@ponythewhite/base-context 1.0.0 → 1.0.2

package/CHANGELOG.md

@@ -1,5 +1,42 @@
 # Changelog
 
+## [1.0.2] - 2026-09-17
+
+- Fixed RPC requests waiting for a timeout after the agent process exits, and removed the shutdown delay for an already-exited process.
+
+## [1.0.1] - 2026-09-16
+
+- Made the Synerise base-context installer the primary setup path, with automatic prerequisites, optional shell-profile updates, and clear launch instructions.
+- Changed fresh managed Python environments to Python 3.13 and refused cross-minor snapshots before restore without overwriting saved state.
+- Removed remote telemetry, trace uploads and sharing, their configuration, and the `/traces` command; retained local diagnostics.
+- Removed Prime account, provider, team, and billing integration and the bundled Prime products skill.
+- Required provider-first fresh setup and explicit model selection, persisted the choice and login credentials, and removed scope-based model defaults.
+- Preserved supported saved model selections while authentication needs setup.
+- Fixed authentication failures to mark the credential used by the request rather than a replacement credential.
+- Restored final exact progress notification for explicit saved-session list requests.
+- Removed upstream-only Linear ticket checks and directed contribution links to Base Context.
+- Prevented uv bootstrap from editing shell profiles before the installer’s explicit PATH consent.
+- Prevented repeated installer runs from adding duplicate PATH entries for managed Node.js.
+- Propagated failed RPC prompt admission to SDK callers without treating successful acknowledgement as completion.
+- Removed remote session sharing through `/share` and stale `/traces` menu entries while retaining local `/export`.
+- Removed stale Prime-team wording from model authentication errors.
+- Added persisted `/agents N` control with a default of four concurrent subagents across the root family, including running and idle residents but excluding the main agent and inactive sessions. Lowering the limit only restricts new admissions.
+- Preserved subagent capacity across daemon worker recovery and saved-session resumes, and advanced the daemon contract to protocol 13 / schema 49.
+- Released newly bound context sources when request admission closes during disposal, preventing child shutdown from hanging.
+- Kept independently resumed live subagents visible and routable when their parent publishes inactive saved-child entries.
+- Fixed Bedrock provider loading in the bundled Node CLI while retaining native request-attempt accounting and local logs.
+- Fixed returning to the agents view during chat initialization without disposing its connection too early or applying late heartbeat updates.
+- Fixed failed terminal turns being marked completed by using a factual error recap with the existing needs-input state, without a classifier request.
+- Scoped daemon discovery and shutdown candidates to the current base-context state root, including owned custom sockets, rather than every matching process on the machine (adapted from [Prime Agent #2361](https://github.com/PrimeIntellect-ai/prime-agent/pull/2361)).
+- Fixed credential rejection attribution when a custom authentication header overrides the selected API key.
+- Handled current Python kernel pipe errors without crashing the worker (adapted from Prime Agent #2213).
+- Prevented blocked input-pump idle waits from starving the I/O needed to resume queued work.
+- Allowed Python `rlm.delete_subagent` to accept the handle returned by `await rlm(...)`, without changing the callable spawn API or subagent limits.
+- Kept drafts alive during pending attaches and handled asynchronous discard failures without crashing other sessions.
+- Rejected missing CLI option values, invalid modes and invalid thinking levels while retaining extension flags and `--` handling.
+- Corrected RPC examples to protocol 13 and schema revision 49, and documented actual provider authentication availability.
+- Reported background scheduler storage failures instead of letting an unhandled promise rejection terminate the daemon worker.
+
 ## [1.0.0] - 2026-09-15
 
 - Added native tool intent and finalized execution evidence to session history without duplicate results, with assistant source order restored for parallel tools.

package/NOTICE

@@ -14,9 +14,4 @@ this fork is not an official Prime Intellect release.
 Prime Agent builds on Mario Zechner's Pi project:
 https://github.com/badlogic/pi-mono
 
-We also credit Prime Intellect's PrimeRL authors for their open-source
-reinforcement-learning work: https://github.com/PrimeIntellect-ai/prime-rl
-This acknowledgement does not imply that PrimeRL is a runtime dependency
-or that upstream authors endorse this fork.
-
 Third-party dependencies retain their own licenses and notices.

package/README.md

@@ -4,27 +4,44 @@
 
 An MIT-licensed coding and research agent for tasks that outgrow a chat window. Base Context combines a persistent Python workspace, recursive agents, and source-backed context management. It is developed by [Synerise](https://synerise.com) as a fork of [Prime Agent](https://github.com/PrimeIntellect-ai/prime-agent).
 
-[Repository](https://github.com/BaseModelAI/base-context) · [Quickstart](docs/quickstart.md) · [Documentation](docs/index.md) · [Why this fork](docs/fork-philosophy.md)
+[Repository](https://github.com/BaseModelAI/base-context) · [Quickstart](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/quickstart.md) · [Documentation](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/index.md) · [Why this fork](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/fork-philosophy.md)
 
-[![One historical SDK study: Base Context achieved 89/90 runtime-clean strict finishes versus Prime Agent 64/90, with 2 versus 26 additional attempts and 24.16% less cumulative attempt time.](https://raw.githubusercontent.com/BaseModelAI/base-context/v1.0.0/packages/coding-agent/docs/images/benchmarks/benchmark-overview.png)](https://github.com/BaseModelAI/base-context/blob/v1.0.0/benchmarks/python-realworld-30/REPORT.md)
+[![One historical SDK study: Base Context achieved 89/90 runtime-clean strict finishes versus Prime Agent 64/90, with 2 versus 26 additional attempts and 24.16% less cumulative attempt time.](https://raw.githubusercontent.com/BaseModelAI/base-context/v1.0.1/packages/coding-agent/docs/images/benchmarks/benchmark-overview.png)](https://github.com/BaseModelAI/base-context/blob/v1.0.1/benchmarks/python-realworld-30/REPORT.md)
 
-*One study, not a universal ranking: historical SDK/shared-Bash results, not a fresh measurement of release 1.0.0. “Clean” adds runtime requirements to task correctness. [Methodology and full results](https://github.com/BaseModelAI/base-context/blob/v1.0.0/benchmarks/python-realworld-30/REPORT.md).*
+*One study, not a universal ranking: historical SDK/shared-Bash results, not a fresh measurement of this release. “Clean” adds runtime requirements to task correctness. [Methodology and full results](https://github.com/BaseModelAI/base-context/blob/v1.0.1/benchmarks/python-realworld-30/REPORT.md).*
 
-## Install
+## Install (recommended)
 
-Use Node.js **22.12+ on the 22.x line, or 23.3+**, and npm:
+On **macOS or Linux**, run the interactive installer:
 
 ```bash
-npm install -g @ponythewhite/base-context
-cd /path/to/project
+curl -fsSL https://github.com/BaseModelAI/base-context/releases/latest/download/install.sh | bash
+```
+
+**Use this instead of the npm route below. You do not install Python or `uv` first.** The installer checks Node.js/npm and offers to install a compatible version if needed. It installs `uv` when needed, prepares the agent's Python environment, and installs the bundled runtime **before activating the CLI**.
+
+Follow the final PATH/activation command printed by the installer, then start work:
+
+```bash
+cd /path/to/your/project
 base-context
 ```
 
-Run `/login` to configure an authorized provider, then `/model` to select a model. You can also set the provider's API-key environment variable before launch. See [providers](docs/providers.md); model catalog entries do not grant subscription access.
+In the terminal UI, select and authenticate with a supported provider using `/login`, then choose a model with `/model`. Both choices are explicit. The installer prepares the application; you use only the selected provider's account and authentication. See [provider setup](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/providers.md).
+
+### npm alternative: for users who already manage Node.js
 
-Install [uv](https://docs.astral.sh/uv/getting-started/installation/) for the Python workspace. First use prepares Python 3.11 and the bundled `base-context-runtime` under `~/.base-context/runtime`. Initial setup can download dependencies. The Python import remains `rlm`.
+Use Node.js **22.12+ on the 22.x line, or 23.3+**, and npm. These must already be installed **before** this route:
 
-For source builds, versioned owned installations, manual Python environments, updates, and rollback, see [installation](docs/installation.md). Prime Agent installers do not install Base Context.
+```bash
+npm install -g @ponythewhite/base-context
+cd /path/to/your/project
+BASE_CONTEXT_INSTALL_UV=1 base-context
+```
+
+`npm install` installs the CLI; it does not prepare Python at that step. Starting a normal CLI session begins preparing the managed Python environment in the background. `BASE_CONTEXT_INSTALL_UV=1` lets that bootstrap install `uv` if it is missing. **No manual Python installation is needed.** Initial setup needs network access and can take a little longer; later launches reuse the environment.
+
+The environment-variable syntax above is for Bash/Zsh. See [installation](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/installation.md) for PowerShell, manual Python environments, updates, and rollback.
 
 ## Why Base Context?
 
@@ -47,7 +64,7 @@ Our different bet is **how to manage the context around that work**. A longer tr
 
 Think of retained history as a **project notebook**, and the model's context as your **desk**. You do not need every notebook page on the desk at once. You do need the current task, the relevant evidence, and a way to fetch an earlier page. That is a design analogy—not a promise that all information is retained forever or that summaries are lossless.
 
-![Base Context architecture: retain source history, select and recover a task-aware working set, then send the supported model request. Stable epochs preserve accepted context choices.](https://raw.githubusercontent.com/BaseModelAI/base-context/v1.0.0/docs/images/benchmarks/context-working-set.png)
+![Base Context architecture: retain source history, select and recover a task-aware working set, then send the supported model request. Stable epochs preserve accepted context choices.](https://raw.githubusercontent.com/BaseModelAI/base-context/v1.0.1/packages/coding-agent/docs/images/benchmarks/context-working-set.png)
 
 ### How the design delivers
 
@@ -64,9 +81,9 @@ Task state is selected recorded evidence, not automatically current truth. Budge
 
 These choices favor **recoverable evidence and controlled working sets**, even when that requires more structure or refusing a request that cannot meet its configured contract. They describe this fork's emphasis—not a claim that every capability is absent from every upstream version.
 
-The fork also owns its package, `base-context` command, `~/.base-context` state, and Python runtime distribution. SSE is the default transport; other supported transports are opt-in. Remote telemetry and trace sharing are off by default and require explicit configuration.
+The fork also owns its package, `base-context` command, `~/.base-context` state, and Python runtime distribution. SSE is the default transport; other supported transports are opt-in. Diagnostics stay local; there is no telemetry upload feature.
 
-There is no promise of unlimited context, universal provider support, guaranteed savings, or lossless summaries. Read [context management](docs/context-management.md) and [fork philosophy](docs/fork-philosophy.md) for the contracts and limits.
+There is no promise of unlimited context, universal provider support, guaranteed savings, or lossless summaries. Read [context management](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/context-management.md) and [fork philosophy](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/fork-philosophy.md) for the contracts and limits.
 
 ## Use it
 
@@ -82,7 +99,9 @@ For independent work, ask the agent to delegate:
 Delegate the API review to a subagent. Update the documentation while it runs, then read its reply before integrating the findings.
 ```
 
-The agent works through a persistent Python REPL. `await rlm(...)` returns a **child admission handle**, not the child's answer. Child results arrive through explicit messages or files. See [RLM programming](docs/rlm.md).
+The agent works through a persistent Python REPL. `await rlm(...)` returns a **child admission handle**, not the child's answer. Child results arrive through explicit messages or files. See [RLM programming](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/rlm.md).
+
+Use `/agents N` to save the concurrent subagent limit globally (default **4**); `/agents` shows its current value. Pending admissions and running or idle subagents count across the root family; the main agent and inactive sessions do not. Lowering the limit never stops existing subagents. A value of **0** blocks new admissions.
 
 Add an `AGENTS.md` file for project instructions. Use `/settings` for common options or `.base-context/settings.json` for project configuration. Global state lives in `~/.base-context`; `BASE_CONTEXT_HOME` selects a separate absolute root.
 
@@ -95,33 +114,31 @@ base-context doctor
 base-context shutdown
 ```
 
-See [usage and CLI](docs/usage.md), [settings](docs/settings.md), and [long-running agents](docs/long-running-agents.md) for complete instructions.
+See [usage and CLI](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/usage.md), [settings](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/settings.md), and [long-running agents](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/long-running-agents.md) for complete instructions.
 
 ## Benchmark evidence
 
-![Runtime-clean strict finishes for all three models: Sol 30/30 versus 18/30; Astra 30/30 versus 19/30; DeepSeek 29/30 versus 27/30. Base Context is first in each pair; all chart scales run from zero to thirty.](https://raw.githubusercontent.com/BaseModelAI/base-context/v1.0.0/packages/coding-agent/docs/images/benchmarks/benchmark-models.png)
+![Runtime-clean strict finishes for all three models: Sol 30/30 versus 18/30; Astra 30/30 versus 19/30; DeepSeek 29/30 versus 27/30. Base Context is first in each pair; all chart scales run from zero to thirty.](https://raw.githubusercontent.com/BaseModelAI/base-context/v1.0.1/packages/coding-agent/docs/images/benchmarks/benchmark-models.png)
 
 One frozen SDK-level study covered 30 Python tasks, three model selections, and two agents: 180 cells and 208 attempts. Base Context reached **90/90 terminal strict passes**, versus **87/90** for stock Prime Agent, with **2 versus 26 additional task attempts** and **24.16% less cumulative attempt lifecycle time**.
 
-The harness used a shared Bash tool, not the native Python/RLM workflow. This is cumulative attempt time, not campaign wall time, CPU time, or user-perceived latency. The study used logical `medium` effort, a fixed single-deferred-retry policy, and historical package `0.1.0` at mixed source revisions (`84a7e6f` for Sol/Astra; `077f463` for DeepSeek). Release `1.0.0` builds on the latter source line; it has not been newly measured in this study. Sixteen attempts have unknown cost; full fees and a whole-campaign cost advantage are unknown. Reported prices are API-equivalent estimates, not cash charges.
+The harness used a shared Bash tool, not the native Python/RLM workflow. This is cumulative attempt time, not campaign wall time, CPU time, or user-perceived latency. The study used logical `medium` effort, a fixed single-deferred-retry policy, and historical package `0.1.0` at mixed source revisions (`84a7e6f` for Sol/Astra; `077f463` for DeepSeek). Current releases build on the latter source line; they have not been newly measured in this study. Sixteen attempts have unknown cost; full fees and a whole-campaign cost advantage are unknown. Reported prices are API-equivalent estimates, not cash charges.
 
-Read the [full report and methodology](https://github.com/BaseModelAI/base-context/blob/main/benchmarks/python-realworld-30/REPORT.md) and [reproduction guide](https://github.com/BaseModelAI/base-context/blob/main/benchmarks/python-realworld-30/REPRODUCE.md).
+Read the [full report and methodology](https://github.com/BaseModelAI/base-context/blob/v1.0.1/benchmarks/python-realworld-30/REPORT.md) and [reproduction guide](https://github.com/BaseModelAI/base-context/blob/v1.0.1/benchmarks/python-realworld-30/REPRODUCE.md).
 
 ## Integrate and extend
 
-- [SDK](docs/sdk.md), [JSON mode](docs/json.md), [RPC](docs/rpc.md), and [ACP](docs/acp.md)
-- [Python and markdown skills](docs/skills.md), [MCP](docs/mcp-integrations.md), and [extensions](docs/extensions.md)
-- [Custom models](docs/models.md), [custom providers](docs/custom-provider.md), and [packages](docs/packages.md)
-- [Sessions](docs/sessions.md), [compaction](docs/compaction.md), and [architecture](docs/architecture.md)
+- [SDK](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/sdk.md), [JSON mode](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/json.md), [RPC](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/rpc.md), and [ACP](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/acp.md)
+- [Python and markdown skills](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/skills.md), [MCP](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/mcp-integrations.md), and [extensions](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/extensions.md)
+- [Custom models](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/models.md), [custom providers](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/custom-provider.md), and [packages](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/packages.md)
+- [Sessions](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/sessions.md), [compaction](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/compaction.md), and [architecture](https://github.com/BaseModelAI/base-context/blob/v1.0.1/packages/coding-agent/docs/architecture.md)
 
 ## Trust and data
 
 The agent executes generated Python and project commands with your user permissions. Worker and kernel processes are **not a security sandbox**. Use an external sandbox for untrusted work and review skills and extensions before loading them.
 
-Retained history and artifacts can contain sensitive information. Compaction is not deletion or secret removal. Do not copy upstream credential stores. Telemetry and trace sharing are off by default and need explicit configuration. SSE is the default provider transport.
-
-## Credits and license
+Retained history and artifacts can contain sensitive information. Compaction is not deletion or secret removal. Do not copy upstream credential stores. Diagnostics stay local; there is no telemetry upload feature. SSE is the default provider transport.
 
-Base Context preserves the MIT-licensed work of **Prime Agent** and **Prime Intellect**, and the **Pi / pi-mono** lineage created by **Mario Zechner**. We also acknowledge Prime Intellect's separate **[PrimeRL](https://github.com/PrimeIntellect-ai/prime-rl)** project; it is not a direct dependency of this CLI.
+## License
 
-See [LICENSE](LICENSE), [contributing](https://github.com/BaseModelAI/base-context/blob/main/CONTRIBUTING.md), and [security](https://github.com/BaseModelAI/base-context/blob/main/SECURITY.md).
+MIT. See [LICENSE](LICENSE) and [NOTICE](NOTICE) for copyright and license terms.

package/dist/NOTICE

@@ -14,9 +14,4 @@ this fork is not an official Prime Intellect release.
 Prime Agent builds on Mario Zechner's Pi project:
 https://github.com/badlogic/pi-mono
 
-We also credit Prime Intellect's PrimeRL authors for their open-source
-reinforcement-learning work: https://github.com/PrimeIntellect-ai/prime-rl
-This acknowledgement does not imply that PrimeRL is a runtime dependency
-or that upstream authors endorse this fork.
-
 Third-party dependencies retain their own licenses and notices.

package/dist/base-context-runtime/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "base-context-runtime"
-version = "1.0.0"
+version = "1.0.2"
 description = "Kernel-side runtime shim for Base Context recursion"
 license = "MIT"
 license-files = ["LICENSE", "NOTICE"]

package/dist/base-context-runtime/src/rlm/__init__.py

@@ -178,16 +178,18 @@ async def list_subagents() -> list[RLMSubagent]:
     return [_subagent_from_payload(entry) for entry in entries]
 
 
-async def delete_subagent(target: str | RLMSubagent) -> RLMSubagent:
-    """Delete one running or retained direct child from the current parent session."""
-    if isinstance(target, RLMSubagent):
+async def delete_subagent(target: str | RLMSubagent | RLMSpawnHandle) -> RLMSubagent:
+    """Delete a direct child by spawn handle, subagent row, or child id/session name."""
+    if isinstance(target, (RLMSpawnHandle, RLMSubagent)):
         selector = target.rlm_child_id
     elif isinstance(target, str):
         selector = target.strip()
         if not selector:
             raise ValueError("target must not be empty")
     else:
-        raise TypeError(f"target must be str or RLMSubagent, got {type(target).__name__}")
+        raise TypeError(
+            f"target must be RLMSpawnHandle, RLMSubagent, or str, got {type(target).__name__}"
+        )
     payload = await host_request("rlm.delete_subagent", {"target": selector})
     return _subagent_from_payload(payload.get("subagent"), "rlm.delete_subagent")
 
@@ -257,7 +259,7 @@ class _RLMCallable:
     async def list_subagents(self) -> list[RLMSubagent]:
         return await list_subagents()
 
-    async def delete_subagent(self, target: str | RLMSubagent) -> RLMSubagent:
+    async def delete_subagent(self, target: str | RLMSubagent | RLMSpawnHandle) -> RLMSubagent:
         return await delete_subagent(target)
 
     async def __call__(self, prompt: str, **kwargs: Any) -> RLMSpawnHandle:

package/dist/base-context-runtime/test/test_subagent_registry.py

@@ -182,6 +182,33 @@ class RlmSubagentRegistryTest(unittest.TestCase):
             {"target": "sub-a1b2c3d4"},
         )
 
+    def test_deletes_subagent_by_spawn_handle(self) -> None:
+        handle = rlm_module.RLMSpawnHandle(
+            rlm_child_id="sub-a1b2c3d4",
+            name="api-reviewer",
+            session_dir=Path("/tmp/parent/sub-a1b2c3d4"),
+            model="deepseek/deepseek-v4-flash",
+        )
+        deleted_payload = {
+            "rlm_child_id": handle.rlm_child_id,
+            "active_session_id": None,
+            "session_id": None,
+            "session_name": handle.name,
+            "session_dir": str(handle.session_dir),
+            "status": "completed",
+        }
+        host_request = AsyncMock(return_value={"subagent": deleted_payload})
+
+        with patch.object(rlm_module, "host_request", host_request):
+            deleted = asyncio.run(rlm_module.rlm.delete_subagent(handle))
+
+        self.assertEqual(deleted.rlm_child_id, handle.rlm_child_id)
+        self.assertEqual(deleted.session_name, handle.name)
+        host_request.assert_awaited_once_with(
+            "rlm.delete_subagent",
+            {"target": "sub-a1b2c3d4"},
+        )
+
     def test_rejects_invalid_delete_response_and_target(self) -> None:
         host_request = AsyncMock(return_value={"subagent": {"status": "completed"}})
 
@@ -191,7 +218,7 @@ class RlmSubagentRegistryTest(unittest.TestCase):
 
         with self.assertRaisesRegex(ValueError, "target must not be empty"):
             asyncio.run(rlm_module.delete_subagent("   "))
-        with self.assertRaisesRegex(TypeError, "target must be str or RLMSubagent"):
+        with self.assertRaisesRegex(TypeError, "target must be RLMSpawnHandle, RLMSubagent, or str"):
             asyncio.run(rlm_module.delete_subagent(123))
 
     def test_rejects_invalid_registry_payload(self) -> None:

package/dist/base-context-runtime/uv.lock

@@ -367,7 +367,7 @@ wheels = [
 
 [[package]]
 name = "base-context-runtime"
-version = "1.0.0"
+version = "1.0.2"
 source = { editable = "." }
 dependencies = [
     { name = "mcp" },

package/dist/build-info.json

@@ -1,5 +1,5 @@
 {
-  "sourceCommit": "3b6e414e5d955f03bdebf866f39b85853a896d13",
+  "sourceCommit": "4192025d919683d61a402b666b4790e26118df53",
   "sourceDirty": false,
   "buildHost": {
     "platform": "linux",

package/dist/bundle/amazon-bedrock.js

@@ -0,0 +1,21 @@
+import { createRequire as __piBundleCreateRequire } from 'node:module'; const require = __piBundleCreateRequire(import.meta.url);
+import {
+  writeFileLogEntry
+} from "./chunk-HKINIBP4.js";
+import "./chunk-27QCJ2AZ.js";
+import "./chunk-5QCJ5DQU.js";
+import "./chunk-CRBVRZBV.js";
+import "./chunk-GRJFXLBM.js";
+import "./chunk-2T7M7VJ4.js";
+import "./chunk-DGNX7HQ3.js";
+import "./chunk-W46VZFI6.js";
+import "./chunk-BAWSWWEU.js";
+
+// dist/node/amazon-bedrock.js
+import { bedrockProviderModule, setLogSink } from "@ponythewhite/base-context-ai/bedrock-provider";
+setLogSink(writeFileLogEntry);
+var { streamBedrock, streamSimpleBedrock } = bedrockProviderModule;
+export {
+  streamBedrock,
+  streamSimpleBedrock
+};

package/dist/bundle/{anthropic-YY4S534V.js → anthropic-UV5GZDGU.js}

@@ -15,15 +15,15 @@ import {
   parseJsonWithRepair,
   parseStreamingJson
 } from "./chunk-2T7M7VJ4.js";
-import {
-  getEnvApiKey
-} from "./chunk-SC5KG5R4.js";
 import {
   adjustMaxTokensForThinking,
   buildBaseOptions,
   sanitizeSurrogates,
   transformMessages
-} from "./chunk-MZHTNNVA.js";
+} from "./chunk-4GXPNXOJ.js";
+import {
+  getEnvApiKey
+} from "./chunk-DGNX7HQ3.js";
 import {
   AssistantMessageEventStream,
   ProviderAttemptTracker,
@@ -36,7 +36,7 @@ import {
   streamFailureFromStopReason,
   streamFailureMessage,
   truncateRawPayload
-} from "./chunk-IW77OXFS.js";
+} from "./chunk-W46VZFI6.js";
 import "./chunk-BAWSWWEU.js";
 
 // ../../node_modules/@anthropic-ai/sdk/internal/tslib.mjs

package/dist/bundle/{azure-openai-responses-JRP6MIM3.js → azure-openai-responses-WSNPXLYC.js}

@@ -3,7 +3,7 @@ import {
   convertResponsesMessages,
   convertResponsesTools,
   processResponsesStream
-} from "./chunk-U7AUVG3B.js";
+} from "./chunk-VSIYFWYI.js";
 import "./chunk-NQB5DCFC.js";
 import {
   AzureOpenAI
@@ -12,12 +12,12 @@ import {
   headersToRecord
 } from "./chunk-CL7XFWST.js";
 import "./chunk-2T7M7VJ4.js";
-import {
-  getEnvApiKey
-} from "./chunk-SC5KG5R4.js";
 import {
   buildBaseOptions
-} from "./chunk-MZHTNNVA.js";
+} from "./chunk-4GXPNXOJ.js";
+import {
+  getEnvApiKey
+} from "./chunk-DGNX7HQ3.js";
 import {
   AssistantMessageEventStream,
   ProviderAttemptTracker,
@@ -25,7 +25,7 @@ import {
   formatStreamFailureMessage,
   recordStreamFailure,
   streamFailureFromStopReason
-} from "./chunk-IW77OXFS.js";
+} from "./chunk-W46VZFI6.js";
 import "./chunk-BAWSWWEU.js";
 
 // ../ai/src/providers/azure-openai-responses.ts

package/dist/bundle/{bundled-modules-P5HARVZA.js → bundled-modules-PWOGKUQQ.js}

@@ -6,7 +6,6 @@ import {
   AssistantMessageComponent,
   AuthStorage,
   BashExecutionComponent,
-  BorderedLoader,
   BranchSummaryMessageComponent,
   ClientPromptStashStore,
   CompactionSummaryMessageComponent,
@@ -52,10 +51,11 @@ import {
   oauth_exports,
   runPrintMode,
   runRpcMode
-} from "./chunk-BYVP7LWV.js";
+} from "./chunk-44PEHPOG.js";
 import {
   AgentSession,
   CURRENT_SESSION_VERSION,
+  CancellableLoader,
   CompactionCommittedError,
   Container,
   DAEMON_PROTOCOL_INFO,
@@ -69,13 +69,14 @@ import {
   ExtensionRunner,
   InferenceCoordinator,
   IpythonKernelProvisioner,
+  Loader,
   NativeRecoveryBudgetRefusal,
   SelectList,
   SessionManager,
+  Spacer,
+  Text,
   Theme,
-  VERSION,
   buildSessionContext,
-  build_exports,
   calculateContextTokens,
   collectEntriesForBranchSummary,
   compact,
@@ -102,7 +103,6 @@ import {
   formatSkillsForPrompt,
   generateBranchSummary,
   generateSummary,
-  getAgentDir,
   getAvailableThemes,
   getLanguageFromPath,
   getLastAssistantUsage,
@@ -133,9 +133,8 @@ import {
   renderDiff,
   serializeConversation,
   shouldCompact,
-  src_exports,
-  src_exports2,
-  src_exports3,
+  src_exports as src_exports2,
+  src_exports2 as src_exports3,
   stripFrontmatter,
   truncateHead,
   truncateLine,
@@ -144,26 +143,33 @@ import {
   withFileMutationQueue,
   wrapRegisteredTool,
   wrapRegisteredTools
-} from "./chunk-73DCPU3D.js";
+} from "./chunk-IADCPVEI.js";
 import "./chunk-CFNASUYO.js";
+import "./chunk-HKINIBP4.js";
+import {
+  VERSION,
+  build_exports,
+  getAgentDir,
+  src_exports
+} from "./chunk-27QCJ2AZ.js";
 import "./chunk-5QCJ5DQU.js";
 import {
   compile_exports,
   value_exports2 as value_exports
 } from "./chunk-CRBVRZBV.js";
-import "./chunk-A53O363M.js";
+import "./chunk-GRJFXLBM.js";
 import "./chunk-2T7M7VJ4.js";
 import "./chunk-UIRUOAIQ.js";
 import "./chunk-HXOBFFJX.js";
-import "./chunk-SC5KG5R4.js";
-import "./chunk-IW77OXFS.js";
+import "./chunk-DGNX7HQ3.js";
+import "./chunk-W46VZFI6.js";
 import {
   __export
 } from "./chunk-BAWSWWEU.js";
 
 // dist/index.js
-var index_exports = {};
-__export(index_exports, {
+var dist_exports = {};
+__export(dist_exports, {
   AgentMessageComponent: () => AgentMessageComponent,
   AgentSession: () => AgentSession,
   AgentSessionRuntime: () => AgentSessionRuntime,
@@ -302,6 +308,55 @@ __export(index_exports, {
   wrapRegisteredTools: () => wrapRegisteredTools
 });
 
+// dist/modes/interactive/components/bordered-loader.js
+var BorderedLoader = class extends Container {
+  loader;
+  cancellable;
+  signalController;
+  constructor(tui, theme, message, options) {
+    super();
+    this.cancellable = options?.cancellable ?? true;
+    const borderColor = (s) => theme.fg("border", s);
+    this.addChild(new DynamicBorder(borderColor));
+    if (this.cancellable) {
+      this.loader = new CancellableLoader(tui, (s) => theme.fg("muted", s), (s) => theme.fg("muted", s), message);
+    } else {
+      this.signalController = new AbortController();
+      this.loader = new Loader(tui, (s) => theme.fg("muted", s), (s) => theme.fg("muted", s), message);
+    }
+    this.addChild(this.loader);
+    if (this.cancellable) {
+      this.addChild(new Spacer(1));
+      this.addChild(new Text(keyHint("tui.select.cancel", "cancel"), 1, 0));
+    }
+    this.addChild(new Spacer(1));
+    this.addChild(new DynamicBorder(borderColor));
+  }
+  get signal() {
+    if (this.cancellable) {
+      return this.loader.signal;
+    }
+    return this.signalController?.signal ?? new AbortController().signal;
+  }
+  set onAbort(fn) {
+    if (this.cancellable) {
+      this.loader.onAbort = fn;
+    }
+  }
+  handleInput(data) {
+    if (this.cancellable) {
+      this.loader.handleInput(data);
+    }
+  }
+  dispose() {
+    if ("dispose" in this.loader && typeof this.loader.dispose === "function") {
+      this.loader.dispose();
+    } else if ("stop" in this.loader && typeof this.loader.stop === "function") {
+      this.loader.stop();
+    }
+  }
+};
+
 // dist/modes/interactive/components/show-images-selector.js
 var SHOW_IMAGES_SELECT_LIST_LAYOUT = {
   minPrimaryColumnWidth: 12,
@@ -384,12 +439,12 @@ var VIRTUAL_MODULES = {
   "@ponythewhite/base-context-tui": src_exports3,
   "@ponythewhite/base-context-ai": src_exports,
   "@ponythewhite/base-context-ai/oauth": oauth_exports,
-  "@ponythewhite/base-context": index_exports,
+  "@ponythewhite/base-context": dist_exports,
   "@mariozechner/pi-agent-core": src_exports2,
   "@mariozechner/pi-tui": src_exports3,
   "@mariozechner/pi-ai": src_exports,
   "@mariozechner/pi-ai/oauth": oauth_exports,
-  "@mariozechner/pi-coding-agent": index_exports
+  "@mariozechner/pi-coding-agent": dist_exports
 };
 export {
   VIRTUAL_MODULES