jeo-code 0.8.18 → 0.8.19

package/CHANGELOG.md

@@ -6,6 +6,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 The README mirrors the latest 5 entries — regenerate with `bun run changelog:sync`.
 
+## [0.8.19] - 2026-07-09
+_"프롬프트 라우팅에 안티그라비티 프로바이더의 경우, 안티그라비티용 소넷과 오퍼스도 3.5급으로 라우팅될수있도록해줘" — Antigravity re-exports 3 distinct model families (Anthropic Claude, Google Gemini, OpenAI GPT-OSS) behind one credential, structurally unlike every other provider. `high`/`complex` tier auto-select always resolved to Google's Gemini 3.5 rows: Anthropic's real 64,000-token output ceiling lost a same-thinking-tier tie to Google's 65,536 by a margin with no practical significance, and Gemini's 1M-token context further outranked Claude's real 200K window — so `antigravity/claude-sonnet-4-6`/`antigravity/claude-opus-4-6-thinking` were NEVER reachable through auto-select, even though both were already correctly `sizeClass`-tagged into the `high`/`complex` pools._
+
+### Fixed
+- **Antigravity's Claude Sonnet/Opus are now reachable at the `high`/`complex` routing tiers** (`src/agent/prompt-router.ts`) — added `antigravityCompanyPoolPick`: when Antigravity is credentialed and a tier's pool spans 2+ companies, session-stably spreads across ONE candidate per company (the strongest Antigravity row for that company) instead of the plain single-winner comparator always resolving to Google's Gemini rows. Wired into `resolveTierModel`'s `high`/`complex` branches, after the explicit `routing.crossProviderPool` opt-in but before the plain single-winner fallback — purely additive, default-on specifically for Antigravity's multi-vendor case (every other provider represents exactly one vendor and is unaffected).
+
+### Added
+- **Antigravity multi-company routing coverage** (`test/prompt-router.test.ts`) — 3 new tests: reachability across all 3 re-exported companies (Anthropic/Google/OpenAI) over 50 sessions, session-stability (same session always resolves to the same pick), and an explicit `roles.high` override still winning over the new spread. Updated one test whose assertion had locked in the old Gemini-only behavior — its actual invariant (never leak a bare public `gemini-*` id) still holds and is now asserted more precisely.
+
+### Verified
+- Live: `resolveTierModel("high"/"complex", …)` across 5 distinct real sessionIds confirms `antigravity/claude-sonnet-4-6` reachable at `high` and `antigravity/claude-opus-4-6-thinking` reachable at `complex`, spread across all 3 companies.
+- `bun test test/prompt-router.test.ts test/prompt-router-tiers.test.ts test/model-manager.test.ts test/launch-prompt-routing.test.ts test/model-provider-mapping.test.ts` — 159 pass / 0 fail.
+- Full `bun test` — 2893 pass / 0 fail across 291 files.
+- `bun run typecheck` — no errors.
+
 ## [0.8.18] - 2026-07-09
 _"라우팅 매 프롬프트마다 변경되는지, route why 의도대로 동작하는지 검증하고 배포할게있는지 확인해줘" — 2 parallel subagents live-verified per-prompt routing + `/route why` (including the previously-only-unit-tested post-call equivalent-model fallback path, now reproduced against a real mock 500-error server) and audited a disconnected `RouteHistory` class shipped as incomplete scaffolding in 0.8.16/0.8.17 (class + 11 tests existed, but nothing ever called it — no `/route history` subcommand, no wiring into the turn loop). Live testing also surfaced one real bug: the TUI footer's model/provider label never updated after a mid-turn fallback, staying on the pre-fallback model for the rest of that turn's render even though the backend decision (`lastRouteDecision`) was already correct._
 

package/README.ja.md

@@ -265,11 +265,11 @@ CI は `.github/workflows/npm-publish.yml` で公開します — GitHub リリ
 ## 変更履歴 (Changelog)
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.8.19]** (2026-07-09) — "프롬프트 라우팅에 안티그라비티 프로바이더의 경우, 안티그라비티용 소넷과 오퍼스도 3.5급으로 라우팅될수있도록해줘" — Antigravity re-exports 3 distinct model families (Anthropic Claude, Google Gemini, OpenAI GPT-OSS) behind one credential, structurally unlike every other provider. `high`/`complex` tier auto-select always resolved to Google's Gemini 3.5 rows: Anthropic's real 64,000-token output ceiling lost a same-thinking-tier tie to Google's 65,536 by a margin with no practical significance, and Gemini's 1M-token context further outranked Claude's real 200K window — so `antigravity/claude-sonnet-4-6`/`antigravity/claude-opus-4-6-thinking` were NEVER reachable through auto-select, even though both were already correctly `sizeClass`-tagged into the `high`/`complex` pools.
 - **[0.8.18]** (2026-07-09) — "라우팅 매 프롬프트마다 변경되는지, route why 의도대로 동작하는지 검증하고 배포할게있는지 확인해줘" — 2 parallel subagents live-verified per-prompt routing + `/route why` (including the previously-only-unit-tested post-call equivalent-model fallback path, now reproduced against a real mock 500-error server) and audited a disconnected `RouteHistory` class shipped as incomplete scaffolding in 0.8.16/0.8.17 (class + 11 tests existed, but nothing ever called it — no `/route history` subcommand, no wiring into the turn loop). Live testing also surfaced one real bug: the TUI footer's model/provider label never updated after a mid-turn fallback, staying on the pre-fallback model for the rest of that turn's render even though the backend decision (`lastRouteDecision`) was already correct.
 - **[0.8.17]** (2026-07-09) — "서브에이전트 이용해. 병렬루 부가 나머지모두 개선하고 검증후 배포까지하자" — a prior 4-way parallel subagent cross-verification pass found the README `[!CAUTION]` routing-lock block (added this session for "프롬프트 라우팅이 한번 정해지면 안바뀌는데?") was directionally correct but had 2 real inaccuracies, plus a concrete test-coverage gap: every existing pin/override regression test drove the lock via the `--model` CLI startup flag, never via typing `/model <name>` as an interactive slash command mid-session, and none proved persistence across 3+ consecutive turns.
 - **[0.8.16]** (2026-07-09) — "jeo code의 computer use 기능은 / 명령어로 설정할수있도록하고, 실동작검증까지진행해줘" — the desktop-automation `computer` tool was only togglable by hand-editing `computer.enabled` in `~/.jeo/config.json`; there was no in-session way to enable it for a single run without leaving a permanent config change behind.
 - **[0.8.15]** (2026-07-09) — "api 인증 모델도 사용가능하도록 딥리서치하고 개선해줘" — traced the API-authenticated GPT path to a routing catalog gap: `GET /models` with an OpenAI API key or an OpenAI-compatible provider returned real account-scoped model ids, and `/model` could show/pin them, but prompt routing still built its auto-select pools from static `MODEL_CATALOG` rows only. A custom `openaiBaseUrl` made this worse by treating every public OpenAI row as servable, so routing could choose `gpt-*`/`o*` ids that the configured local/Azure/proxy endpoint never listed.
-- **[0.8.14]** (2026-07-09) — "GPT 연결 실패가 여전히 무응답으로 끝나는 케이스를 잡아줘" — traced the remaining GPT/routing failure path to post-call continuity: the route target could be credentialed and servable, but still end a turn with a recoverable terminal result such as plain `OpenAI returned no content`, rate/quota pressure, model unavailable/not found, or an invalid agent-tool response. jeo already knew the prompt tier and equivalent credentialed fallbacks, but it only applied that fallback before the model call. After the call failed, the turn stopped instead of trying another model in the same tier.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.ko.md

@@ -266,11 +266,11 @@ CI는 `.github/workflows/npm-publish.yml`로 배포합니다 — GitHub 릴리
 ## 변경 이력 (Changelog)
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.8.19]** (2026-07-09) — "프롬프트 라우팅에 안티그라비티 프로바이더의 경우, 안티그라비티용 소넷과 오퍼스도 3.5급으로 라우팅될수있도록해줘" — Antigravity re-exports 3 distinct model families (Anthropic Claude, Google Gemini, OpenAI GPT-OSS) behind one credential, structurally unlike every other provider. `high`/`complex` tier auto-select always resolved to Google's Gemini 3.5 rows: Anthropic's real 64,000-token output ceiling lost a same-thinking-tier tie to Google's 65,536 by a margin with no practical significance, and Gemini's 1M-token context further outranked Claude's real 200K window — so `antigravity/claude-sonnet-4-6`/`antigravity/claude-opus-4-6-thinking` were NEVER reachable through auto-select, even though both were already correctly `sizeClass`-tagged into the `high`/`complex` pools.
 - **[0.8.18]** (2026-07-09) — "라우팅 매 프롬프트마다 변경되는지, route why 의도대로 동작하는지 검증하고 배포할게있는지 확인해줘" — 2 parallel subagents live-verified per-prompt routing + `/route why` (including the previously-only-unit-tested post-call equivalent-model fallback path, now reproduced against a real mock 500-error server) and audited a disconnected `RouteHistory` class shipped as incomplete scaffolding in 0.8.16/0.8.17 (class + 11 tests existed, but nothing ever called it — no `/route history` subcommand, no wiring into the turn loop). Live testing also surfaced one real bug: the TUI footer's model/provider label never updated after a mid-turn fallback, staying on the pre-fallback model for the rest of that turn's render even though the backend decision (`lastRouteDecision`) was already correct.
 - **[0.8.17]** (2026-07-09) — "서브에이전트 이용해. 병렬루 부가 나머지모두 개선하고 검증후 배포까지하자" — a prior 4-way parallel subagent cross-verification pass found the README `[!CAUTION]` routing-lock block (added this session for "프롬프트 라우팅이 한번 정해지면 안바뀌는데?") was directionally correct but had 2 real inaccuracies, plus a concrete test-coverage gap: every existing pin/override regression test drove the lock via the `--model` CLI startup flag, never via typing `/model <name>` as an interactive slash command mid-session, and none proved persistence across 3+ consecutive turns.
 - **[0.8.16]** (2026-07-09) — "jeo code의 computer use 기능은 / 명령어로 설정할수있도록하고, 실동작검증까지진행해줘" — the desktop-automation `computer` tool was only togglable by hand-editing `computer.enabled` in `~/.jeo/config.json`; there was no in-session way to enable it for a single run without leaving a permanent config change behind.
 - **[0.8.15]** (2026-07-09) — "api 인증 모델도 사용가능하도록 딥리서치하고 개선해줘" — traced the API-authenticated GPT path to a routing catalog gap: `GET /models` with an OpenAI API key or an OpenAI-compatible provider returned real account-scoped model ids, and `/model` could show/pin them, but prompt routing still built its auto-select pools from static `MODEL_CATALOG` rows only. A custom `openaiBaseUrl` made this worse by treating every public OpenAI row as servable, so routing could choose `gpt-*`/`o*` ids that the configured local/Azure/proxy endpoint never listed.
-- **[0.8.14]** (2026-07-09) — "GPT 연결 실패가 여전히 무응답으로 끝나는 케이스를 잡아줘" — traced the remaining GPT/routing failure path to post-call continuity: the route target could be credentialed and servable, but still end a turn with a recoverable terminal result such as plain `OpenAI returned no content`, rate/quota pressure, model unavailable/not found, or an invalid agent-tool response. jeo already knew the prompt tier and equivalent credentialed fallbacks, but it only applied that fallback before the model call. After the call failed, the turn stopped instead of trying another model in the same tier.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.md

@@ -273,11 +273,11 @@ Huge thanks to [gajae-code](https://github.com/Yeachan-Heo/gajae-code) for the i
 ## Changelog
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.8.19]** (2026-07-09) — "프롬프트 라우팅에 안티그라비티 프로바이더의 경우, 안티그라비티용 소넷과 오퍼스도 3.5급으로 라우팅될수있도록해줘" — Antigravity re-exports 3 distinct model families (Anthropic Claude, Google Gemini, OpenAI GPT-OSS) behind one credential, structurally unlike every other provider. `high`/`complex` tier auto-select always resolved to Google's Gemini 3.5 rows: Anthropic's real 64,000-token output ceiling lost a same-thinking-tier tie to Google's 65,536 by a margin with no practical significance, and Gemini's 1M-token context further outranked Claude's real 200K window — so `antigravity/claude-sonnet-4-6`/`antigravity/claude-opus-4-6-thinking` were NEVER reachable through auto-select, even though both were already correctly `sizeClass`-tagged into the `high`/`complex` pools.
 - **[0.8.18]** (2026-07-09) — "라우팅 매 프롬프트마다 변경되는지, route why 의도대로 동작하는지 검증하고 배포할게있는지 확인해줘" — 2 parallel subagents live-verified per-prompt routing + `/route why` (including the previously-only-unit-tested post-call equivalent-model fallback path, now reproduced against a real mock 500-error server) and audited a disconnected `RouteHistory` class shipped as incomplete scaffolding in 0.8.16/0.8.17 (class + 11 tests existed, but nothing ever called it — no `/route history` subcommand, no wiring into the turn loop). Live testing also surfaced one real bug: the TUI footer's model/provider label never updated after a mid-turn fallback, staying on the pre-fallback model for the rest of that turn's render even though the backend decision (`lastRouteDecision`) was already correct.
 - **[0.8.17]** (2026-07-09) — "서브에이전트 이용해. 병렬루 부가 나머지모두 개선하고 검증후 배포까지하자" — a prior 4-way parallel subagent cross-verification pass found the README `[!CAUTION]` routing-lock block (added this session for "프롬프트 라우팅이 한번 정해지면 안바뀌는데?") was directionally correct but had 2 real inaccuracies, plus a concrete test-coverage gap: every existing pin/override regression test drove the lock via the `--model` CLI startup flag, never via typing `/model <name>` as an interactive slash command mid-session, and none proved persistence across 3+ consecutive turns.
 - **[0.8.16]** (2026-07-09) — "jeo code의 computer use 기능은 / 명령어로 설정할수있도록하고, 실동작검증까지진행해줘" — the desktop-automation `computer` tool was only togglable by hand-editing `computer.enabled` in `~/.jeo/config.json`; there was no in-session way to enable it for a single run without leaving a permanent config change behind.
 - **[0.8.15]** (2026-07-09) — "api 인증 모델도 사용가능하도록 딥리서치하고 개선해줘" — traced the API-authenticated GPT path to a routing catalog gap: `GET /models` with an OpenAI API key or an OpenAI-compatible provider returned real account-scoped model ids, and `/model` could show/pin them, but prompt routing still built its auto-select pools from static `MODEL_CATALOG` rows only. A custom `openaiBaseUrl` made this worse by treating every public OpenAI row as servable, so routing could choose `gpt-*`/`o*` ids that the configured local/Azure/proxy endpoint never listed.
-- **[0.8.14]** (2026-07-09) — "GPT 연결 실패가 여전히 무응답으로 끝나는 케이스를 잡아줘" — traced the remaining GPT/routing failure path to post-call continuity: the route target could be credentialed and servable, but still end a turn with a recoverable terminal result such as plain `OpenAI returned no content`, rate/quota pressure, model unavailable/not found, or an invalid agent-tool response. jeo already knew the prompt tier and equivalent credentialed fallbacks, but it only applied that fallback before the model call. After the call failed, the turn stopped instead of trying another model in the same tier.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.zh.md

@@ -265,11 +265,11 @@ CI 通过 `.github/workflows/npm-publish.yml` 发布 — GitHub 发布 release
 ## 更新日志 (Changelog)
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.8.19]** (2026-07-09) — "프롬프트 라우팅에 안티그라비티 프로바이더의 경우, 안티그라비티용 소넷과 오퍼스도 3.5급으로 라우팅될수있도록해줘" — Antigravity re-exports 3 distinct model families (Anthropic Claude, Google Gemini, OpenAI GPT-OSS) behind one credential, structurally unlike every other provider. `high`/`complex` tier auto-select always resolved to Google's Gemini 3.5 rows: Anthropic's real 64,000-token output ceiling lost a same-thinking-tier tie to Google's 65,536 by a margin with no practical significance, and Gemini's 1M-token context further outranked Claude's real 200K window — so `antigravity/claude-sonnet-4-6`/`antigravity/claude-opus-4-6-thinking` were NEVER reachable through auto-select, even though both were already correctly `sizeClass`-tagged into the `high`/`complex` pools.
 - **[0.8.18]** (2026-07-09) — "라우팅 매 프롬프트마다 변경되는지, route why 의도대로 동작하는지 검증하고 배포할게있는지 확인해줘" — 2 parallel subagents live-verified per-prompt routing + `/route why` (including the previously-only-unit-tested post-call equivalent-model fallback path, now reproduced against a real mock 500-error server) and audited a disconnected `RouteHistory` class shipped as incomplete scaffolding in 0.8.16/0.8.17 (class + 11 tests existed, but nothing ever called it — no `/route history` subcommand, no wiring into the turn loop). Live testing also surfaced one real bug: the TUI footer's model/provider label never updated after a mid-turn fallback, staying on the pre-fallback model for the rest of that turn's render even though the backend decision (`lastRouteDecision`) was already correct.
 - **[0.8.17]** (2026-07-09) — "서브에이전트 이용해. 병렬루 부가 나머지모두 개선하고 검증후 배포까지하자" — a prior 4-way parallel subagent cross-verification pass found the README `[!CAUTION]` routing-lock block (added this session for "프롬프트 라우팅이 한번 정해지면 안바뀌는데?") was directionally correct but had 2 real inaccuracies, plus a concrete test-coverage gap: every existing pin/override regression test drove the lock via the `--model` CLI startup flag, never via typing `/model <name>` as an interactive slash command mid-session, and none proved persistence across 3+ consecutive turns.
 - **[0.8.16]** (2026-07-09) — "jeo code의 computer use 기능은 / 명령어로 설정할수있도록하고, 실동작검증까지진행해줘" — the desktop-automation `computer` tool was only togglable by hand-editing `computer.enabled` in `~/.jeo/config.json`; there was no in-session way to enable it for a single run without leaving a permanent config change behind.
 - **[0.8.15]** (2026-07-09) — "api 인증 모델도 사용가능하도록 딥리서치하고 개선해줘" — traced the API-authenticated GPT path to a routing catalog gap: `GET /models` with an OpenAI API key or an OpenAI-compatible provider returned real account-scoped model ids, and `/model` could show/pin them, but prompt routing still built its auto-select pools from static `MODEL_CATALOG` rows only. A custom `openaiBaseUrl` made this worse by treating every public OpenAI row as servable, so routing could choose `gpt-*`/`o*` ids that the configured local/Azure/proxy endpoint never listed.
-- **[0.8.14]** (2026-07-09) — "GPT 연결 실패가 여전히 무응답으로 끝나는 케이스를 잡아줘" — traced the remaining GPT/routing failure path to post-call continuity: the route target could be credentialed and servable, but still end a turn with a recoverable terminal result such as plain `OpenAI returned no content`, rate/quota pressure, model unavailable/not found, or an invalid agent-tool response. jeo already knew the prompt tier and equivalent credentialed fallbacks, but it only applied that fallback before the model call. After the call failed, the turn stopped instead of trying another model in the same tier.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jeo-code",
-  "version": "0.8.18",
+  "version": "0.8.19",
   "description": "Clean, highly optimized AI coding agent using spec-first loop",
   "type": "module",
   "main": "src/cli.ts",

package/src/agent/prompt-router.ts

@@ -456,6 +456,44 @@ function crossProviderPoolPick(tier: PromptTier, config: RoutingConfig, sessionI
   return pool.length > 0 ? selectFromPool(pool, sessionId) : null;
 }
 
+/** Antigravity re-exports THREE distinct model families (Anthropic's Claude
+ *  Sonnet/Opus, Google's own Gemini, OpenAI's GPT-OSS) behind one credential —
+ *  structurally different from every other provider, which represents exactly
+ *  one vendor. The tier's plain single-winner pick (`strongestMidTierCredentialed`/
+ *  `strongestCredentialed`) always resolves to Google's rows for `high`/`complex`:
+ *  Anthropic's real 64,000-token output ceiling loses a same-thinking-tier tie to
+ *  Google's 65,536 by a margin with no practical significance, and Gemini's
+ *  1M-token context further outranks Claude's real (but comparatively narrower)
+ *  200K window — so `antigravity/claude-sonnet-4-6`/`antigravity/claude-opus-4-6-thinking`
+ *  were NEVER reachable through auto-select even though both are already
+ *  correctly `sizeClass`-tagged into the `high`/`complex` pools (`tierModelPool`).
+ *  This applies the SAME session-stable spread `routing.crossProviderPool` uses
+ *  globally (opt-in, default off), but scoped and DEFAULT-ON specifically for
+ *  Antigravity's multi-company pool: one candidate per COMPANY (the strongest
+ *  Antigravity row for that company, by `compareStrengthAscending`) so Google's
+ *  several same-tier Gemini rows don't crowd out the single Anthropic/OpenAI
+ *  candidate 3-to-1 — each vendor Antigravity re-exports gets an equal, genuinely
+ *  reachable share. `null` when Antigravity isn't credentialed this turn, the
+ *  tier's Antigravity-only pool is empty, or it spans only ONE company (nothing
+ *  to spread — the plain single-winner path already picks correctly and this
+ *  would just add indirection). */
+function antigravityCompanyPoolPick(tier: PromptTier, config: RoutingConfig, sessionId: string | undefined): string | null {
+  if (!hasAntigravityOauth(config)) return null;
+  const pool = tierModelPool(tier, config)
+    .map(id => catalogMetadata(id))
+    .filter((m): m is CatalogModel => !!m && m.provider === "antigravity");
+  if (pool.length === 0) return null;
+  const byCompany = new Map<string, CatalogModel>();
+  for (const m of pool) {
+    const company = m.company ?? "antigravity";
+    const current = byCompany.get(company);
+    if (!current || compareStrengthAscending(current, m) < 0) byCompany.set(company, m);
+  }
+  if (byCompany.size < 2) return null; // one vendor in this tier -> nothing to spread across
+  const representatives = [...byCompany.values()].map(m => m.canonical).sort();
+  return selectFromPool(representatives, sessionId);
+}
+
 // --- Tier -> model/thinking resolution (inline off config.roles/config.routing.tiers; no static map). ---
 
 /** Exported for `jeo doctor`'s routing-preview diagnostic (same resolution the real
@@ -466,8 +504,8 @@ export function resolveTierModel(tier: PromptTier, config: RoutingConfig, sessio
   if (configured) return configured;
   if (tier === "trivial") return config.roles?.smol || crossProviderPoolPick(tier, config, sessionId) || cheapestCredentialed(config) || config.defaultModel;
   if (tier === "standard") return config.roles?.medium || config.roles?.high || crossProviderPoolPick(tier, config, sessionId) || config.defaultModel;
-  if (tier === "high") return config.roles?.high || config.roles?.medium || crossProviderPoolPick(tier, config, sessionId) || strongestMidTierCredentialed(config) || config.defaultModel;
-  if (tier === "complex") return config.roles?.xhigh || config.roles?.slow || crossProviderPoolPick(tier, config, sessionId) || strongestCredentialed(config) || config.defaultModel;
+  if (tier === "high") return config.roles?.high || config.roles?.medium || crossProviderPoolPick(tier, config, sessionId) || antigravityCompanyPoolPick(tier, config, sessionId) || strongestMidTierCredentialed(config) || config.defaultModel;
+  if (tier === "complex") return config.roles?.xhigh || config.roles?.slow || crossProviderPoolPick(tier, config, sessionId) || antigravityCompanyPoolPick(tier, config, sessionId) || strongestCredentialed(config) || config.defaultModel;
   return config.defaultModel;
 }
 

package/src/agent/session.ts

@@ -13,6 +13,11 @@ export interface SessionHeader {
   /** Model id pinned to this session; restored on resume so each session can carry
    *  its own model independent of the global default (per-session model selection). */
   model?: string;
+  /** Unsent input-box draft at last save, restored into the prompt on `/resume`
+   *  (gajae-code parity: an in-progress prompt survives a quit + resume instead
+   *  of being silently lost). Absent/empty when there was nothing to restore. */
+  draft?: string;
+
 }
 
 export interface SessionEntry {
@@ -416,6 +421,25 @@ export async function updateSessionModel(id: string, model: string, cwd = proces
     return true;
   }, cwd);
 }
+/**
+ * Persist (or clear) the unsent input-box draft in a session's JSONL header so a
+ * later `/resume` restores it into the prompt (gajae-code parity). Empty text
+ * clears any stale draft rather than writing an empty string. No-op (no write)
+ * when the header's current draft already matches. Throws a clear Error if the
+ * session file does not exist.
+ */
+export async function updateSessionDraft(id: string, draft: string, cwd = process.cwd()): Promise<void> {
+  const trimmed = draft.trim();
+  const next = trimmed ? trimmed : undefined;
+  await rewriteSessionHeader(id, header => {
+    if ((header.draft ?? undefined) === next) return false;
+    if (next === undefined) delete header.draft;
+    else header.draft = next;
+    return true;
+  }, cwd);
+}
+
+
 
 /**
  * Delete a session file.

package/src/commands/launch/session-slash.ts

@@ -47,6 +47,11 @@ export interface SessionSlashResult {
   lastUserInput?: string;
   /** Present only when changed (by `/session resume` re-seeding retry state). */
   lastReply?: string;
+  /** Present only when changed (by `/session resume` restoring header.draft into
+   *  the prompt box). */
+  draft?: string;
+
+
 }
 
 /**
@@ -62,14 +67,18 @@ export async function runSessionSlash(input: string, ctx: SessionSlashCtx): Prom
   let sessionModelChanged = false;
   let lastUserInput: string | undefined;
   let lastReply: string | undefined;
+  let draft: string | undefined;
+
 
   const result = (): SessionSlashResult => ({
     sessionId,
     ...(sessionModelChanged ? { sessionModel } : {}),
     ...(lastUserInput !== undefined ? { lastUserInput } : {}),
     ...(lastReply !== undefined ? { lastReply } : {}),
+    ...(draft !== undefined ? { draft } : {}),
   });
 
+
   const tokens = input.substring(8).trim().split(/\s+/).filter(Boolean);
   const sub = (tokens[0] ?? "").toLowerCase();
 
@@ -125,6 +134,10 @@ export async function runSessionSlash(input: string, ctx: SessionSlashCtx): Prom
         sessionId = rid;
         // Restore the model this session was last using (per-session model).
         if (header.model) { sessionModel = header.model; sessionModelChanged = true; }
+        // Restore the unsent input-box draft left over at last save, if any.
+        if (header.draft) draft = header.draft;
+
+
         // Seed /retry + reply marker from the last user/assistant turn.
         lastUserInput = ""; lastReply = "";
         for (let k = history.length - 1; k >= 1; k--) {

package/src/commands/launch.ts

@@ -110,6 +110,8 @@ import {
   latestSessionId,
   exportSession,
   updateSessionModel,
+  updateSessionDraft,
+
   appendCompaction,
   resolveSessionRef,
 } from "../agent/session";
@@ -752,6 +754,10 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
 
   // pi-style session persistence: resume an existing session or create a new one.
   let sessionId: string | undefined;
+  // Unsent input-box draft restored from a resumed session's header (gajae-code
+  // parity) — folded into the very first prompt once `queuedPromptInput` exists.
+  let restoredDraft = "";
+
   let compactionSeq = 0;
   if (!flags.noSession) {
     if (flags.resumeInteractive) {
@@ -770,6 +776,8 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
             for (const m of messages) history.push(m);
             sessionId = id;
             if (!initialSessionModel && header.model) sessionModel = header.model;
+            if (header.draft) restoredDraft = header.draft;
+
             const modelNote = sessionModel ? ` · model ${sessionModel}` : "";
             console.log(`Resumed session ${id} (${messages.length} messages).${modelNote}`);
           } catch (err) {
@@ -808,6 +816,8 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
           // Restore the model this session was last using unless the CLI explicitly
           // pinned one (flags.model/role/provider → initialSessionModel wins).
           if (!initialSessionModel && header.model) sessionModel = header.model;
+          if (header.draft) restoredDraft = header.draft;
+
           const modelNote = sessionModel ? ` · model ${sessionModel}` : "";
           console.log(`Resumed session ${id} (${messages.length} messages).${modelNote}`);
         } catch (err) {
@@ -2183,6 +2193,12 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
   // (the user explicitly invoked them — no second Enter).
   const pendingMidTurnCommands: string[] = [];
   const queuedPromptInput: PromptInputQueue = { pendingLines: pendingStdinLines, partial: "", pastedLines: [], inPaste: false };
+  // Fold a restored `/resume` draft (session header `draft`) into the very first
+  // prompt the same way a queued mid-turn prefill lands — visible in the input box,
+  // Enter to run, Esc/Ctrl+U to discard (gajae-code parity: an in-progress prompt
+  // survives a quit + resume instead of being silently lost).
+  if (restoredDraft) queuedPromptInput.partial = restoredDraft;
+
   queueBusyInput = (chunk: string) => captureLivePromptInputChunk(queuedPromptInput, chunk);
   queueBusyPasteActive = () => queuedPromptInput.inPaste;
   queueBusySnapshot = () => ({
@@ -2194,6 +2210,19 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
     // default routes to the immediate-dispatch queue served at the top of the loop.
     (jeoEnv("NO_MIDTURN_DISPATCH") === "1" ? pendingStdinLines : pendingMidTurnCommands).push(line);
   };
+  // Persist (or clear) the unsent input-box draft into the session header so a later
+  // `/resume` restores it (gajae-code parity). Best-effort, mirrors persistSessionModel;
+  // reads readline's own line buffer since that is the single source of truth for
+  // "what's currently typed but not submitted" (same buffer the box-render/prefill
+  // logic above reads from).
+  const persistSessionDraft = async (): Promise<void> => {
+    if (flags.noSession || !sessionId) return;
+    const current = (rl as unknown as { line?: string }).line ?? queuedPromptInput.partial ?? "";
+    try {
+      await updateSessionDraft(sessionId, current, cwd);
+    } catch { /* best-effort */ }
+  };
+
   // Bracketed-paste line routing at the PROMPT: readline strips the 2004 markers
   // and replays pasted lines as synthetic keypresses, emitting paste-start /
   // paste-end around them. Lines submitted INSIDE that window are intentional
@@ -3695,6 +3724,8 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
     if (pickResult.sessionModel !== undefined) sessionModel = pickResult.sessionModel;
     if (pickResult.lastUserInput !== undefined) lastUserInput = pickResult.lastUserInput;
     if (pickResult.lastReply !== undefined) lastReply = pickResult.lastReply;
+    if (pickResult.draft) queuedPromptInput.partial = pickResult.draft;
+
   }
 
   // Remote (Telegram) free-text reply / config command routing — wired here,
@@ -3896,7 +3927,9 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
         });
         sessionId = result.sessionId;
         if (result.sessionModel !== undefined) sessionModel = result.sessionModel;
-        if (result.lastUserInput !== undefined) lastUserInput = result.lastUserInput;
+        if (result.lastReply !== undefined) lastReply = result.lastReply;
+        if (result.draft) queuedPromptInput.partial = result.draft;
+
         if (result.lastReply !== undefined) lastReply = result.lastReply;
         continue;
       }
@@ -4598,7 +4631,9 @@ export async function runLaunchCommand(args: string[]): Promise<void> {
       }
     }
   disarmPreview(); // clear footer + restore full-screen scrolling before leaving the REPL
+  await persistSessionDraft();
   if (hardExitOnLoopEnd) {
+
     try {
       disarmPreview();
       out.write("\x1b[?25h\n");