feat(adapters/gemini-local): add auth detection, turn-limit handling, sandbox, and approval modes

Incorporate improvements from PR #13 and #105 into the gemini-local adapter:

- Add detectGeminiAuthRequired() for runtime auth failure detection with
  errorCode: "gemini_auth_required" on execution results
- Add isGeminiTurnLimitResult() to detect exit code 53 / turn_limit status
  and clear session to prevent stuck sessions on next heartbeat
- Add describeGeminiFailure() for structured error messages from parsed
  result events including errors array extraction
- Return parsed resultEvent in resultJson instead of raw stdout/stderr
- Add isRetry guard to prevent stale session ID fallback after retry
- Replace boolean yolo with approvalMode string (default/auto_edit/yolo)
  with backwards-compatible config.yolo fallback
- Add sandbox config option (--sandbox / --sandbox=none)
- Add GOOGLE_GENAI_USE_GCA auth detection in environment test
- Consolidate auth detection regex into shared detectGeminiAuthRequired()
- Add gemini-2.0-flash and gemini-2.0-flash-lite model IDs

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Aaron
2026-03-09 15:16:15 +00:00
parent ec445e4cc9
commit 4e5f67ef96
6 changed files with 142 additions and 22 deletions

View File

@@ -7,6 +7,8 @@ export const models = [
{ id: "gemini-2.5-pro", label: "Gemini 2.5 Pro" },
{ id: "gemini-2.5-flash", label: "Gemini 2.5 Flash" },
{ id: "gemini-2.5-flash-lite", label: "Gemini 2.5 Flash Lite" },
{ id: "gemini-2.0-flash", label: "Gemini 2.0 Flash" },
{ id: "gemini-2.0-flash-lite", label: "Gemini 2.0 Flash Lite" },
];
export const agentConfigurationDoc = `# gemini_local agent configuration
@@ -28,7 +30,8 @@ Core fields:
- instructionsFilePath (string, optional): absolute path to a markdown instructions file prepended to the run prompt
- promptTemplate (string, optional): run prompt template
- model (string, optional): Gemini model id. Defaults to auto.
- yolo (boolean, optional): pass --approval-mode yolo for unattended operation
- approvalMode (string, optional): "default", "auto_edit", or "yolo" (default: "default")
- sandbox (boolean, optional): run in sandbox mode (default: false, passes --sandbox=none)
- command (string, optional): defaults to "gemini"
- extraArgs (string[], optional): additional CLI args
- env (object, optional): KEY=VALUE environment variables

View File

@@ -19,7 +19,13 @@ import {
runChildProcess,
} from "@paperclipai/adapter-utils/server-utils";
import { DEFAULT_GEMINI_LOCAL_MODEL } from "../index.js";
import { isGeminiUnknownSessionError, parseGeminiJsonl } from "./parse.js";
import {
describeGeminiFailure,
detectGeminiAuthRequired,
isGeminiTurnLimitResult,
isGeminiUnknownSessionError,
parseGeminiJsonl,
} from "./parse.js";
import { firstNonEmptyLine } from "./utils.js";
const __moduleDir = path.dirname(fileURLToPath(import.meta.url));
@@ -93,7 +99,8 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
);
const command = asString(config.command, "gemini");
const model = asString(config.model, DEFAULT_GEMINI_LOCAL_MODEL).trim();
const yolo = asBoolean(config.yolo, false);
const approvalMode = asString(config.approvalMode, asBoolean(config.yolo, false) ? "yolo" : "default");
const sandbox = asBoolean(config.sandbox, false);
const workspaceContext = parseObject(context.paperclipWorkspace);
const workspaceCwd = asString(workspaceContext.cwd, "");
@@ -211,7 +218,7 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
}
const commandNotes = (() => {
const notes: string[] = ["Prompt is passed to Gemini as the final positional argument."];
if (yolo) notes.push("Added --approval-mode yolo for unattended execution.");
if (approvalMode !== "default") notes.push(`Added --approval-mode ${approvalMode} for unattended execution.`);
if (!instructionsFilePath) return notes;
if (instructionsPrefix.length > 0) {
notes.push(
@@ -242,7 +249,12 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
const args = ["--output-format", "stream-json"];
if (resumeSessionId) args.push("--resume", resumeSessionId);
if (model && model !== DEFAULT_GEMINI_LOCAL_MODEL) args.push("--model", model);
if (yolo) args.push("--approval-mode", "yolo");
if (approvalMode !== "default") args.push("--approval-mode", approvalMode);
if (sandbox) {
args.push("--sandbox");
} else {
args.push("--sandbox=none");
}
if (extraArgs.length > 0) args.push(...extraArgs);
args.push(prompt);
return args;
@@ -290,18 +302,31 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
parsed: ReturnType<typeof parseGeminiJsonl>;
},
clearSessionOnMissingSession = false,
isRetry = false,
): AdapterExecutionResult => {
const authMeta = detectGeminiAuthRequired({
parsed: attempt.parsed.resultEvent,
stdout: attempt.proc.stdout,
stderr: attempt.proc.stderr,
});
if (attempt.proc.timedOut) {
return {
exitCode: attempt.proc.exitCode,
signal: attempt.proc.signal,
timedOut: true,
errorMessage: `Timed out after ${timeoutSec}s`,
errorCode: authMeta.requiresAuth ? "gemini_auth_required" : null,
clearSession: clearSessionOnMissingSession,
};
}
const resolvedSessionId = attempt.parsed.sessionId ?? runtimeSessionId ?? runtime.sessionId ?? null;
const clearSessionForTurnLimit = isGeminiTurnLimitResult(attempt.parsed.resultEvent, attempt.proc.exitCode);
// On retry, don't fall back to old session ID — the old session was stale
const canFallbackToRuntimeSession = !isRetry;
const resolvedSessionId = attempt.parsed.sessionId
?? (canFallbackToRuntimeSession ? (runtimeSessionId ?? runtime.sessionId ?? null) : null);
const resolvedSessionParams = resolvedSessionId
? ({
sessionId: resolvedSessionId,
@@ -313,8 +338,12 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
: null;
const parsedError = typeof attempt.parsed.errorMessage === "string" ? attempt.parsed.errorMessage.trim() : "";
const stderrLine = firstNonEmptyLine(attempt.proc.stderr);
const structuredFailure = attempt.parsed.resultEvent
? describeGeminiFailure(attempt.parsed.resultEvent)
: null;
const fallbackErrorMessage =
parsedError ||
structuredFailure ||
stderrLine ||
`Gemini exited with code ${attempt.proc.exitCode ?? -1}`;
@@ -323,6 +352,7 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
signal: attempt.proc.signal,
timedOut: false,
errorMessage: (attempt.proc.exitCode ?? 0) === 0 ? null : fallbackErrorMessage,
errorCode: authMeta.requiresAuth ? "gemini_auth_required" : null,
usage: attempt.parsed.usage,
sessionId: resolvedSessionId,
sessionParams: resolvedSessionParams,
@@ -331,12 +361,12 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
model,
billingType,
costUsd: attempt.parsed.costUsd,
resultJson: {
resultJson: attempt.parsed.resultEvent ?? {
stdout: attempt.proc.stdout,
stderr: attempt.proc.stderr,
},
summary: attempt.parsed.summary,
clearSession: Boolean(clearSessionOnMissingSession && !resolvedSessionId),
clearSession: clearSessionForTurnLimit || Boolean(clearSessionOnMissingSession && !resolvedSessionId),
};
};
@@ -353,7 +383,7 @@ export async function execute(ctx: AdapterExecutionContext): Promise<AdapterExec
`[paperclip] Gemini resume session "${sessionId}" is unavailable; retrying with a fresh session.\n`,
);
const retry = await runAttempt(null);
return toResult(retry, true);
return toResult(retry, true, true);
}
return toResult(initial);

View File

@@ -1,6 +1,12 @@
export { execute } from "./execute.js";
export { testEnvironment } from "./test.js";
export { parseGeminiJsonl, isGeminiUnknownSessionError } from "./parse.js";
export {
parseGeminiJsonl,
isGeminiUnknownSessionError,
describeGeminiFailure,
detectGeminiAuthRequired,
isGeminiTurnLimitResult,
} from "./parse.js";
import type { AdapterSessionCodec } from "@paperclipai/adapter-utils";
function readNonEmptyString(value: unknown): string | null {

View File

@@ -77,6 +77,7 @@ export function parseGeminiJsonl(stdout: string) {
const messages: string[] = [];
let errorMessage: string | null = null;
let costUsd: number | null = null;
let resultEvent: Record<string, unknown> | null = null;
const usage = {
inputTokens: 0,
cachedInputTokens: 0,
@@ -101,6 +102,7 @@ export function parseGeminiJsonl(stdout: string) {
}
if (type === "result") {
resultEvent = event;
accumulateUsage(usage, event.usage ?? event.usageMetadata);
const resultText =
asString(event.result, "").trim() ||
@@ -151,6 +153,7 @@ export function parseGeminiJsonl(stdout: string) {
usage,
costUsd,
errorMessage,
resultEvent,
};
}
@@ -165,3 +168,75 @@ export function isGeminiUnknownSessionError(stdout: string, stderr: string): boo
haystack,
);
}
function extractGeminiErrorMessages(parsed: Record<string, unknown>): string[] {
const messages: string[] = [];
const errorMsg = asString(parsed.error, "").trim();
if (errorMsg) messages.push(errorMsg);
const raw = Array.isArray(parsed.errors) ? parsed.errors : [];
for (const entry of raw) {
if (typeof entry === "string") {
const msg = entry.trim();
if (msg) messages.push(msg);
continue;
}
if (typeof entry !== "object" || entry === null || Array.isArray(entry)) continue;
const obj = entry as Record<string, unknown>;
const msg = asString(obj.message, "") || asString(obj.error, "") || asString(obj.code, "");
if (msg) {
messages.push(msg);
continue;
}
try {
messages.push(JSON.stringify(obj));
} catch {
// skip non-serializable entry
}
}
return messages;
}
export function describeGeminiFailure(parsed: Record<string, unknown>): string | null {
const status = asString(parsed.status, "");
const errors = extractGeminiErrorMessages(parsed);
const detail = errors[0] ?? "";
const parts = ["Gemini run failed"];
if (status) parts.push(`status=${status}`);
if (detail) parts.push(detail);
return parts.length > 1 ? parts.join(": ") : null;
}
const GEMINI_AUTH_REQUIRED_RE = /(?:not\s+authenticated|please\s+authenticate|api[_ ]?key\s+(?:required|missing|invalid)|authentication\s+required|unauthorized|invalid\s+credentials|GEMINI_API_KEY|GOOGLE_API_KEY|not\s+logged\s+in|login\s+required|run\s+`?gemini\s+auth(?:\s+login)?`?\s+first)/i;
export function detectGeminiAuthRequired(input: {
parsed: Record<string, unknown> | null;
stdout: string;
stderr: string;
}): { requiresAuth: boolean } {
const errors = extractGeminiErrorMessages(input.parsed ?? {});
const messages = [...errors, input.stdout, input.stderr]
.join("\n")
.split(/\r?\n/)
.map((line) => line.trim())
.filter(Boolean);
const requiresAuth = messages.some((line) => GEMINI_AUTH_REQUIRED_RE.test(line));
return { requiresAuth };
}
export function isGeminiTurnLimitResult(
parsed: Record<string, unknown> | null | undefined,
exitCode?: number | null,
): boolean {
if (exitCode === 53) return true;
if (!parsed) return false;
const status = asString(parsed.status, "").trim().toLowerCase();
if (status === "turn_limit" || status === "max_turns") return true;
const error = asString(parsed.error, "").trim();
return /turn\s*limit|max(?:imum)?\s+turns?/i.test(error);
}

View File

@@ -15,7 +15,7 @@ import {
runChildProcess,
} from "@paperclipai/adapter-utils/server-utils";
import { DEFAULT_GEMINI_LOCAL_MODEL } from "../index.js";
import { parseGeminiJsonl } from "./parse.js";
import { detectGeminiAuthRequired, parseGeminiJsonl } from "./parse.js";
import { firstNonEmptyLine } from "./utils.js";
function summarizeStatus(checks: AdapterEnvironmentCheck[]): AdapterEnvironmentTestResult["status"] {
@@ -41,9 +41,6 @@ function summarizeProbeDetail(stdout: string, stderr: string, parsedError: strin
return clean.length > max ? `${clean.slice(0, max - 1)}` : clean;
}
const GEMINI_AUTH_REQUIRED_RE =
/(?:authentication\s+required|not\s+authenticated|not\s+logged\s+in|login\s+required|unauthorized|invalid(?:\s+or\s+missing)?\s+api(?:[_\s-]?key)?|gemini[_\s-]?api[_\s-]?key|google[_\s-]?api[_\s-]?key|run\s+`?gemini\s+auth(?:\s+login)?`?\s+first|api(?:[_\s-]?key)?(?:\s+is)?\s+required)/i;
export async function testEnvironment(
ctx: AdapterEnvironmentTestContext,
): Promise<AdapterEnvironmentTestResult> {
@@ -94,15 +91,19 @@ export async function testEnvironment(
const hostGeminiApiKey = process.env.GEMINI_API_KEY;
const configGoogleApiKey = env.GOOGLE_API_KEY;
const hostGoogleApiKey = process.env.GOOGLE_API_KEY;
const hasGca = env.GOOGLE_GENAI_USE_GCA === "true" || process.env.GOOGLE_GENAI_USE_GCA === "true";
if (
isNonEmpty(configGeminiApiKey) ||
isNonEmpty(hostGeminiApiKey) ||
isNonEmpty(configGoogleApiKey) ||
isNonEmpty(hostGoogleApiKey)
isNonEmpty(hostGoogleApiKey) ||
hasGca
) {
const source = isNonEmpty(configGeminiApiKey) || isNonEmpty(configGoogleApiKey)
? "adapter config env"
: "server environment";
const source = hasGca
? "Google account login (GCA)"
: isNonEmpty(configGeminiApiKey) || isNonEmpty(configGoogleApiKey)
? "adapter config env"
: "server environment";
checks.push({
code: "gemini_api_key_present",
level: "info",
@@ -114,7 +115,7 @@ export async function testEnvironment(
code: "gemini_api_key_missing",
level: "warn",
message: "No Gemini API key was detected. Gemini runs may fail until auth is configured.",
hint: "Set GEMINI_API_KEY or GOOGLE_API_KEY in adapter env/shell, or run `gemini auth` / `gemini auth login`.",
hint: "Set GEMINI_API_KEY or GOOGLE_API_KEY in adapter env/shell, run `gemini auth` / `gemini auth login`, or set GOOGLE_GENAI_USE_GCA=true for Google account auth.",
});
}
@@ -158,7 +159,11 @@ export async function testEnvironment(
);
const parsed = parseGeminiJsonl(probe.stdout);
const detail = summarizeProbeDetail(probe.stdout, probe.stderr, parsed.errorMessage);
const authEvidence = `${parsed.errorMessage ?? ""}\n${probe.stdout}\n${probe.stderr}`.trim();
const authMeta = detectGeminiAuthRequired({
parsed: parsed.resultEvent,
stdout: probe.stdout,
stderr: probe.stderr,
});
if (probe.timedOut) {
checks.push({
@@ -183,7 +188,7 @@ export async function testEnvironment(
hint: "Try `gemini --output-format json \"Respond with hello.\"` manually to inspect full output.",
}),
});
} else if (GEMINI_AUTH_REQUIRED_RE.test(authEvidence)) {
} else if (authMeta.requiresAuth) {
checks.push({
code: "gemini_hello_probe_auth_required",
level: "warn",

View File

@@ -67,7 +67,8 @@ export function buildGeminiLocalConfig(v: CreateConfigValues): Record<string, un
}
}
if (Object.keys(env).length > 0) ac.env = env;
if (v.dangerouslyBypassSandbox) ac.yolo = true;
if (v.dangerouslyBypassSandbox) ac.approvalMode = "yolo";
ac.sandbox = !v.dangerouslyBypassSandbox;
if (v.command) ac.command = v.command;
if (v.extraArgs) ac.extraArgs = parseCommaArgs(v.extraArgs);
return ac;