like some of the setup logic across those tests could be
consolidated.

In my testing, it seems that this might be slightly out of distribution
for the model, as I had to explicitly tell it to run `apply_patch` and
that it had the permissions to write those files (initially, it just
showed me a diff and told me to apply it myself). Nevertheless, I think
this is a good starting point.
This commit is contained in:
Michael Bolin
2025-04-17 15:39:26 -07:00
committed by GitHub
parent d5eed65963
commit ae5b1b5cb5
22 changed files with 103 additions and 13 deletions

View File

@@ -88,6 +88,7 @@ describe("cancel before first function_call", () => {
const { _test } = (await import("openai")) as any;
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,

View File

@@ -99,6 +99,7 @@ describe("cancel clears previous_response_id", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: () => {},
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -92,6 +92,7 @@ describe("Agent cancellation race", () => {
const items: Array<any> = [];
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "any",
instructions: "",
config: { model: "any", instructions: "" },

View File

@@ -91,6 +91,7 @@ describe("Agent cancellation", () => {
instructions: "",
config: { model: "any", instructions: "" },
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (item) => {
received.push(item);
},
@@ -136,6 +137,7 @@ describe("Agent cancellation", () => {
const received: Array<any> = [];
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "any",
instructions: "",
config: { model: "any", instructions: "" },

View File

@@ -118,6 +118,7 @@ describe("function_call_output includes original call ID", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: () => {},
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -56,6 +56,7 @@ describe("AgentLoop generic network/server errors", () => {
const received: Array<any> = [];
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
@@ -99,6 +100,7 @@ describe("AgentLoop generic network/server errors", () => {
const received: Array<any> = [];
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,

View File

@@ -34,6 +34,7 @@ describe("Agent interrupt and continue", () => {
// Create the agent
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "test-model",
instructions: "",
approvalPolicy: { mode: "auto" } as any,

View File

@@ -58,6 +58,7 @@ describe("AgentLoop invalid request / 4xx errors", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => received.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -58,6 +58,7 @@ describe("AgentLoop max_tokens too large error", () => {
const received: Array<any> = [];
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,

View File

@@ -109,6 +109,7 @@ describe("AgentLoop network resilience", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => received.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),
@@ -150,6 +151,7 @@ describe("AgentLoop network resilience", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => received.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -112,6 +112,7 @@ describe("AgentLoop", () => {
expect(config.instructions).toContain("Hello docs!");
const agent = new AgentLoop({
additionalWritableRoots: [],
model: "o3", // arbitrary
instructions: config.instructions,
config,

View File

@@ -79,6 +79,7 @@ describe("AgentLoop ratelimit handling", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => received.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -97,6 +97,7 @@ describe("AgentLoop automatic retry on 5xx errors", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => received.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),
@@ -134,6 +135,7 @@ describe("AgentLoop automatic retry on 5xx errors", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => received.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -82,7 +82,14 @@ describe("Agent terminate (hard cancel)", () => {
it("suppresses function_call_output and stops processing once terminate() is invoked", async () => {
// Simulate a longrunning exec that would normally resolve with output.
vi.spyOn(handleExec, "handleExecCommand").mockImplementation(
async (_args, _config, _policy, _getConf, abortSignal) => {
async (
_args,
_config,
_policy,
_additionalWritableRoots,
_getConf,
abortSignal,
) => {
// Wait until the abort signal is fired or 2s (whichever comes first).
await new Promise<void>((resolve) => {
if (abortSignal?.aborted) {
@@ -106,6 +113,7 @@ describe("Agent terminate (hard cancel)", () => {
instructions: "",
config: { model: "any", instructions: "" },
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (item) => received.push(item),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),
@@ -141,6 +149,7 @@ describe("Agent terminate (hard cancel)", () => {
instructions: "",
config: { model: "any", instructions: "" },
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: () => {},
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -107,6 +107,7 @@ describe("thinking time counter", () => {
model: "any",
instructions: "",
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (i) => items.push(i),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),

View File

@@ -53,10 +53,12 @@ describe("handleExecCommand invalid executable", () => {
const policy = { mode: "auto" } as any;
const getConfirmation = async () => ({ review: "yes" } as any);
const additionalWritableRoots: Array<string> = [];
const { outputText, metadata } = await handleExecCommand(
execInput,
config,
policy,
additionalWritableRoots,
getConfirmation,
);