auto-model-router 0.8.1 → 0.10.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -7,14 +7,14 @@
7
7
  },
8
8
  "metadata": {
9
9
  "description": "auto-model-router: a local cost/complexity-aware model router for Oh My Pi, backed by OpenRouter",
10
- "version": "0.8.1",
10
+ "version": "0.10.0",
11
11
  "pluginRoot": "."
12
12
  },
13
13
  "plugins": [
14
14
  {
15
15
  "name": "auto-model-router",
16
16
  "description": "Local cost/complexity-aware model router for Oh My Pi, backed by OpenRouter. Runs in-process, routes per turn by price and task complexity, with budget caps, mid-stream escalation, and cache-aware hysteresis.",
17
- "version": "0.8.1",
17
+ "version": "0.10.0",
18
18
  "author": {
19
19
  "name": "drewappling",
20
20
  "email": "drewappling@gmail.com"
package/README.md CHANGED
@@ -1312,18 +1312,33 @@ refresh token beside the key (`--refresh-token`, `--key-expires`, `--refresh-exp
1312
1312
  new key a day before expiry, at session start, and re-writes every config `connect`
1313
1313
  wrote; `auto-model-router refresh` does the same by hand (`--force` to do it early), and
1314
1314
  `auto-model-router token` prints a key that is good right now, refreshing first if needed —
1315
- the shape a harness key-helper wants (Claude Code's `apiKeyHelper`). The remote keeps the
1316
- old key valid until its own expiry, so a session still holding it is never cut. A refresh
1317
- token presented twice means the credential was copied: the remote revokes that device, and
1318
- the machine onboards again.
1315
+ the shape a harness key-helper wants. The remote keeps the old key valid until its own
1316
+ expiry, so a session still holding it is never cut. A refresh token presented twice means
1317
+ the credential was copied: the remote revokes that device, and the machine onboards again.
1318
+
1319
+ The refresh token, the long-lived secret, does not sit in a file: `connect` puts it in the
1320
+ operating system's credential store — DPAPI on Windows (ciphertext in
1321
+ `<router home>/refresh.dpapi`, decryptable only by that Windows user on that machine),
1322
+ the login keychain on macOS, the Secret Service on Linux — and `remote.json` only names
1323
+ which store holds it. `<router home>/refresh.token`, owner-readable only, is the fallback
1324
+ when no store is usable (a CI box), and the note at the end of `connect` says when that
1325
+ happened. **Claude Code** gets its settings file written instead of an environment: the
1326
+ `env` block carries `ANTHROPIC_BASE_URL`, and `apiKeyHelper` runs
1327
+ `auto-model-router token`, so no key is in its environment or on disk for it and a
1328
+ short-lived key rotates underneath a running session.
1319
1329
 
1320
1330
  In remote mode omp sends `X-Agentdox-Scope` derived from the workspace folder, so one
1321
1331
  remote router serves every repo on the machine with that repo's shared context. The remote
1322
1332
  decides what to do with it: a team edition that pins a scope on the member's group
1323
- overrides it, and one that pins none follows the workspace. That header rides on the roles
1324
- the extensions register; the **main** model's handle comes from `models.yml`, which is
1325
- machine-wide, so it carries a scope only if you pass `--scope <slug>` to `connect` right
1326
- for a single-project machine, wrong for one with several repos.
1333
+ overrides it, and one that pins none follows the workspace. The roles the extensions
1334
+ register carry the header directly. The **main** model's handle comes from `models.yml`,
1335
+ which is machine-wide and resolved before extensions load, so `connect` writes that
1336
+ entry's header value as the *name* of an environment variable,
1337
+ `AUTO_MODEL_ROUTER_SCOPE`; omp resolves it from the environment on every request, and
1338
+ the embed extension sets it from the workspace as it loads. If the variable is unset omp
1339
+ sends the name itself, which the router does not accept as a scope (a scope is a
1340
+ lowercase slug) and falls back to its default. `--scope <slug>` pins one project for the
1341
+ whole machine instead; a refresh keeps a pin, and `connect --scope ""` removes it.
1327
1342
 
1328
1343
  ## Multiple coding harnesses, one router
1329
1344
 
@@ -1461,6 +1476,17 @@ move while the loop runs. So the router buffers the assistant's narration across
1461
1476
  the loop and writes it once, together with the closing synthesis, when the
1462
1477
  assistant actually yields back to the user.
1463
1478
 
1479
+ ### Utility calls get neither the block nor a transcript
1480
+
1481
+ A harness drives more than the agent's conversation through this provider: omp
1482
+ asks for a session title and a complexity rating with `model: auto`. Those
1483
+ calls answer *about* the conversation, carry no tool schemas, and gain nothing
1484
+ from the project block, yet each one paid the whole block — measured at ~6k
1485
+ prompt tokens per call, 42k across one turn's seven side calls. So the tool
1486
+ array is the discriminator for both directions: a request with no tools is
1487
+ neither recorded nor injected. `context.injectWithoutTools: true` restores
1488
+ injection into tool-less requests for a deliberately tool-less agent.
1489
+
1464
1490
  Write-backs are queued, bounded, and never awaited: agentdox is an enrichment,
1465
1491
  not a dependency. If it is unreachable the turn routes and dispatches normally,
1466
1492
  and a pinned block keeps being served.
@@ -169,6 +169,16 @@ scope now correct the bridge degrades to **inert** for other projects. Correct a
169
169
  means the bridge only helps projects the configured token actually grants. A multi-scope token
170
170
  would fix that, at the cost of one credential reaching every project.
171
171
 
172
+ ## 6b. FIXED — the same discriminator gates injection (0.10)
173
+
174
+ Recording skipped tool-less calls from 0.2.11, but injection did not: every scoped
175
+ request received the block, so omp's title and rating calls each carried ~6k tokens of
176
+ project context they could not use. Measured 2026-09-08 on one omp turn: seven side
177
+ calls, 23,412 chars each, 42,417 prompt tokens of context in total. `turn.ts` now
178
+ resolves the block only when `req.tools.length > 0`, the same test recording uses, and
179
+ leaves the conversation's pin untouched otherwise. `context.injectWithoutTools` (default
180
+ off) is the escape hatch for an agent that deliberately ships no tools.
181
+
172
182
  ## 7. Also worth doing
173
183
 
174
184
  - **Context pollution from test turns.** `context_assemble` includes recent session messages,
@@ -22,8 +22,16 @@ export interface RemoteRouter {
22
22
  userId: string;
23
23
  name: string;
24
24
  joinedAtMs: number;
25
- /** Present when the remote issues short-lived keys: trades for the next key (see src/cli/refresh.ts). */
25
+ /**
26
+ * Present only in a remote.json written before the credential store existed:
27
+ * the token inline. New files name the store instead (`refreshTokenStore`) and
28
+ * the token is read from it when a refresh happens (src/cli/refresh.ts).
29
+ */
26
30
  refreshToken?: string;
31
+ /** Which OS store holds the refresh token: dpapi (Windows), keychain (macOS), secret-service (Linux) or file. */
32
+ refreshTokenStore?: "dpapi" | "keychain" | "secret-service" | "file";
33
+ /** The account the store files it under (`<userId>@<remote host>`). */
34
+ refreshAccount?: string;
27
35
  keyExpiresAtMs?: number;
28
36
  refreshExpiresAtMs?: number;
29
37
  /** What the remote calls this machine. */
@@ -46,6 +54,8 @@ export function parseRemoteRouter(text: string): RemoteRouter | null {
46
54
  name: typeof raw.name === "string" ? raw.name : "",
47
55
  joinedAtMs: typeof raw.joinedAtMs === "number" ? raw.joinedAtMs : 0,
48
56
  ...(typeof raw.refreshToken === "string" && raw.refreshToken !== "" ? { refreshToken: raw.refreshToken } : {}),
57
+ ...(raw.refreshTokenStore === "dpapi" || raw.refreshTokenStore === "keychain" || raw.refreshTokenStore === "secret-service" || raw.refreshTokenStore === "file" ? { refreshTokenStore: raw.refreshTokenStore } : {}),
58
+ ...(typeof raw.refreshAccount === "string" && raw.refreshAccount !== "" ? { refreshAccount: raw.refreshAccount } : {}),
49
59
  ...(typeof raw.keyExpiresAtMs === "number" ? { keyExpiresAtMs: raw.keyExpiresAtMs } : {}),
50
60
  ...(typeof raw.refreshExpiresAtMs === "number" ? { refreshExpiresAtMs: raw.refreshExpiresAtMs } : {}),
51
61
  ...(typeof raw.device === "string" && raw.device !== "" ? { device: raw.device } : {}),
@@ -55,6 +65,22 @@ export function parseRemoteRouter(text: string): RemoteRouter | null {
55
65
  }
56
66
  }
57
67
 
68
+ /** True when this machine can trade for a new key: a refresh token inline, or a store that holds one. */
69
+ export function hasRefresh(remote: RemoteRouter): boolean {
70
+ return (remote.refreshToken !== undefined && remote.refreshToken !== "") || remote.refreshTokenStore !== undefined;
71
+ }
72
+
73
+ /** The account a remote user's refresh token is filed under in the OS store. */
74
+ export function refreshAccountOf(url: string, userId: string): string {
75
+ let host = url;
76
+ try {
77
+ host = new URL(url).host;
78
+ } catch {
79
+ /* keep the raw url */
80
+ }
81
+ return `${userId === "" ? "member" : userId}@${host}`;
82
+ }
83
+
58
84
  export function readRemoteRouter(routerHome: string): RemoteRouter | null {
59
85
  for (const path of [remoteFilePath(routerHome), join(routerHome, LEGACY_FILE)]) {
60
86
  if (!existsSync(path)) continue;
@@ -34,6 +34,19 @@ import type { RouterConfig } from "../src/config/types.ts";
34
34
  import type { ExtensionAPI } from "@oh-my-pi/pi-coding-agent";
35
35
 
36
36
  import { EMBED_DUMMY_API_KEY, EMBED_PROVIDER_ID, buildProviderConfig, deriveAgentdoxScope, embedPortPath, modelsYmlPort, probeEmbed, readEmbedPort, resolveEmbedPort, writeEmbedPort } from "./embed-logic.ts";
37
+ import { SCOPE_ENV } from "../src/context/scope.ts";
38
+
39
+ // The workspace's scope for the MAIN model. omp builds that handle from
40
+ // models.yml before this file loads, so its X-Agentdox-Scope cannot come from
41
+ // the provider we register; instead the managed entry names SCOPE_ENV as the
42
+ // header's value, omp resolves that from the environment on every request,
43
+ // and this module runs inside omp's process — so setting it here reaches
44
+ // every turn of this session, main and side roles alike. A workspace that
45
+ // derives no scope (no folder name) leaves whatever the shell set.
46
+ {
47
+ const workspaceScope = deriveAgentdoxScope(process.cwd());
48
+ if (workspaceScope !== "") process.env[SCOPE_ENV] = workspaceScope;
49
+ }
37
50
 
38
51
  /** omp's models.yml as text, or "" when it does not exist / cannot be read. */
39
52
  function readModelsYml(): string {
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "auto-model-router",
3
- "version": "0.8.1",
3
+ "version": "0.10.0",
4
4
  "private": false,
5
5
  "description": "Local cost/complexity-aware model router for Oh My Pi, backed by OpenRouter",
6
6
  "type": "module",
@@ -15,6 +15,7 @@
15
15
  * keys survive.
16
16
  */
17
17
 
18
+ import { SCOPE_ENV } from "../context/scope.ts";
18
19
  import { copyFileSync, existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
19
20
  import { homedir } from "node:os";
20
21
  import { dirname, join } from "node:path";
@@ -59,16 +60,20 @@ export interface SpliceResult {
59
60
  */
60
61
  /**
61
62
  * Headers omp attaches to every request through this provider. Both are
62
- * optional: absent harness id ⇒ single-harness defaults, absent agentdox scope
63
- * the router falls back to its own `context.defaultScope`.
63
+ * optional: absent harness id ⇒ single-harness defaults; the agentdox scope
64
+ * names `SCOPE_ENV`, which omp resolves from its environment per request and
65
+ * the embed extension sets from the workspace folder, so the MAIN model's turns
66
+ * carry the repository's own scope rather than one for the whole machine. When
67
+ * the variable is unset the router ignores the literal and falls back to its
68
+ * own `context.defaultScope`.
64
69
  */
65
70
  function providerHeaders(cfg: RouterConfig): Record<string, string> {
66
71
  const headers: Record<string, string> = {};
67
72
  if (cfg.server.harnessId !== undefined && cfg.server.harnessId !== "") {
68
73
  headers["X-Omp-Harness"] = cfg.server.harnessId;
69
74
  }
70
- if (cfg.context.enabled && cfg.context.defaultScope !== "") {
71
- headers["X-Agentdox-Scope"] = cfg.context.defaultScope;
75
+ if (cfg.context.enabled) {
76
+ headers["X-Agentdox-Scope"] = SCOPE_ENV;
72
77
  }
73
78
  return headers;
74
79
  }
@@ -301,6 +301,7 @@ export const WIZARD_SECTIONS: readonly SectionSpec[] = [
301
301
  { path: "context.sessionLimit", label: "Session items", kind: "number", min: 0 },
302
302
  { path: "context.briefChars", label: "Brief chars", kind: "number", min: 0 },
303
303
  { path: "context.recordTurns", label: "Record turns back", kind: "boolean" },
304
+ { path: "context.injectWithoutTools", label: "Inject into tool-less calls", kind: "boolean" },
304
305
  { path: "context.maxQueue", label: "Write-back queue", kind: "number", min: 1 },
305
306
  ],
306
307
  },
@@ -14,7 +14,9 @@
14
14
  * $HERMES_HOME/plugins and .env points them at the remote
15
15
  * Codex ~/.codex/config.toml gains the auto-model-router provider
16
16
  * Aider ~/.aider.conf.yml gains the base URL, key and model
17
- * Claude Code ANTHROPIC_BASE_URL / ANTHROPIC_API_KEY (printed; --profile persists)
17
+ * Claude Code ~/.claude/settings.json gains the base URL (its `env` block) and
18
+ * `apiKeyHelper` running `auto-model-router token`, so no key
19
+ * sits in its environment or on disk for it
18
20
  *
19
21
  * Every write is idempotent and announced. `--profile` persists the
20
22
  * environment lines (shell rc on POSIX, user environment on Windows).
@@ -26,7 +28,9 @@ import { appendFileSync, cpSync, existsSync, mkdirSync, readFileSync, writeFileS
26
28
  import { homedir } from "node:os";
27
29
  import { dirname, join, resolve } from "node:path";
28
30
  import { fileURLToPath } from "node:url";
29
- import { remoteFilePath } from "../../omp-extension/remote-logic.ts";
31
+ import { refreshAccountOf, remoteFilePath } from "../../omp-extension/remote-logic.ts";
32
+ import { SCOPE_ENV } from "../context/scope.ts";
33
+ import { pickStore, saveRefreshToken, type StoreDeps, type StoreKind } from "./credential-store.ts";
30
34
  import { flagString, type CliArgs } from "./args.ts";
31
35
 
32
36
  export interface ConnectOptions {
@@ -44,10 +48,13 @@ export interface ConnectOptions {
44
48
  packageDir: string;
45
49
  /** Cost figures omp shows for the remote's virtual models, USD per million tokens. */
46
50
  blend?: { inputPerMtok: number; outputPerMtok: number };
47
- /** Adds `X-Agentdox-Scope` to omp's models.yml entry. Machine-wide: only for a single-project machine. Undefined keeps what the managed block already has. */
51
+ /** Pins `X-Agentdox-Scope` in omp's models.yml entry to one slug, machine-wide. Without it the entry follows the workspace (see renderRemoteModelsYml). Undefined keeps what the managed block already has. */
48
52
  agentdoxScope?: string;
49
53
  /** Short-lived credential fields from a remote that issues them; absent for a permanent key. */
50
54
  refreshToken?: string;
55
+ /** Which store takes the refresh token; picked from the platform when absent. Tests inject a backend. */
56
+ store?: StoreKind;
57
+ storeDeps?: StoreDeps;
51
58
  keyExpiresAtMs?: number;
52
59
  refreshExpiresAtMs?: number;
53
60
  device?: string;
@@ -135,10 +142,13 @@ const MODELS_YML_END = " # END auto-model-router (remote)";
135
142
  * neither problem — the URL and the key are stable — and the entry is what makes
136
143
  * omp's main model resolvable at startup, before extensions load.
137
144
  *
138
- * `scope` adds `X-Agentdox-Scope` to every request through this provider. It is
139
- * off by default on purpose: the file is machine-wide, so a scope here would
140
- * label turns from every workspace with one project. The extensions still send
141
- * the workspace's own scope on the roles that resolve after they load.
145
+ * `X-Agentdox-Scope` on this entry reaches the MAIN model's turns, which the
146
+ * extensions' own registration cannot (they load after the handle is built).
147
+ * The file is machine-wide, so a literal slug here would label every
148
+ * workspace's turns with one project; by default the value is the NAME of
149
+ * `SCOPE_ENV`, which omp resolves from its environment per request, and the
150
+ * embed extension sets that variable from the workspace folder as it loads.
151
+ * `scope` pins a literal slug instead, for a single-project machine.
142
152
  */
143
153
  export function renderRemoteModelsYml(url: string, key: string, blend: { inputPerMtok: number; outputPerMtok: number }, scope = ""): string {
144
154
  const round = (v: number): number => Math.round(v * 1e4) / 1e4;
@@ -156,7 +166,7 @@ export function renderRemoteModelsYml(url: string, key: string, blend: { inputPe
156
166
  " api: openai-completions",
157
167
  ` apiKey: ${key}`,
158
168
  ];
159
- if (scope !== "") lines.push(" headers:", ` X-Agentdox-Scope: ${scope}`);
169
+ lines.push(" headers:", ` X-Agentdox-Scope: ${scope !== "" ? scope : SCOPE_ENV}`);
160
170
  lines.push(" models:");
161
171
  for (const m of REMOTE_MODEL_ROWS) {
162
172
  lines.push(
@@ -195,14 +205,15 @@ export function mergeModelsYml(before: string, blockText: string): string {
195
205
  return `${body.replace(/\s*$/, "")}${eol}providers:${eol}${block}${eol}`;
196
206
  }
197
207
 
198
- /** The `X-Agentdox-Scope` the managed block carries, or "" when none. */
208
+ /** The literal `X-Agentdox-Scope` the managed block pins, or "" when it follows the workspace (or has none). */
199
209
  export function existingBlockScope(text: string): string {
200
210
  const begin = text.indexOf(MODELS_YML_BEGIN);
201
211
  if (begin < 0) return "";
202
212
  const end = text.indexOf(MODELS_YML_END, begin);
203
213
  const block = text.slice(begin, end < 0 ? text.length : end);
204
214
  const m = /X-Agentdox-Scope:\s*(\S+)/.exec(block);
205
- return m?.[1] ?? "";
215
+ const value = m?.[1] ?? "";
216
+ return value === SCOPE_ENV ? "" : value;
206
217
  }
207
218
 
208
219
  /** True when the file already defines our provider outside a block we manage. */
@@ -223,6 +234,16 @@ export function connectRemote(o: ConnectOptions): ConnectReport {
223
234
  const rh = routerHomeOf(o);
224
235
  report.remoteFile = remoteFilePath(rh);
225
236
  const previous = existsSync(report.remoteFile) ? (JSON.parse(readFileSync(report.remoteFile, "utf8")) as Record<string, unknown>) : {};
237
+ // The refresh token is the long-lived secret: it goes to the OS credential store, and
238
+ // remote.json only says which one. The access key stays in the file: it is short-lived,
239
+ // and the extensions need it without a subprocess on every poll.
240
+ let refreshTokenStore: StoreKind | undefined;
241
+ const refreshAccount = refreshAccountOf(o.url, o.userId);
242
+ if (o.refreshToken !== undefined && o.refreshToken !== "" && !o.dryRun) {
243
+ const wanted = o.store ?? pickStore(o.platform, o.pathHas);
244
+ refreshTokenStore = saveRefreshToken(rh, refreshAccount, o.refreshToken, wanted, o.storeDeps ?? { pathHas: o.pathHas });
245
+ if (refreshTokenStore !== wanted) report.notes.push(`the ${wanted} credential store was not usable; the refresh token is in ${join(rh, "refresh.token")} (owner-readable only)`);
246
+ } else if (o.refreshToken !== undefined && o.refreshToken !== "") refreshTokenStore = o.store ?? pickStore(o.platform, o.pathHas);
226
247
  write(
227
248
  report.remoteFile,
228
249
  `${JSON.stringify(
@@ -232,7 +253,7 @@ export function connectRemote(o: ConnectOptions): ConnectReport {
232
253
  userId: o.userId,
233
254
  name: o.name,
234
255
  joinedAtMs: typeof previous.joinedAtMs === "number" ? previous.joinedAtMs : Date.now(),
235
- ...(o.refreshToken !== undefined && o.refreshToken !== "" ? { refreshToken: o.refreshToken } : {}),
256
+ ...(refreshTokenStore !== undefined ? { refreshTokenStore, refreshAccount } : {}),
236
257
  ...(o.keyExpiresAtMs !== undefined ? { keyExpiresAtMs: o.keyExpiresAtMs } : {}),
237
258
  ...(o.refreshExpiresAtMs !== undefined ? { refreshExpiresAtMs: o.refreshExpiresAtMs } : {}),
238
259
  ...(o.device !== undefined && o.device !== "" ? { device: o.device } : {}),
@@ -300,11 +321,33 @@ export function connectRemote(o: ConnectOptions): ConnectReport {
300
321
  report.configured.push(`Aider (${aiderConf})`);
301
322
  } else report.skipped.push("Aider (not found)");
302
323
 
303
- // 6. Claude Code: environment only.
304
- if (wants(o, "claude") && o.pathHas("claude")) {
305
- report.envLines.push(`ANTHROPIC_BASE_URL=${o.url}`, `ANTHROPIC_API_KEY=${o.key}`);
306
- report.configured.push("Claude Code (environment)");
307
- } else report.skipped.push("Claude Code (not on PATH)");
324
+ // 6. Claude Code: its settings file carries the base URL (the `env` block) and a key
325
+ // helper, a command it runs for the key — so the key is never in its environment or
326
+ // on disk for it. Without a refresh token the helper still works (it prints the key it
327
+ // holds); the helper is what lets a short-lived key rotate underneath a running session.
328
+ const claudeDir = join(o.home, ".claude");
329
+ if (wants(o, "claude") && (o.pathHas("claude") || existsSync(claudeDir))) {
330
+ const settingsPath = join(claudeDir, "settings.json");
331
+ let settings: Record<string, unknown> = {};
332
+ const before = existsSync(settingsPath) ? readFileSync(settingsPath, "utf8") : "";
333
+ try {
334
+ settings = before === "" ? {} : (JSON.parse(before) as Record<string, unknown>);
335
+ } catch {
336
+ report.notes.push(`${settingsPath} is not valid JSON; left alone — set env.ANTHROPIC_BASE_URL and apiKeyHelper by hand`);
337
+ settings = {};
338
+ }
339
+ const env: Record<string, unknown> = { ...((settings.env as Record<string, unknown> | undefined) ?? {}), ANTHROPIC_BASE_URL: o.url };
340
+ // Never leave a stale key beside the helper: the helper is the source now.
341
+ delete env.ANTHROPIC_API_KEY;
342
+ const entry = resolve(o.packageDir, "src", "index.ts").replaceAll("\\", "/");
343
+ const next = { ...settings, env, apiKeyHelper: `bun run "${entry}" token` };
344
+ const after = `${JSON.stringify(next, null, 2)}\n`;
345
+ if (after !== before) {
346
+ if (before !== "" && !o.dryRun) writeFileSync(`${settingsPath}.${new Date().toISOString().replaceAll(":", "-")}.bak`, before, "utf8");
347
+ write(settingsPath, after);
348
+ }
349
+ report.configured.push(`Claude Code (${settingsPath}: env.ANTHROPIC_BASE_URL + apiKeyHelper; open a new session)`);
350
+ } else report.skipped.push("Claude Code (not on PATH and no ~/.claude)");
308
351
  report.envLines.unshift(`AUTO_MODEL_ROUTER_URL=${o.url}`, `AUTO_MODEL_ROUTER_API_KEY=${o.key}`);
309
352
  report.envLines = [...new Set(report.envLines)];
310
353
 
@@ -0,0 +1,157 @@
1
+ /**
2
+ * Where the refresh token lives on a member's machine.
3
+ *
4
+ * The access key has to sit in harness config files (a harness needs a literal
5
+ * bearer), and it is short-lived. The refresh token is the long-lived secret,
6
+ * so it goes to the operating system's credential store instead of a file:
7
+ *
8
+ * Windows DPAPI (CurrentUser scope): the token is encrypted so that only
9
+ * this Windows user on this machine can decrypt it, and the
10
+ * ciphertext is kept in `<router home>/refresh.dpapi`. Built in;
11
+ * no module to install. The plaintext passes to PowerShell through
12
+ * an environment variable, never an argument.
13
+ * macOS the login keychain, through `security` (service
14
+ * `auto-model-router`, one account per remote user).
15
+ * Linux the Secret Service through `secret-tool` when it is installed.
16
+ * file `<router home>/refresh.token`, owner-readable only — the fallback
17
+ * when none of the above works, and the choice on CI boxes.
18
+ *
19
+ * `remote.json` records which store holds it (`refreshTokenStore`) and the
20
+ * account name; it never holds the token itself once a store other than
21
+ * `file` is in use. A remote.json written before this existed may still carry
22
+ * the token inline; reading honours that until the next refresh moves it.
23
+ */
24
+
25
+ import { spawnSync } from "node:child_process";
26
+ import { chmodSync, existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
27
+ import { join } from "node:path";
28
+
29
+ export type StoreKind = "dpapi" | "keychain" | "secret-service" | "file";
30
+
31
+ const SERVICE = "auto-model-router";
32
+
33
+ /** The store this platform offers, given which tools are on PATH. */
34
+ export function pickStore(platform: string, pathHas: (bin: string) => boolean): StoreKind {
35
+ if (platform === "win32") return pathHas("powershell") || pathHas("pwsh") ? "dpapi" : "file";
36
+ if (platform === "darwin") return pathHas("security") ? "keychain" : "file";
37
+ if (platform === "linux") return pathHas("secret-tool") ? "secret-service" : "file";
38
+ return "file";
39
+ }
40
+
41
+ const powershell = (pathHas: (bin: string) => boolean): string => (pathHas("pwsh") ? "pwsh" : "powershell");
42
+
43
+ function dpapiProtect(secret: string, pathHas: (bin: string) => boolean): string {
44
+ const r = spawnSync(
45
+ powershell(pathHas),
46
+ ["-NoProfile", "-NonInteractive", "-Command", "Add-Type -AssemblyName System.Security; [Convert]::ToBase64String([System.Security.Cryptography.ProtectedData]::Protect([Text.Encoding]::UTF8.GetBytes($env:AMR_SECRET), $null, 'CurrentUser'))"],
47
+ { encoding: "utf8", env: { ...process.env, AMR_SECRET: secret } },
48
+ );
49
+ if (r.status !== 0 || r.stdout.trim() === "") throw new Error(`DPAPI protect failed: ${r.stderr.trim() || r.status}`);
50
+ return r.stdout.trim();
51
+ }
52
+
53
+ function dpapiUnprotect(blob: string, pathHas: (bin: string) => boolean): string {
54
+ const r = spawnSync(
55
+ powershell(pathHas),
56
+ ["-NoProfile", "-NonInteractive", "-Command", "Add-Type -AssemblyName System.Security; [Text.Encoding]::UTF8.GetString([System.Security.Cryptography.ProtectedData]::Unprotect([Convert]::FromBase64String($env:AMR_BLOB), $null, 'CurrentUser'))"],
57
+ { encoding: "utf8", env: { ...process.env, AMR_BLOB: blob } },
58
+ );
59
+ if (r.status !== 0) throw new Error(`DPAPI unprotect failed: ${r.stderr.trim() || r.status}`);
60
+ return r.stdout.replace(/\r?\n$/, "");
61
+ }
62
+
63
+ const filePath = (routerHome: string): string => join(routerHome, "refresh.token");
64
+ const dpapiPath = (routerHome: string): string => join(routerHome, "refresh.dpapi");
65
+
66
+ export interface StoreDeps {
67
+ pathHas?: (bin: string) => boolean;
68
+ /** Injected in tests to stand in for the platform tools. */
69
+ backend?: { save(account: string, secret: string): void; load(account: string): string | null; remove(account: string): void };
70
+ }
71
+
72
+ /**
73
+ * Saves the refresh token and returns the store that took it. A store that
74
+ * fails (keychain locked, tool missing) falls back to the file, so a member is
75
+ * never left without a refresh token; the caller records what was used.
76
+ */
77
+ export function saveRefreshToken(routerHome: string, account: string, secret: string, kind: StoreKind, deps: StoreDeps = {}): StoreKind {
78
+ const pathHas = deps.pathHas ?? ((bin) => Bun.which(bin) !== null);
79
+ mkdirSync(routerHome, { recursive: true });
80
+ try {
81
+ if (deps.backend !== undefined) {
82
+ deps.backend.save(account, secret);
83
+ return kind;
84
+ }
85
+ switch (kind) {
86
+ case "dpapi":
87
+ writeFileSync(dpapiPath(routerHome), `${dpapiProtect(secret, pathHas)}\n`, { encoding: "utf8", mode: 0o600 });
88
+ rmSync(filePath(routerHome), { force: true });
89
+ return "dpapi";
90
+ case "keychain": {
91
+ const r = spawnSync("security", ["add-generic-password", "-U", "-a", account, "-s", SERVICE, "-w", secret], { encoding: "utf8" });
92
+ if (r.status !== 0) throw new Error(r.stderr.trim());
93
+ rmSync(filePath(routerHome), { force: true });
94
+ return "keychain";
95
+ }
96
+ case "secret-service": {
97
+ const r = spawnSync("secret-tool", ["store", `--label=${SERVICE} ${account}`, "service", SERVICE, "account", account], { encoding: "utf8", input: secret });
98
+ if (r.status !== 0) throw new Error(r.stderr.trim());
99
+ rmSync(filePath(routerHome), { force: true });
100
+ return "secret-service";
101
+ }
102
+ case "file":
103
+ break;
104
+ }
105
+ } catch {
106
+ // fall through to the file
107
+ }
108
+ writeFileSync(filePath(routerHome), `${secret}\n`, { encoding: "utf8", mode: 0o600 });
109
+ try {
110
+ chmodSync(filePath(routerHome), 0o600);
111
+ } catch {
112
+ /* Windows */
113
+ }
114
+ return "file";
115
+ }
116
+
117
+ /** The refresh token from the store `remote.json` names, or null when it is gone. */
118
+ export function loadRefreshToken(routerHome: string, account: string, kind: StoreKind, deps: StoreDeps = {}): string | null {
119
+ const pathHas = deps.pathHas ?? ((bin) => Bun.which(bin) !== null);
120
+ try {
121
+ if (deps.backend !== undefined) return deps.backend.load(account);
122
+ switch (kind) {
123
+ case "dpapi": {
124
+ const p = dpapiPath(routerHome);
125
+ if (!existsSync(p)) return null;
126
+ return dpapiUnprotect(readFileSync(p, "utf8").trim(), pathHas);
127
+ }
128
+ case "keychain": {
129
+ const r = spawnSync("security", ["find-generic-password", "-a", account, "-s", SERVICE, "-w"], { encoding: "utf8" });
130
+ return r.status === 0 ? r.stdout.replace(/\r?\n$/, "") : null;
131
+ }
132
+ case "secret-service": {
133
+ const r = spawnSync("secret-tool", ["lookup", "service", SERVICE, "account", account], { encoding: "utf8" });
134
+ return r.status === 0 && r.stdout !== "" ? r.stdout.replace(/\r?\n$/, "") : null;
135
+ }
136
+ case "file": {
137
+ const p = filePath(routerHome);
138
+ return existsSync(p) ? readFileSync(p, "utf8").trim() : null;
139
+ }
140
+ }
141
+ } catch {
142
+ return null;
143
+ }
144
+ return null;
145
+ }
146
+
147
+ /** Forgets the token everywhere it might be. */
148
+ export function removeRefreshToken(routerHome: string, account: string, deps: StoreDeps = {}): void {
149
+ rmSync(filePath(routerHome), { force: true });
150
+ rmSync(dpapiPath(routerHome), { force: true });
151
+ if (deps.backend !== undefined) {
152
+ deps.backend.remove(account);
153
+ return;
154
+ }
155
+ if (process.platform === "darwin") spawnSync("security", ["delete-generic-password", "-a", account, "-s", SERVICE], { encoding: "utf8" });
156
+ if (process.platform === "linux") spawnSync("secret-tool", ["clear", "service", SERVICE, "account", account], { encoding: "utf8" });
157
+ }
@@ -16,7 +16,8 @@
16
16
  import { homedir } from "node:os";
17
17
  import { dirname, resolve } from "node:path";
18
18
  import { fileURLToPath } from "node:url";
19
- import { readRemoteRouter, type RemoteRouter } from "../../omp-extension/remote-logic.ts";
19
+ import { hasRefresh, readRemoteRouter, refreshAccountOf, type RemoteRouter } from "../../omp-extension/remote-logic.ts";
20
+ import { loadRefreshToken, type StoreDeps } from "./credential-store.ts";
20
21
  import { routerHome } from "../../omp-extension/router-url.ts";
21
22
  import type { CliArgs } from "./args.ts";
22
23
  import { connectRemote } from "./connect.ts";
@@ -34,7 +35,7 @@ export interface RefreshedCredential {
34
35
 
35
36
  /** True when the credential can and should be traded now: it has a refresh token and its key is near or past expiry. */
36
37
  export function shouldRefresh(remote: RemoteRouter, nowMs = Date.now()): boolean {
37
- if (remote.refreshToken === undefined || remote.refreshToken === "") return false;
38
+ if (!hasRefresh(remote)) return false;
38
39
  if (remote.keyExpiresAtMs === undefined) return false;
39
40
  return remote.keyExpiresAtMs - nowMs <= REFRESH_AHEAD_MS;
40
41
  }
@@ -49,13 +50,21 @@ export class RefreshError extends Error {
49
50
  }
50
51
  }
51
52
 
53
+ /** The refresh token: inline from an older remote.json, else from the OS store remote.json names. */
54
+ export function resolveRefreshToken(remote: RemoteRouter, routerHome: string, storeDeps: StoreDeps = {}): string | null {
55
+ if (remote.refreshToken !== undefined && remote.refreshToken !== "") return remote.refreshToken;
56
+ if (remote.refreshTokenStore === undefined) return null;
57
+ return loadRefreshToken(routerHome, remote.refreshAccount ?? refreshAccountOf(remote.url, remote.userId), remote.refreshTokenStore, storeDeps);
58
+ }
59
+
52
60
  /** Trades the refresh token at the remote for the next credential. */
53
- export async function refreshCredential(remote: RemoteRouter, fetchImpl: typeof fetch = fetch): Promise<RefreshedCredential> {
54
- if (remote.refreshToken === undefined || remote.refreshToken === "") throw new RefreshError("no_refresh_token", "this machine holds no refresh token; onboard it again with a setup token");
61
+ export async function refreshCredential(remote: RemoteRouter, fetchImpl: typeof fetch = fetch, routerHomeDir: string = routerHome(), storeDeps: StoreDeps = {}): Promise<RefreshedCredential> {
62
+ const token = resolveRefreshToken(remote, routerHomeDir, storeDeps);
63
+ if (token === null || token === "") throw new RefreshError("no_refresh_token", "this machine holds no refresh token (or its credential store no longer has it); onboard it again with a setup token");
55
64
  const res = await fetchImpl(`${remote.url}/auth/refresh`, {
56
65
  method: "POST",
57
66
  headers: { "content-type": "application/json" },
58
- body: JSON.stringify({ refreshToken: remote.refreshToken }),
67
+ body: JSON.stringify({ refreshToken: token }),
59
68
  signal: AbortSignal.timeout(15_000),
60
69
  });
61
70
  const body = (await res.json().catch(() => null)) as { key?: unknown; keyExpiresAtMs?: unknown; refreshToken?: unknown; refreshExpiresAtMs?: unknown; device?: unknown; error?: { code?: string; message?: string } } | null;
@@ -76,8 +85,9 @@ export async function refreshCredential(remote: RemoteRouter, fetchImpl: typeof
76
85
  * harness configs `connect` manages. Returns the fresh credential. `home` and
77
86
  * `packageDir` are injectable for tests.
78
87
  */
79
- export async function refreshAndRewrite(opts: { remote: RemoteRouter; fetchImpl?: typeof fetch; home?: string; packageDir?: string; env?: Record<string, string | undefined>; platform?: string; pathHas?: (bin: string) => boolean }): Promise<RefreshedCredential> {
80
- const fresh = await refreshCredential(opts.remote, opts.fetchImpl ?? fetch);
88
+ export async function refreshAndRewrite(opts: { remote: RemoteRouter; fetchImpl?: typeof fetch; home?: string; packageDir?: string; env?: Record<string, string | undefined>; platform?: string; pathHas?: (bin: string) => boolean; routerHome?: string; storeDeps?: StoreDeps }): Promise<RefreshedCredential> {
89
+ const rh = opts.routerHome ?? routerHome();
90
+ const fresh = await refreshCredential(opts.remote, opts.fetchImpl ?? fetch, rh, opts.storeDeps ?? {});
81
91
  const home = opts.home ?? (process.env.HOME !== undefined && process.env.HOME !== "" ? process.env.HOME : homedir());
82
92
  const packageDir = opts.packageDir ?? resolve(dirname(fileURLToPath(import.meta.url)), "..", "..");
83
93
  connectRemote({
@@ -97,6 +107,9 @@ export async function refreshAndRewrite(opts: { remote: RemoteRouter; fetchImpl?
97
107
  packageDir,
98
108
  platform: opts.platform ?? process.platform,
99
109
  pathHas: opts.pathHas ?? ((bin) => Bun.which(bin) !== null),
110
+ // The store that already holds it keeps it; a machine never silently changes store.
111
+ ...(opts.remote.refreshTokenStore !== undefined ? { store: opts.remote.refreshTokenStore } : {}),
112
+ ...(opts.storeDeps !== undefined ? { storeDeps: opts.storeDeps } : {}),
100
113
  // undefined keeps whatever scope the managed models.yml block already carries.
101
114
  });
102
115
  return fresh;
@@ -263,6 +263,14 @@ export const DEFAULT_CONFIG: RouterConfig = {
263
263
  // the query-relevant memory/docs tail. 0 omits the brief.
264
264
  briefChars: 12_000,
265
265
  recordTurns: true,
266
+ // A request with no tool schemas is a harness utility call — omp asks for
267
+ // a session title or a complexity rating through the same provider — and
268
+ // it answers ABOUT the conversation, so the project block cannot help it.
269
+ // Recording already skips those calls (turn.ts); injection did not, and
270
+ // one omp turn measured seven side calls at ~6k tokens of context each,
271
+ // 42k prompt tokens for nothing. Off: the block goes only to turns that
272
+ // ship tools. On: every scoped turn gets it, as before 0.10.
273
+ injectWithoutTools: false,
266
274
  maxQueue: 64,
267
275
  },
268
276
  compaction: {
@@ -192,6 +192,7 @@ const context = z.strictObject({
192
192
  sessionLimit: z.number().int().nonnegative().optional(),
193
193
  briefChars: z.number().int().nonnegative().optional(),
194
194
  recordTurns: z.boolean().optional(),
195
+ injectWithoutTools: z.boolean().optional(),
195
196
  maxQueue: z.number().int().positive().optional(),
196
197
  });
197
198
 
@@ -769,6 +769,13 @@ export interface ContextConfig {
769
769
  briefChars: number;
770
770
  /** Write settled turns back to agentdox sessions, tagged with the served model. */
771
771
  recordTurns: boolean;
772
+ /**
773
+ * Inject the block into turns that carry NO tool schemas too. Those are
774
+ * harness utility calls (titles, ratings), which recording already skips;
775
+ * default off, because each one paid the whole block for an answer that is
776
+ * about the conversation, not part of it.
777
+ */
778
+ injectWithoutTools: boolean;
772
779
  /** Bound on queued write-backs; excess turns are dropped, never buffered unbounded. */
773
780
  maxQueue: number;
774
781
  }
@@ -0,0 +1,29 @@
1
+ /**
2
+ * The agentdox scope a request names, and where omp's main model gets it from.
3
+ *
4
+ * omp resolves its MAIN model from `models.yml` at startup, before any extension
5
+ * loads, so the `X-Agentdox-Scope` header on that provider entry cannot be set
6
+ * by the extension per workspace — but omp resolves a header VALUE that names
7
+ * an environment variable from the environment on every request, and the
8
+ * extension runs inside omp's process. So `connect` (and the local sync) write
9
+ * the header's value as the NAME below, and the extension sets that variable
10
+ * from the workspace folder when it loads. One machine-wide file, one scope
11
+ * per repository.
12
+ */
13
+ export const SCOPE_ENV = "AUTO_MODEL_ROUTER_SCOPE";
14
+
15
+ /**
16
+ * A scope is an agentdox project slug: lowercase, starting alphanumeric. The
17
+ * shape matters because omp sends the header's literal value when the variable
18
+ * it names is unset — `AUTO_MODEL_ROUTER_SCOPE` itself — and that must never
19
+ * become a project. Uppercase never passes, so the sentinel cannot.
20
+ */
21
+ export function isScopeSlug(value: string): boolean {
22
+ return /^[a-z0-9][a-z0-9._-]{0,127}$/.test(value);
23
+ }
24
+
25
+ /** The scope to trust from a request header: a slug, or "" for anything else. */
26
+ export function acceptScope(raw: string | null | undefined): string {
27
+ const s = (raw ?? "").trim();
28
+ return isScopeSlug(s) ? s : "";
29
+ }
package/src/index.ts CHANGED
@@ -27,7 +27,7 @@ Usage: auto-model-router <command> [options]
27
27
  stats Show routed spend, per-model share, and escalation rates
28
28
  report Usage analytics: providers, models, tiers, cost, speed, cache hit rate
29
29
  export One row per day, harness and model as CSV (--json for rows)
30
- connect Point this machine at a remote router (--url, --key[, --refresh-token]; --scope labels a single-project machine; --profile persists the environment)
30
+ connect Point this machine at a remote router (--url, --key[, --refresh-token]; --scope pins one project for the whole machine (default: each workspace's own); --profile persists the environment)
31
31
  refresh Trade the refresh token for a new access key and re-write every harness config (--force: even when not near expiry)
32
32
  token Print an access key that is good right now, refreshing first if needed (for a harness key-helper)
33
33
  models Show what each complexity tier would consider, and why
@@ -136,6 +136,12 @@ export async function runTurn(
136
136
  // Request header wins; the configured default covers harnesses that send none.
137
137
  const doxScope = req.agentdoxScope !== "" ? req.agentdoxScope : config.context.defaultScope;
138
138
  const doxActive = bridge.enabled && doxScope !== "";
139
+ // Injection shares recording's discriminator (explained at the record call
140
+ // below): a tool-less harness utility call answers ABOUT the conversation
141
+ // and gains nothing from the project block, yet paid its full ~6k tokens on
142
+ // every title and rating — 42k prompt tokens across one omp turn's seven
143
+ // side calls. `context.injectWithoutTools` restores the old behaviour.
144
+ const doxInject = doxActive && (req.tools.length > 0 || config.context.injectWithoutTools);
139
145
 
140
146
  // The trigger list is the source of truth for enabled signals, except
141
147
  // length_stop, which rides on its own toggle (escalation.escalateOnLengthStop).
@@ -209,7 +215,7 @@ export async function runTurn(
209
215
  // turn's prefix is already cold — a model switch or a retry — so the
210
216
  // injected bytes stay identical while the cache is worth keeping.
211
217
  let contextBlock: string | undefined;
212
- if (doxActive) {
218
+ if (doxInject) {
213
219
  const pin = await bridge.resolve({
214
220
  scope: doxScope,
215
221
  conversationKey: req.conversationKey,
@@ -233,6 +239,7 @@ export async function runTurn(
233
239
  log.debug("agentdox context", {
234
240
  active: doxActive,
235
241
  scope: doxScope === "" ? "(none)" : doxScope,
242
+ utilityCall: doxActive && !doxInject,
236
243
  injected: contextBlock !== undefined,
237
244
  chars: contextBlock?.length ?? 0,
238
245
  });
@@ -1,3 +1,4 @@
1
+ import { acceptScope } from "../../context/scope.ts";
1
2
  import type {
2
3
  RequestPolicy,
3
4
  CompactionEdit,
@@ -329,8 +330,10 @@ export function parseChatRequest(body: unknown, headers: Headers): NormRequest {
329
330
  const ompSessionId = (headers.get("x-omp-session") ?? "").trim();
330
331
 
331
332
  // agentdox project scope. Selects whose shared context is injected; absent
332
- // the server falls back to its configured default scope.
333
- const agentdoxScope = (headers.get("x-agentdox-scope") ?? "").trim();
333
+ // or not a slug (omp sends the literal env-var NAME when the variable that
334
+ // models.yml names is unset — see src/context/scope.ts) the server falls
335
+ // back to its configured default scope.
336
+ const agentdoxScope = acceptScope(headers.get("x-agentdox-scope"));
334
337
 
335
338
  // Subagent marker from the embed extension (sessions without a UI).
336
339
  const isSubagent = (headers.get("x-omp-subagent") ?? "").trim() === "1";
@@ -73,7 +73,7 @@ function mkConfig(escalation: Partial<EscalationConfig> = {}): RouterConfig {
73
73
  hysteresis: { holdTurns: 2, holdTurnsAfterEscalation: 4, switchMargin: 1.5, cacheWarmTtlMs: 600_000, maxDowngradePerTurn: 1, breakHoldOnMechanical: false, switchHorizonTurns: 1, confirmUpgradesBelowConfidence: 0.6 },
74
74
  exploration: { enabled: false, rates: {}, stickyPolicy: "never", holdTurns: { enabled: false, values: [2, 3, 4] } },
75
75
  cache: { injectBreakpoints: true, maxBreakpoints: 4, minPromptTokens: 1024, milestoneTokens: 20_000 },
76
- context: { enabled: false, baseUrl: "", token: "", defaultScope: "", timeoutMs: 3_000, maxStalenessMs: 900_000, maxBlockChars: 24_000, memoryLimit: 8, docsLimit: 2, sessionLimit: 6, briefChars: 0, recordTurns: false, maxQueue: 64 },
76
+ context: { enabled: false, baseUrl: "", token: "", defaultScope: "", timeoutMs: 3_000, maxStalenessMs: 900_000, maxBlockChars: 24_000, memoryLimit: 8, docsLimit: 2, sessionLimit: 6, briefChars: 0, recordTurns: false, injectWithoutTools: false, maxQueue: 64 },
77
77
  compaction: { enabled: false, budgetTokens: 40_000, floorRatio: 1, fitToWindow: true, protectRecentTurns: 4, maxToolResultBytes: 4_096, keepHeadBytes: 512, keepTailBytes: 512, elideSupersededReads: true, collapseDuplicateResults: true, replanGrowthRatio: 1, digestToolResults: false, digestMaxPerTurn: 2 },
78
78
  budget: { onExceeded: "downgrade" },
79
79
  report: { baselines: [], dailySummary: false },
@@ -1,12 +1,15 @@
1
1
  import { describe, expect, test } from "bun:test";
2
- import { existsSync, mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
2
+ import { existsSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, rmSync, writeFileSync } from "node:fs";
3
3
  import { tmpdir } from "node:os";
4
4
  import { join } from "node:path";
5
5
 
6
6
  import { addExtensions, codexBlock, connectRemote, setDotenv, type ConnectOptions } from "../src/cli/connect.ts";
7
7
  import { parseRemoteRouter, readRemoteRouter, remoteProviderRegistration } from "../omp-extension/remote-logic.ts";
8
8
  import { existingBlockScope, hasForeignRouterProvider, mergeModelsYml, renderRemoteModelsYml } from "../src/cli/connect.ts";
9
- import { refreshAndRewrite, refreshCredential, RefreshError, shouldRefresh } from "../src/cli/refresh.ts";
9
+ import { SCOPE_ENV } from "../src/context/scope.ts";
10
+ import { refreshAndRewrite, refreshCredential, RefreshError, resolveRefreshToken, shouldRefresh } from "../src/cli/refresh.ts";
11
+ import { loadRefreshToken, pickStore, removeRefreshToken, saveRefreshToken } from "../src/cli/credential-store.ts";
12
+ import { hasRefresh, refreshAccountOf } from "../omp-extension/remote-logic.ts";
10
13
 
11
14
  /**
12
15
  * Remote mode: remote.json puts the omp extensions on a router elsewhere,
@@ -78,7 +81,8 @@ describe("connect", () => {
78
81
  expect(readFileSync(join(home, ".codex", "config.toml"), "utf8")).toContain("[model_providers.auto-model-router]");
79
82
  expect(readFileSync(join(home, ".aider.conf.yml"), "utf8")).toContain("openai-api-base: https://team.example/v1");
80
83
  expect(r1.configured.join("\n")).toMatch(/omp[\s\S]*Hermes[\s\S]*Codex[\s\S]*Aider[\s\S]*Claude Code/);
81
- expect(r1.envLines).toEqual(["AUTO_MODEL_ROUTER_URL=https://team.example", "AUTO_MODEL_ROUTER_API_KEY=amrt_key", "ANTHROPIC_BASE_URL=https://team.example", "ANTHROPIC_API_KEY=amrt_key"]);
84
+ // Claude Code is configured through its settings file now, so nothing ANTHROPIC_* rides in the environment.
85
+ expect(r1.envLines).toEqual(["AUTO_MODEL_ROUTER_URL=https://team.example", "AUTO_MODEL_ROUTER_API_KEY=amrt_key"]);
82
86
  // Running again changes nothing.
83
87
  const snapshot = [ompCfg, readFileSync(join(home, ".codex", "config.toml"), "utf8"), readFileSync(join(home, ".aider.conf.yml"), "utf8")];
84
88
  connectRemote(o);
@@ -99,7 +103,8 @@ describe("connect", () => {
99
103
  connectRemote(o2);
100
104
  const rc = readFileSync(join(h2, ".zshrc"), "utf8");
101
105
  expect(rc.split("# auto-model-router remote").length).toBe(2);
102
- expect(rc).toContain("export ANTHROPIC_BASE_URL=https://team.example");
106
+ expect(rc).toContain("export AUTO_MODEL_ROUTER_URL=https://team.example");
107
+ expect(rc).not.toContain("ANTHROPIC_");
103
108
  rmSync(home, { recursive: true, force: true });
104
109
  rmSync(h2, { recursive: true, force: true });
105
110
  });
@@ -110,7 +115,7 @@ describe("omp models.yml for a remote router", () => {
110
115
  const NL = String.fromCharCode(10);
111
116
  const yaml = (...lines: string[]): string => lines.join(NL) + NL;
112
117
 
113
- test("the block names the remote, the key and the three virtual models; a scope is opt-in", () => {
118
+ test("the block names the remote, the key and the three virtual models; the scope follows the workspace unless pinned", () => {
114
119
  const block = renderRemoteModelsYml("https://team.example/", "amrt_k", BLEND);
115
120
  expect(block).toContain("baseUrl: https://team.example/v1");
116
121
  expect(block).toContain("apiKey: amrt_k");
@@ -118,9 +123,15 @@ describe("omp models.yml for a remote router", () => {
118
123
  expect(block).toContain("- id: auto-cheap");
119
124
  expect(block).toContain("- id: auto-max");
120
125
  expect(block).toContain("cost: { input: 1.1, output: 4.4, cacheRead: 0.11, cacheWrite: 1.375 }");
121
- // Machine-wide file: no scope unless the caller asks for one.
122
- expect(block).not.toContain("X-Agentdox-Scope");
123
- expect(renderRemoteModelsYml("https://team.example", "k", BLEND, "omp-router")).toContain("X-Agentdox-Scope: omp-router");
126
+ // Machine-wide file: the header names the env var the extension sets per
127
+ // workspace, so the MAIN model's turns carry each repo's own scope.
128
+ expect(block).toContain(`X-Agentdox-Scope: ${SCOPE_ENV}`);
129
+ expect(existingBlockScope(mergeModelsYml("", block))).toBe("");
130
+ // --scope pins one slug for the whole machine.
131
+ const pinned = renderRemoteModelsYml("https://team.example", "k", BLEND, "omp-router");
132
+ expect(pinned).toContain("X-Agentdox-Scope: omp-router");
133
+ expect(pinned).not.toContain(SCOPE_ENV);
134
+ expect(existingBlockScope(mergeModelsYml("", pinned))).toBe("omp-router");
124
135
  });
125
136
 
126
137
  test("merging keeps other providers, replaces our own block, and is idempotent", () => {
@@ -203,16 +214,21 @@ describe("short-lived remote credentials", () => {
203
214
  const { connectRemote } = await import("../src/cli/connect.ts");
204
215
  connectRemote({ url: "https://team.example", key: "amrt_old", userId: "u_ada", name: "Ada", refreshToken: "amrr_r1", keyExpiresAtMs: 1, refreshExpiresAtMs: 2, device: "laptop", agentdoxScope: "omp-router", profile: false, dryRun: false, only: ["omp"], env, home, packageDir: process.cwd(), platform: "linux", pathHas: () => false });
205
216
  const before = JSON.parse(readFileSync(join(routerHome, "remote.json"), "utf8")) as Record<string, unknown>;
206
- expect(before).toMatchObject({ key: "amrt_old", refreshToken: "amrr_r1", keyExpiresAtMs: 1, device: "laptop" });
217
+ // The refresh token is in the store (the file here: no platform tool on PATH); remote.json only names it.
218
+ expect(before).toMatchObject({ key: "amrt_old", keyExpiresAtMs: 1, device: "laptop", refreshTokenStore: "file", refreshAccount: "u_ada@team.example" });
219
+ expect(before.refreshToken).toBeUndefined();
220
+ expect(readFileSync(join(routerHome, "refresh.token"), "utf8").trim()).toBe("amrr_r1");
207
221
  const models0 = readFileSync(join(agent, "models.yml"), "utf8");
208
222
  expect(models0).toContain("apiKey: amrt_old");
209
223
  expect(existingBlockScope(models0)).toBe("omp-router");
210
224
  // Then a refresh, which knows nothing about the scope.
211
225
  const fetchImpl = (async () => Response.json({ key: "amrt_new", keyExpiresAtMs: 50, refreshToken: "amrr_r2", refreshExpiresAtMs: 90 })) as unknown as typeof fetch;
212
- const fresh = await refreshAndRewrite({ remote: parseRemoteRouter(readFileSync(join(routerHome, "remote.json"), "utf8"))!, fetchImpl, home, packageDir: process.cwd(), env, platform: "linux", pathHas: () => false });
226
+ const fresh = await refreshAndRewrite({ remote: parseRemoteRouter(readFileSync(join(routerHome, "remote.json"), "utf8"))!, fetchImpl, home, packageDir: process.cwd(), env, platform: "linux", pathHas: () => false, routerHome });
213
227
  expect(fresh.key).toBe("amrt_new");
214
228
  const after = JSON.parse(readFileSync(join(routerHome, "remote.json"), "utf8")) as Record<string, unknown>;
215
- expect(after).toMatchObject({ key: "amrt_new", refreshToken: "amrr_r2", keyExpiresAtMs: 50, refreshExpiresAtMs: 90, device: "laptop", joinedAtMs: before.joinedAtMs });
229
+ expect(after).toMatchObject({ key: "amrt_new", keyExpiresAtMs: 50, refreshExpiresAtMs: 90, device: "laptop", joinedAtMs: before.joinedAtMs, refreshTokenStore: "file" });
230
+ expect(after.refreshToken).toBeUndefined();
231
+ expect(readFileSync(join(routerHome, "refresh.token"), "utf8").trim()).toBe("amrr_r2");
216
232
  const models1 = readFileSync(join(agent, "models.yml"), "utf8");
217
233
  expect(models1).toContain("apiKey: amrt_new");
218
234
  expect(models1).not.toContain("amrt_old");
@@ -222,3 +238,91 @@ describe("short-lived remote credentials", () => {
222
238
  }
223
239
  });
224
240
  });
241
+
242
+ describe("the refresh token lives in the OS credential store", () => {
243
+ const NL = String.fromCharCode(10);
244
+ // A fake backend stands in for DPAPI / the keychain / secret-service.
245
+ const vault = new Map<string, string>();
246
+ const backend = { save: (a: string, s: string) => void vault.set(a, s), load: (a: string) => vault.get(a) ?? null, remove: (a: string) => void vault.delete(a) };
247
+
248
+ test("the store is picked from the platform and its tools; the file is the fallback everywhere", () => {
249
+ expect(pickStore("win32", (b) => b === "powershell")).toBe("dpapi");
250
+ expect(pickStore("win32", () => false)).toBe("file");
251
+ expect(pickStore("darwin", (b) => b === "security")).toBe("keychain");
252
+ expect(pickStore("linux", (b) => b === "secret-tool")).toBe("secret-service");
253
+ expect(pickStore("linux", () => false)).toBe("file");
254
+ expect(refreshAccountOf("https://team.example:8790/", "u_ada")).toBe("u_ada@team.example:8790");
255
+ });
256
+
257
+ test("save/load through a store, and the file fallback keeps the token owner-readable", () => {
258
+ const home = mkdtempSync(join(tmpdir(), "amr-store-"));
259
+ try {
260
+ expect(saveRefreshToken(home, "u@t", "amrr_x", "keychain", { backend })).toBe("keychain");
261
+ expect(loadRefreshToken(home, "u@t", "keychain", { backend })).toBe("amrr_x");
262
+ expect(existsSync(join(home, "refresh.token"))).toBe(false); // nothing on disk
263
+ expect(saveRefreshToken(home, "u@t", "amrr_f", "file")).toBe("file");
264
+ expect(loadRefreshToken(home, "u@t", "file")).toBe("amrr_f");
265
+ removeRefreshToken(home, "u@t", { backend });
266
+ expect(loadRefreshToken(home, "u@t", "keychain", { backend })).toBeNull();
267
+ expect(existsSync(join(home, "refresh.token"))).toBe(false);
268
+ } finally {
269
+ rmSync(home, { recursive: true, force: true });
270
+ }
271
+ });
272
+
273
+ test("connect files the token in the store and remote.json only names it; refresh reads it back; an older inline token still works", async () => {
274
+ const home = mkdtempSync(join(tmpdir(), "amr-store2-"));
275
+ const agent = join(home, ".omp", "agent");
276
+ mkdirSync(agent, { recursive: true });
277
+ writeFileSync(join(agent, "config.yml"), "extensions: []" + NL);
278
+ const rh = join(home, ".auto-model-router");
279
+ const env = { HOME: home, PI_CODING_AGENT_DIR: agent, AUTO_MODEL_ROUTER_HOME: rh, HERMES_HOME: join(home, "no-hermes") };
280
+ try {
281
+ const { connectRemote } = await import("../src/cli/connect.ts");
282
+ connectRemote({ url: "https://team.example", key: "amrt_k1", userId: "u_ada", name: "Ada", refreshToken: "amrr_r1", keyExpiresAtMs: 1, refreshExpiresAtMs: 2, device: "laptop", store: "keychain", storeDeps: { backend }, profile: false, dryRun: false, only: ["omp"], env, home, packageDir: process.cwd(), platform: "darwin", pathHas: () => false });
283
+ const written = readFileSync(join(rh, "remote.json"), "utf8");
284
+ expect(written).not.toContain("amrr_r1");
285
+ const remote = parseRemoteRouter(written)!;
286
+ expect(remote).toMatchObject({ refreshTokenStore: "keychain", refreshAccount: "u_ada@team.example" });
287
+ expect(remote.refreshToken).toBeUndefined();
288
+ expect(hasRefresh(remote)).toBe(true);
289
+ expect(resolveRefreshToken(remote, rh, { backend })).toBe("amrr_r1");
290
+ // The refresh trades the stored token and files the new one in the same store.
291
+ const seen: string[] = [];
292
+ const fetchImpl = (async (_u: string | URL | Request, init?: RequestInit) => {
293
+ seen.push(String(init?.body));
294
+ return Response.json({ key: "amrt_k2", keyExpiresAtMs: 50, refreshToken: "amrr_r2", refreshExpiresAtMs: 90 });
295
+ }) as unknown as typeof fetch;
296
+ await refreshAndRewrite({ remote, fetchImpl, home, packageDir: process.cwd(), env, platform: "darwin", pathHas: () => false, routerHome: rh, storeDeps: { backend } });
297
+ expect(seen[0]).toBe(JSON.stringify({ refreshToken: "amrr_r1" }));
298
+ expect(vault.get("u_ada@team.example")).toBe("amrr_r2");
299
+ expect(readFileSync(join(rh, "remote.json"), "utf8")).not.toContain("amrr_r2");
300
+ // An older remote.json with the token inline is honoured until its next refresh.
301
+ const legacy = parseRemoteRouter(JSON.stringify({ url: "https://t", key: "k", refreshToken: "amrr_inline" }))!;
302
+ expect(resolveRefreshToken(legacy, rh)).toBe("amrr_inline");
303
+ } finally {
304
+ rmSync(home, { recursive: true, force: true });
305
+ }
306
+ });
307
+
308
+ test("connect writes Claude Code's settings: base URL in env, a key helper instead of a key, other settings kept", async () => {
309
+ const home = mkdtempSync(join(tmpdir(), "amr-claude-"));
310
+ const claude = join(home, ".claude");
311
+ mkdirSync(claude, { recursive: true });
312
+ writeFileSync(join(claude, "settings.json"), JSON.stringify({ theme: "dark", env: { ANTHROPIC_API_KEY: "sk-old", FOO: "bar" } }, null, 2) + NL);
313
+ const env = { HOME: home, PI_CODING_AGENT_DIR: join(home, "no-omp"), AUTO_MODEL_ROUTER_HOME: join(home, ".auto-model-router"), HERMES_HOME: join(home, "no-hermes") };
314
+ try {
315
+ const { connectRemote } = await import("../src/cli/connect.ts");
316
+ const r = connectRemote({ url: "https://team.example", key: "amrt_k", userId: "u_ada", name: "Ada", profile: false, dryRun: false, only: ["claude"], env, home, packageDir: "/pkg", platform: "linux", pathHas: () => false });
317
+ expect(r.configured.some((c) => c.startsWith("Claude Code ("))).toBe(true);
318
+ const s = JSON.parse(readFileSync(join(claude, "settings.json"), "utf8")) as { theme: string; env: Record<string, string>; apiKeyHelper: string };
319
+ expect(s.theme).toBe("dark");
320
+ expect(s.env).toEqual({ FOO: "bar", ANTHROPIC_BASE_URL: "https://team.example" }); // the stale key is gone
321
+ expect(s.apiKeyHelper).toMatch(/^bun run ".*\/pkg\/src\/index\.ts" token$/); // an absolute path, drive letter and all on Windows
322
+ expect(r.envLines.some((l) => l.startsWith("ANTHROPIC_API_KEY="))).toBe(false);
323
+ expect(readdirSync(claude).some((f) => f.startsWith("settings.json.") && f.endsWith(".bak"))).toBe(true); // the previous file was kept
324
+ } finally {
325
+ rmSync(home, { recursive: true, force: true });
326
+ }
327
+ });
328
+ });
@@ -0,0 +1,31 @@
1
+ import { describe, expect, test } from "bun:test";
2
+ import { SCOPE_ENV, acceptScope, isScopeSlug } from "../src/context/scope.ts";
3
+ import { parseChatRequest } from "../src/wire/openai/request.ts";
4
+
5
+ describe("the scope a request may name", () => {
6
+ test("a slug passes; the env-var sentinel and other non-slugs do not", () => {
7
+ expect(isScopeSlug("omp-router")).toBe(true);
8
+ expect(isScopeSlug("ashlands")).toBe(true);
9
+ expect(isScopeSlug("my.app_v2")).toBe(true);
10
+ // omp sends the header's literal value when the variable it names is
11
+ // unset: the NAME must never become a project.
12
+ expect(isScopeSlug(SCOPE_ENV)).toBe(false);
13
+ expect(isScopeSlug("")).toBe(false);
14
+ expect(isScopeSlug("-leading")).toBe(false);
15
+ expect(isScopeSlug("Has Spaces")).toBe(false);
16
+ expect(isScopeSlug("a".repeat(129))).toBe(false);
17
+ });
18
+
19
+ test("acceptScope trims and rejects", () => {
20
+ expect(acceptScope(" omp-router ")).toBe("omp-router");
21
+ expect(acceptScope(SCOPE_ENV)).toBe("");
22
+ expect(acceptScope(null)).toBe("");
23
+ });
24
+
25
+ test("the wire parser drops a non-slug X-Agentdox-Scope", () => {
26
+ const body = { model: "auto", messages: [{ role: "user", content: "hi" }] };
27
+ const parse = (scope: string) => parseChatRequest(body, new Headers({ "x-agentdox-scope": scope }));
28
+ expect(parse("omp-router").agentdoxScope).toBe("omp-router");
29
+ expect(parse(SCOPE_ENV).agentdoxScope).toBe("");
30
+ });
31
+ });
package/test/turn.test.ts CHANGED
@@ -73,7 +73,7 @@ function mkConfig(escalation: Partial<EscalationConfig> = {}): RouterConfig {
73
73
  hysteresis: { holdTurns: 2, holdTurnsAfterEscalation: 4, switchMargin: 1.5, cacheWarmTtlMs: 600_000, maxDowngradePerTurn: 1, breakHoldOnMechanical: false, switchHorizonTurns: 1, confirmUpgradesBelowConfidence: 0.6 },
74
74
  exploration: { enabled: false, rates: {}, stickyPolicy: "never", holdTurns: { enabled: false, values: [2, 3, 4] } },
75
75
  cache: { injectBreakpoints: true, maxBreakpoints: 4, minPromptTokens: 1024, milestoneTokens: 20_000 },
76
- context: { enabled: false, baseUrl: "", token: "", defaultScope: "", timeoutMs: 3_000, maxStalenessMs: 900_000, maxBlockChars: 24_000, memoryLimit: 8, docsLimit: 2, sessionLimit: 6, briefChars: 0, recordTurns: false, maxQueue: 64 },
76
+ context: { enabled: false, baseUrl: "", token: "", defaultScope: "", timeoutMs: 3_000, maxStalenessMs: 900_000, maxBlockChars: 24_000, memoryLimit: 8, docsLimit: 2, sessionLimit: 6, briefChars: 0, recordTurns: false, injectWithoutTools: false, maxQueue: 64 },
77
77
  compaction: { enabled: false, budgetTokens: 40_000, floorRatio: 1, fitToWindow: true, protectRecentTurns: 4, maxToolResultBytes: 4_096, keepHeadBytes: 512, keepTailBytes: 512, elideSupersededReads: true, collapseDuplicateResults: true, replanGrowthRatio: 1, digestToolResults: false, digestMaxPerTurn: 2 },
78
78
  budget: { onExceeded: "downgrade" },
79
79
  report: { baselines: [], dailySummary: false },
@@ -660,6 +660,81 @@ describe("agentdox write-back sees the shape of the turn", () => {
660
660
  });
661
661
  });
662
662
 
663
+ describe("agentdox injection sees the shape of the turn", () => {
664
+ const AGENT_TOOL = { name: "read", description: "read a file", schemaBytes: 128 };
665
+
666
+ /** A bridge that always has a block, counting how often it is asked. */
667
+ function mkServingBridge(): { bridge: ContextBridge; resolves: number[] } {
668
+ const resolves: number[] = [];
669
+ return {
670
+ resolves,
671
+ bridge: {
672
+ enabled: true,
673
+ resolve: () => {
674
+ resolves.push(1);
675
+ return Promise.resolve({ block: "<project-context>ctx</project-context>", version: "v1", fetchedAtMs: 1 });
676
+ },
677
+ recordTurn: () => {},
678
+ flush: () => Promise.resolve(),
679
+ pruneBlocks: () => 0,
680
+ close: () => {},
681
+ },
682
+ };
683
+ }
684
+
685
+ function oneDispatch() {
686
+ const { router } = mkRouter([mkDecision("trivial", "cheap/model", { escalateTo: null })]);
687
+ const { upstream } = mkUpstream([
688
+ { kind: "chunks", chunks: [startChunk("cheap/model"), textChunk("high"), finishChunk("stop"), usageChunk({}, 0.0001)] },
689
+ ]);
690
+ const { ledger } = mkLedger();
691
+ const { store } = mkConversations();
692
+ const { sink, errors } = mkSink();
693
+ return { router, upstream, ledger, store, sink, errors };
694
+ }
695
+
696
+ test("a harness utility call gets no context block", async () => {
697
+ // Measured on one omp turn: seven tool-less side calls (title, ratings)
698
+ // each received the whole ~6k-token block — 42k prompt tokens for
699
+ // answers ABOUT the conversation. The tool array is the discriminator,
700
+ // exactly as for recording.
701
+ const { router, upstream, ledger, store, sink, errors } = oneDispatch();
702
+ const { bridge, resolves } = mkServingBridge();
703
+ const req: NormRequest = { ...mkReq(), agentdoxScope: "proj", tools: [] };
704
+
705
+ await runTurn(req, sink, { config: mkConfig({ enabled: false }), router, upstream, ledger, conversations: store, catalog, context: bridge }, new AbortController().signal);
706
+
707
+ expect(errors).toHaveLength(0);
708
+ expect(resolves).toHaveLength(0);
709
+ expect(store.load(req.conversationKey).contextVersion).toBeNull();
710
+ });
711
+
712
+ test("an agent turn with tools is injected", async () => {
713
+ const { router, upstream, ledger, store, sink, errors } = oneDispatch();
714
+ const { bridge, resolves } = mkServingBridge();
715
+ const req: NormRequest = { ...mkReq(), agentdoxScope: "proj", tools: [AGENT_TOOL] };
716
+
717
+ await runTurn(req, sink, { config: mkConfig({ enabled: false }), router, upstream, ledger, conversations: store, catalog, context: bridge }, new AbortController().signal);
718
+
719
+ expect(errors).toHaveLength(0);
720
+ expect(resolves).toHaveLength(1);
721
+ expect(store.load(req.conversationKey).contextVersion).toBe("v1");
722
+ });
723
+
724
+ test("context.injectWithoutTools restores injection into tool-less calls", async () => {
725
+ const { router, upstream, ledger, store, sink, errors } = oneDispatch();
726
+ const { bridge, resolves } = mkServingBridge();
727
+ const base = mkConfig({ enabled: false });
728
+ const config: RouterConfig = { ...base, context: { ...base.context, injectWithoutTools: true } };
729
+ const req: NormRequest = { ...mkReq(), agentdoxScope: "proj", tools: [] };
730
+
731
+ await runTurn(req, sink, { config, router, upstream, ledger, conversations: store, catalog, context: bridge }, new AbortController().signal);
732
+
733
+ expect(errors).toHaveLength(0);
734
+ expect(resolves).toHaveLength(1);
735
+ });
736
+ });
737
+
663
738
  describe("spend reaches the conversation total however the dispatch ends", () => {
664
739
  test("a dispatch that dies mid-stream still books what it was billed", async () => {
665
740
  // Live data: 152 aborted dispatches billed $0.9985 — 30% of all spend —