Merge pull request #3518 from jeremymcs/fix/bootstrap-prefer-gsd-preferences

fix(gsd): prefer PREFERENCES.md over settings.json for bootstrap model
This commit is contained in:
Jeremy McSpadden 2026-04-04 19:50:09 -05:00 committed by GitHub
commit 34baf57e40
4 changed files with 186 additions and 9 deletions

View file

@ -80,6 +80,7 @@ import { join } from "node:path";
import { sep as pathSep } from "node:path";
import { resolveProjectRootDbPath } from "./bootstrap/dynamic-tools.js";
import { resolveDefaultSessionModel } from "./preferences-models.js";
import type { WorktreeResolver } from "./worktree-resolver.js";
export interface BootstrapDeps {
@ -155,12 +156,16 @@ export async function bootstrapAutoSession(
// Capture the user's session model before guided-flow dispatch can apply a
// phase-specific planning model for a discuss turn (#2829).
const startModelSnapshot = ctx.model
? {
provider: ctx.model.provider,
id: ctx.model.id,
}
: null;
//
// GSD PREFERENCES.md takes priority over the session model from settings.json
// (#3517). The session model (ctx.model) comes from findInitialModel() which
// reads defaultProvider/defaultModel from ~/.gsd/agent/settings.json. When
// the user has explicit model preferences in PREFERENCES.md, those should win.
const preferredModel = resolveDefaultSessionModel(ctx.model?.provider);
const startModelSnapshot = preferredModel
?? (ctx.model
? { provider: ctx.model.provider, id: ctx.model.id }
: null);
try {
// Validate GSD_PROJECT_ID early so the user gets immediate feedback

View file

@ -107,6 +107,84 @@ export function resolveModelWithFallbacksForUnit(unitType: string): ResolvedMode
};
}
/**
* Resolve the default session model from GSD preferences.
*
* Used at auto-mode bootstrap to override the session model that was
* determined by settings.json (defaultProvider/defaultModel). When
* PREFERENCES.md (or project preferences) configures an `execution` model
* we treat that as the session default. Falls back through execution
* planning first configured model.
*
* Accepts an optional `sessionProvider` for bare model IDs that don't
* include an explicit provider prefix (e.g. `gpt-5.4` instead of
* `openai-codex/gpt-5.4`). When a bare ID is found and sessionProvider
* is available, the session provider is used. Without sessionProvider,
* bare IDs are still returned with provider set to the bare ID itself
* so downstream resolution (resolveModelId) can match it.
*
* Returns `{ provider, id }` or `undefined` if no model preference is
* configured.
*/
export function resolveDefaultSessionModel(
sessionProvider?: string,
): { provider: string; id: string } | undefined {
const prefs = loadEffectiveGSDPreferences();
if (!prefs?.preferences.models) return undefined;
const m = prefs.preferences.models as GSDModelConfigV2;
// Priority: execution → planning → first configured value
const candidates: Array<string | GSDPhaseModelConfig | undefined> = [
m.execution,
m.planning,
m.research,
m.discuss,
m.completion,
m.validation,
m.subagent,
];
for (const cfg of candidates) {
if (!cfg) continue;
// Normalize to provider + id from the various config shapes
let provider: string | undefined;
let id: string;
if (typeof cfg === "string") {
const slashIdx = cfg.indexOf("/");
if (slashIdx !== -1) {
provider = cfg.slice(0, slashIdx);
id = cfg.slice(slashIdx + 1);
} else {
// Bare model ID (e.g. "gpt-5.4") — use session provider as context
provider = sessionProvider;
id = cfg;
}
} else {
// Object config: { model, provider?, fallbacks? }
if (cfg.provider) {
provider = cfg.provider;
} else if (cfg.model.includes("/")) {
const slashIdx = cfg.model.indexOf("/");
provider = cfg.model.slice(0, slashIdx);
id = cfg.model.slice(slashIdx + 1);
return { provider, id };
} else {
provider = sessionProvider;
}
id = cfg.model;
}
if (provider && id) {
return { provider, id };
}
}
return undefined;
}
/**
* Determines the next fallback model to try when the current model fails.
* If the current model is not in the configured list, returns the primary model.

View file

@ -7,8 +7,10 @@ const sourcePath = join(import.meta.dirname, "..", "auto-start.ts");
const source = readFileSync(sourcePath, "utf-8");
test("bootstrapAutoSession snapshots ctx.model before guided-flow entry (#2829)", () => {
const snapshotIdx = source.indexOf("const startModelSnapshot = ctx.model");
assert.ok(snapshotIdx > -1, "auto-start.ts should snapshot ctx.model at bootstrap start");
// #3517 changed the snapshot to prefer GSD preferences, but the ordering
// guarantee still holds: the snapshot must be built before guided-flow.
const snapshotIdx = source.indexOf("const startModelSnapshot = preferredModel");
assert.ok(snapshotIdx > -1, "auto-start.ts should snapshot model at bootstrap start");
const firstDiscussIdx = source.indexOf('await showSmartEntry(ctx, pi, base, { step: requestedStepMode });');
assert.ok(firstDiscussIdx > -1, "auto-start.ts should route through showSmartEntry during guided flow");
@ -26,3 +28,21 @@ test("bootstrapAutoSession restores autoModeStartModel from the early snapshot (
const snapshotRefIdx = source.indexOf("provider: startModelSnapshot.provider", assignmentIdx);
assert.ok(snapshotRefIdx > -1, "autoModeStartModel should be restored from startModelSnapshot");
});
test("bootstrapAutoSession prefers GSD PREFERENCES.md over settings.json for start model (#3517)", () => {
// resolveDefaultSessionModel() should be called before the snapshot is built
const preferredIdx = source.indexOf("const preferredModel = resolveDefaultSessionModel(");
assert.ok(preferredIdx > -1, "auto-start.ts should call resolveDefaultSessionModel()");
// Session provider should be passed for bare model ID resolution
const withProviderIdx = source.indexOf("resolveDefaultSessionModel(ctx.model?.provider)");
assert.ok(withProviderIdx > -1, "auto-start.ts should pass ctx.model?.provider for bare ID resolution");
const snapshotIdx = source.indexOf("const startModelSnapshot = preferredModel");
assert.ok(snapshotIdx > -1, "startModelSnapshot should use preferredModel when available");
assert.ok(
preferredIdx < snapshotIdx,
"resolveDefaultSessionModel() must be called before building startModelSnapshot",
);
});

View file

@ -1,5 +1,6 @@
/**
* Tests for model config isolation between concurrent instances (#650, #1065).
* Tests for model config isolation between concurrent instances (#650, #1065)
* and GSD preferences override of settings.json defaults (#3517).
*/
import { describe, it, beforeEach, afterEach } from "node:test";
@ -155,3 +156,76 @@ describe("session model recovery on error (#1065)", () => {
"Recovery should be skipped when no session model was captured");
});
});
// ─── GSD Preferences override settings.json (#3517) ─────────────────────────
describe("GSD preferences override settings.json for session model (#3517)", () => {
it("preferredModel takes priority over ctx.model when both are available", () => {
// Simulates auto-start.ts logic: preferredModel ?? ctx.model snapshot
const preferredModel = { provider: "openai-codex", id: "gpt-5.4" };
const ctxModel = { provider: "claude-code", id: "claude-sonnet-4-6" };
const startModelSnapshot = preferredModel
?? { provider: ctxModel.provider, id: ctxModel.id };
assert.equal(startModelSnapshot.provider, "openai-codex",
"preferredModel provider should win over ctx.model");
assert.equal(startModelSnapshot.id, "gpt-5.4",
"preferredModel id should win over ctx.model");
});
it("falls back to ctx.model when no GSD preferences are configured", () => {
const preferredModel: { provider: string; id: string } | undefined = undefined;
const ctxModel = { provider: "claude-code", id: "claude-sonnet-4-6" };
const startModelSnapshot = preferredModel
?? { provider: ctxModel.provider, id: ctxModel.id };
assert.equal(startModelSnapshot.provider, "claude-code",
"should fall back to ctx.model provider when no preferences");
assert.equal(startModelSnapshot.id, "claude-sonnet-4-6",
"should fall back to ctx.model id when no preferences");
});
it("handles null ctx.model with no preferences gracefully", () => {
const preferredModel: { provider: string; id: string } | undefined = undefined;
// Use a function to prevent TS from narrowing to `never` in the ternary
function getCtxModel(): { provider: string; id: string } | null { return null; }
const ctxModel = getCtxModel();
const startModelSnapshot = preferredModel
?? (ctxModel ? { provider: ctxModel.provider, id: ctxModel.id } : null);
assert.equal(startModelSnapshot, null,
"should be null when neither preferences nor ctx.model exist");
});
it("bare model ID uses session provider when available", () => {
// Simulates: PREFERENCES.md has "gpt-5.4" (no provider), session is openai-codex
const preferredModel = { provider: "openai-codex", id: "gpt-5.4" }; // from resolveDefaultSessionModel("openai-codex")
const ctxModel = { provider: "openai-codex", id: "claude-sonnet-4-6" };
const startModelSnapshot = preferredModel
?? { provider: ctxModel.provider, id: ctxModel.id };
assert.equal(startModelSnapshot.provider, "openai-codex");
assert.equal(startModelSnapshot.id, "gpt-5.4",
"bare model ID from preferences should still override ctx.model");
});
it("stale settings.json does not leak when preferences are set", () => {
// Scenario: settings.json has claude-code, PREFERENCES.md has openai-codex
const settingsJsonDefault = { provider: "claude-code", id: "claude-sonnet-4-6" };
const preferencesModel = { provider: "openai-codex", id: "gpt-5.4" };
// auto-start.ts captures preferredModel first, which preempts settingsJsonDefault
const startModelSnapshot = preferencesModel ?? settingsJsonDefault;
assert.equal(startModelSnapshot.provider, "openai-codex",
"PREFERENCES.md must override stale settings.json provider");
assert.equal(startModelSnapshot.id, "gpt-5.4",
"PREFERENCES.md must override stale settings.json model");
assert.notEqual(startModelSnapshot.provider, settingsJsonDefault.provider,
"settings.json provider must NOT leak through");
});
});