ci: release packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@yaseenhq/echadron@0.30.0

Minor Changes

• #30 07b4780 Thanks @YaseenHQ! - Show a cache-expiry reminder when resuming a long-idle session or submitting after a long idle stretch.

• #26 a9b1250 Thanks @YaseenHQ! - Improve prompt-cache hit-rate accounting and display cache reads/writes in session usage. Add stable 64-character cache-key handling and configurable Anthropic cache retention (none, short, or long).

• #26 a9b1250 Thanks @YaseenHQ! - Run Echadron's interactive TUI, print mode, doctor, ACP, export, and provider commands on the native agent-core-v2 engine by default. Set ECHADRON_LEGACY_FLAG=1 (or KIMI_CODE_LEGACY_FLAG=1) to use the v1 compatibility path. Remove the dead v1 micro-compaction implementation while preserving historical replay records.

• #7 aef303d Thanks @YaseenHQ! - Make the fork-owned CLI launch as Echadron (echadron/chad/maker) with an isolated
  ~/.echadron data namespace, and add echadron update --models for a
  persistent, validator-aware models.dev catalog refresh. Echadron no longer
  invokes the upstream Kimi Code self-update path by default, and its install
  hook no longer renames or removes an existing kimi executable.

• #1 02606aa Thanks @YaseenHQ! - Add unified account (OAuth) and API-key login routes with Kimi Code, xAI, OpenAI Codex, known catalog providers, and custom registries. Browser and device-code login methods are available for the supported OAuth providers. /logout supports individual and clearly described credential bundles plus separately confirmed provider-configuration removal; the redundant /provider slash command is removed.

• #7 8ee637d Thanks @YaseenHQ! - Allow custom agent profiles to select a model for delegated work. Set model in an agent file, or use inherit to keep the caller's model.

• #1 02606aa Thanks @YaseenHQ! - Add the /tree slash command to browse earlier turns and fork from a selected point. Run /tree to open the turn picker, then press Ctrl-X to copy its highlighted turn without closing it.

• #8 8d820c9 Thanks @YaseenHQ! - Remove the 50 MB size limit on file uploads to the built-in server, so large attachments (for example in the web UI) no longer fail with an upload-too-large error. Uploads now stream to disk instead of being buffered in memory.

• #12 71f7552 Thanks @YaseenHQ! - Support Markdown-defined custom agents on agent-core.

• #12 71f7552 Thanks @YaseenHQ! - Add the /secondary_model slash command to configure the secondary model used by subagents.

Patch Changes

• #16 7ef6f0c Thanks @YaseenHQ! - Restore the last completed turn outcome in the activity view when resuming a session.

• #7 fb9a15d Thanks @YaseenHQ! - Add an internal bash parsing capability that turns shell command strings into syntax trees, in preparation for per-command permission analysis. No user-facing behavior change yet.

• #1 02606aa Thanks @YaseenHQ! - Fix terminal paste restoration, legacy Alt-symbol input, tab rendering, mixed line endings, and cursor cleanup on exit.

• #13 f2830f8 Thanks @YaseenHQ! - Fix request headers not being passed correctly on some requests.

• #1 02606aa Thanks @YaseenHQ! - Show transient request and compaction retries in the activity pane, report aggregate token/cache usage in the footer, and make file-tool paths clickable in supporting terminals.

• #7 aef303d Thanks @YaseenHQ! - Allow updating the subagent secondary model through the configuration API.

• #1 02606aa Thanks @YaseenHQ! - Add Ctrl-X as a shortcut for copying the last assistant response.

• #7 aef303d Thanks @YaseenHQ! - Align ACP with Echadron's live provider/model configuration: forward additional workspace roots, reload configuration after authentication and catalog changes, expose safe model context metadata, and stream context-window usage updates. Rename the terminal OAuth method and ACP identity to Echadron.

• #22 bbd3d57 Thanks @YaseenHQ! - Add an explicit experimental echadron acp-v2 entry point for the draft ACP protocol v2. The stable echadron acp command remains unchanged; the v2 bridge adds batch-capable transport, message IDs, unified tool-call updates, structured plan/state updates, and terminal OAuth metadata.

• #7 aef303d Thanks @YaseenHQ! - Make Echadron (echadron / chad / maker) the fork's user-facing identity while retaining the legacy storage interfaces for compatibility.

• #19 e788920 Thanks @YaseenHQ! - Add /bug as a shorthand alias for /feedback.

• #22 39934ed Thanks @YaseenHQ! - Upgrade Echadron's MCP clients to the July 28, 2026 stateless protocol with automatic negotiation and legacy fallback. Refresh connected tool catalogs when servers publish list changes, validate OAuth callback issuers, and retain SSE compatibility for older servers.

• #18 8b64c43 Thanks @YaseenHQ! - Start MCP connections in the background during agent construction while still waiting for readiness before the first model step.

• #20 975f3a3 Thanks @YaseenHQ! - Show concrete thinking-effort levels on live subagent cards and swarm progress.

• #17 5d294e7 Thanks @YaseenHQ! - Render v2 background-task completion notifications as status cards when replaying sessions.

• #16 0f934f1 Thanks @YaseenHQ! - Prevent long-running print-mode and background-task waits from overflowing the host timer limit and firing immediately.

• #1 d92ccad Thanks @YaseenHQ! - Suppress the Skipped refreshing managed:kimi-code: ... requires login warning when switching models. The refresh orchestrator now treats an unauthenticated managed provider as not-yet-logged-in rather than a refresh failure.

• #7 2419cc3 Thanks @YaseenHQ! - Keep editor scroll indicators readable without overflowing narrow terminals.

• #1 1889925 Thanks @YaseenHQ! - Listen for OAuth browser callbacks on both IPv4 (127.0.0.1) and IPv6 (::1) loopback addresses while keeping the localhost redirect URI. This fixes browser login failures on systems where localhost resolves to ::1 first (e.g., Codex and Anthropic PKCE flows).

• #16 7ef6f0c Thanks @YaseenHQ! - Keep compaction token accounting on the same full-request basis as live context measurements.

• #11 b4f4788 Thanks @YaseenHQ! - Fix sporadic "model is not configured" errors when starting kimi web, caused by the background provider-model refresh transiently clearing the model catalog while the first session was being created.

• #10 b6019f6 Thanks @YaseenHQ! - Preserve the assistant's partial output when a turn is interrupted with Esc, and remind the model that the previous turn was deliberately interrupted.

• #27 f65bb85 Thanks @YaseenHQ! - Prevent one session's subagent tool projection from changing builtin profiles in later sessions.

• #16 b0f4039 Thanks @YaseenHQ! - Allow explicitly OAuth-enabled MCP servers to use OAuth discovery alongside non-secret headers in the v2 connection manager.

• #7 aef303d Thanks @YaseenHQ! - Keep Models.dev adapter and endpoint provenance on imported providers, share the
  persisted catalog between update --models and the v2 server, and honor the
  Echadron home directory when resolving config and server paths.

• #2147 29783e4 Thanks @wbxl2000! - Show a quota consumption note after installing official plugins that bill against plan quota (such as Kimi Datasource).

• #2147 29783e4 Thanks @wbxl2000! - Show an update notice when a turn that used an outdated plugin ends and the Official Marketplace has a newer version; each new version is announced once. Run /plugins to install the latest version.

• #3 55e7b53 Thanks @YaseenHQ! - Preserve opaque OpenAI Responses compaction state across turns and automatically
  use /responses/compact when the active provider exposes that capability,
  falling back to Kimi's existing local summarizer when it does not.

• #7 b17ea1f Thanks @YaseenHQ! - Fail fast on quota/balance-exhausted HTTP 429 errors (e.g. Moonshot exceeded_current_quota_error, OpenAI insufficient_quota) instead of silently retrying for ~3 minutes. Transient rate-limit 429s keep the existing retry, backoff, and Retry-After behavior.

• #30 07b4780 Thanks @YaseenHQ! - Keep the managed Kimi subscription labeled as Kimi Code inside the Echadron host, identify ChatGPT Codex requests as Echadron, make device OAuth network waits and polling sleeps cancel immediately, expose model capabilities when agents choose between primary and secondary subagent models, and add v2 config deprecation guidance without breaking existing Echadron config files.

• #31 7b85759 Thanks @YaseenHQ! - Repair the release plan after the Echadron package rename and make npm publishing independent from optional native signing.

• #15 195f8d2 Thanks @YaseenHQ! - Keep builtin agent profile state isolated between sessions.

• #28 155adc4 Thanks @YaseenHQ! - Subagent UIs now show each subagent's bound model and thinking effort.

• #10 ad546bb Thanks @YaseenHQ! - Remove the blocking block/timeout wait from the TaskOutput tool so checking a background task can no longer stall the conversation; it now always returns an immediate snapshot, and completion still arrives via automatic notification.

• #8 3bbf786 Thanks @YaseenHQ! - Fix sessions missing from the session picker when their cached metadata predates the archived flag.

• #1 02606aa Thanks @YaseenHQ! - Replace decorative terminal emoji with restrained Unicode markers and semantic tool glyphs.

• #7 67a08cf Thanks @YaseenHQ! - web: Fix garbled line numbers in code blocks.

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

• #15 195f8d2 Thanks @YaseenHQ! - Improve upstream-compatible session isolation, MCP OAuth recovery, structured MCP results, and UTF-16 file reading.

• #26 a9b1250 Thanks @YaseenHQ! - Keep live sessions stable across plugin mutations: preserve MCP baselines, freeze agent-tool profile listings, and surface an explicit plugin-change reminder until the user reloads.

@moonshot-ai/agent-core@0.16.0

Minor Changes

• #7 8ee637d Thanks @YaseenHQ! - Allow custom agent profiles to select a model for delegated work. Set model in an agent file, or use inherit to keep the caller's model.

• #1 02606aa Thanks @YaseenHQ! - Add the /tree slash command to browse earlier turns and fork from a selected point. Run /tree to open the turn picker, then press Ctrl-X to copy its highlighted turn without closing it.

Patch Changes

• #13 f2830f8 Thanks @YaseenHQ! - Thread the host identity through the managed auth facades so OAuth token refreshes from inside the core carry the X-Msh-* device headers.

• #1 02606aa Thanks @YaseenHQ! - Show transient request and compaction retries in the activity pane, report aggregate token/cache usage in the footer, and make file-tool paths clickable in supporting terminals.

• #9 84680a4 Thanks @YaseenHQ! - Count validation-rejected tool calls toward the repeat breaker so reminders fire at 3/5/8 and the turn force-stops at 12.

• #26 a9b1250 Thanks @YaseenHQ! - Run Echadron's interactive TUI, print mode, doctor, ACP, export, and provider commands on the native agent-core-v2 engine by default. Set ECHADRON_LEGACY_FLAG=1 (or KIMI_CODE_LEGACY_FLAG=1) to use the v1 compatibility path. Remove the dead v1 micro-compaction implementation while preserving historical replay records.

• #22 39934ed Thanks @YaseenHQ! - Upgrade Echadron's MCP clients to the July 28, 2026 stateless protocol with automatic negotiation and legacy fallback. Refresh connected tool catalogs when servers publish list changes, validate OAuth callback issuers, and retain SSE compatibility for older servers.

• #7 aef303d Thanks @YaseenHQ! - Keep Models.dev adapter and endpoint provenance on imported providers, share the
  persisted catalog between update --models and the v2 server, and honor the
  Echadron home directory when resolving config and server paths.

• #3 55e7b53 Thanks @YaseenHQ! - Preserve opaque OpenAI Responses compaction state across turns and automatically
  use /responses/compact when the active provider exposes that capability,
  falling back to Kimi's existing local summarizer when it does not.

• #1 02606aa Thanks @YaseenHQ! - Add unified account (OAuth) and API-key login routes with Kimi Code, xAI, OpenAI Codex, known catalog providers, and custom registries. Browser and device-code login methods are available for the supported OAuth providers. /logout supports individual and clearly described credential bundles plus separately confirmed provider-configuration removal; the redundant /provider slash command is removed.

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

• Updated dependencies [02606aa, a9b1250, d92ccad, 1889925, f2830f8, 55e7b53, 02606aa, 07b4780, cf99ad9, a9b1250]:

  • @moonshot-ai/protocol@0.6.0
  • @moonshot-ai/kimi-code-oauth@0.3.0
  • @moonshot-ai/kosong@0.5.6

@moonshot-ai/agent-core-v2@0.3.0

Minor Changes

• #1 02606aa Thanks @YaseenHQ! - Add unified account (OAuth) and API-key login routes with Kimi Code, xAI, OpenAI Codex, known catalog providers, and custom registries. Browser and device-code login methods are available for the supported OAuth providers. /logout supports individual and clearly described credential bundles plus separately confirmed provider-configuration removal; the redundant /provider slash command is removed.

• #7 8ee637d Thanks @YaseenHQ! - Allow custom agent profiles to select a model for delegated work. Set model in an agent file, or use inherit to keep the caller's model.

• #13 f2830f8 Thanks @YaseenHQ! - Replace the bootstrap clientVersion with a required clientIdentity host identity object: the OAuth device-flow endpoints now send the full X-Msh-* device headers on every host, telemetry reads the client version from the same source, and the session export manifest gains an optional desktop version field.

Patch Changes

• #1 02606aa Thanks @YaseenHQ! - Show transient request and compaction retries in the activity pane, report aggregate token/cache usage in the footer, and make file-tool paths clickable in supporting terminals.

• #22 39934ed Thanks @YaseenHQ! - Upgrade Echadron's MCP clients to the July 28, 2026 stateless protocol with automatic negotiation and legacy fallback. Refresh connected tool catalogs when servers publish list changes, validate OAuth callback issuers, and retain SSE compatibility for older servers.

• #18 8b64c43 Thanks @YaseenHQ! - Start MCP connections in the background during agent construction while still waiting for readiness before the first model step.

• #16 0f934f1 Thanks @YaseenHQ! - Prevent long-running print-mode and background-task waits from overflowing the host timer limit and firing immediately.

• #2144 a77ee03 Thanks @sailist! - Let embedding hosts customize the agent's product name and reply-style guidance in the system prompt when starting the server.

• #7 aef303d Thanks @YaseenHQ! - Keep Models.dev adapter and endpoint provenance on imported providers, share the
  persisted catalog between update --models and the v2 server, and honor the
  Echadron home directory when resolving config and server paths.

• #3 55e7b53 Thanks @YaseenHQ! - Preserve opaque OpenAI Responses compaction state across turns and automatically
  use /responses/compact when the active provider exposes that capability,
  falling back to Kimi's existing local summarizer when it does not.

• #30 07b4780 Thanks @YaseenHQ! - Keep the managed Kimi subscription labeled as Kimi Code inside the Echadron host, identify ChatGPT Codex requests as Echadron, make device OAuth network waits and polling sleeps cancel immediately, expose model capabilities when agents choose between primary and secondary subagent models, and add v2 config deprecation guidance without breaking existing Echadron config files.

• #7 cf99ad9 Thanks @YaseenHQ! - Derive the /usage plan usage window labels and reset hints from structured usage data instead of preformatted text.

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

• #9 2512022 Thanks @YaseenHQ! - Count validation-rejected tool calls toward the repeat breaker so reminders fire at 3/5/8 and the turn force-stops at 12.

• #26 a9b1250 Thanks @YaseenHQ! - Keep live sessions stable across plugin mutations: preserve MCP baselines, freeze agent-tool profile listings, and surface an explicit plugin-change reminder until the user reloads.

• #26 a9b1250 Thanks @YaseenHQ! - Complete v2 session parity for historical forks, session-only workspace directories, and non-blocking MCP startup.

• Updated dependencies [02606aa, a9b1250, d92ccad, 1889925, f2830f8, 02606aa, 07b4780, cf99ad9, a9b1250]:

  • @moonshot-ai/protocol@0.6.0
  • @moonshot-ai/kimi-code-oauth@0.3.0

@moonshot-ai/kap-server@0.2.0

Minor Changes

• #13 f2830f8 Thanks @YaseenHQ! - Require a hostIdentity when starting the server and derive the default outbound identity headers (User-Agent + X-Msh-*) from it, replacing the hardcoded CLI fallback. The version option is renamed to serverVersion, and session export manifests now record the host product version — plus an optional desktop version for desktop exports — instead of the engine version.

Patch Changes

• #1 02606aa Thanks @YaseenHQ! - Show transient request and compaction retries in the activity pane, report aggregate token/cache usage in the footer, and make file-tool paths clickable in supporting terminals.

• #7 aef303d Thanks @YaseenHQ! - Keep Models.dev adapter and endpoint provenance on imported providers, share the
  persisted catalog between update --models and the v2 server, and honor the
  Echadron home directory when resolving config and server paths.

• #7 cf99ad9 Thanks @YaseenHQ! - Derive the /usage plan usage window labels and reset hints from structured usage data instead of preformatted text.

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

• #26 a9b1250 Thanks @YaseenHQ! - Keep live sessions stable across plugin mutations: preserve MCP baselines, freeze agent-tool profile listings, and surface an explicit plugin-change reminder until the user reloads.

• Updated dependencies [02606aa, a9b1250, 39934ed, 8b64c43, 0f934f1, d92ccad, 1889925, a77ee03, aef303d, f2830f8, 55e7b53, 02606aa, 07b4780, 8ee637d, cf99ad9, a9b1250, f2830f8, 2512022, a9b1250, a9b1250]:

  • @moonshot-ai/protocol@0.6.0
  • @moonshot-ai/agent-core-v2@0.3.0
  • @moonshot-ai/kimi-code-oauth@0.3.0

@moonshot-ai/kimi-code-sdk@0.15.0

Minor Changes

• #26 a9b1250 Thanks @YaseenHQ! - Run Echadron's interactive TUI, print mode, doctor, ACP, export, and provider commands on the native agent-core-v2 engine by default. Set ECHADRON_LEGACY_FLAG=1 (or KIMI_CODE_LEGACY_FLAG=1) to use the v1 compatibility path. Remove the dead v1 micro-compaction implementation while preserving historical replay records.

• #16 21369e3 Thanks @YaseenHQ! - Expose global MCP authorization status through the SDK, including persisted OAuth and bearer-token states.

• #1 02606aa Thanks @YaseenHQ! - Add unified account (OAuth) and API-key login routes with Kimi Code, xAI, OpenAI Codex, known catalog providers, and custom registries. Browser and device-code login methods are available for the supported OAuth providers. /logout supports individual and clearly described credential bundles plus separately confirmed provider-configuration removal; the redundant /provider slash command is removed.

• #13 f2830f8 Thanks @YaseenHQ! - Rename userAgentProduct to productName and require platform in the host identity options exposed by the SDK.

• #1 02606aa Thanks @YaseenHQ! - Add the /tree slash command to browse earlier turns and fork from a selected point. Run /tree to open the turn picker, then press Ctrl-X to copy its highlighted turn without closing it.

Patch Changes

• #26 a9b1250 Thanks @YaseenHQ! - Improve prompt-cache hit-rate accounting and display cache reads/writes in session usage. Add stable 64-character cache-key handling and configurable Anthropic cache retention (none, short, or long).

• #1 02606aa Thanks @YaseenHQ! - Show transient request and compaction retries in the activity pane, report aggregate token/cache usage in the footer, and make file-tool paths clickable in supporting terminals.

• #7 aef303d Thanks @YaseenHQ! - Keep Models.dev adapter and endpoint provenance on imported providers, share the
  persisted catalog between update --models and the v2 server, and honor the
  Echadron home directory when resolving config and server paths.

• #7 b17ea1f Thanks @YaseenHQ! - Fail fast on quota/balance-exhausted HTTP 429 errors (e.g. Moonshot exceeded_current_quota_error, OpenAI insufficient_quota) instead of silently retrying for ~3 minutes. Transient rate-limit 429s keep the existing retry, backoff, and Retry-After behavior.

• #30 07b4780 Thanks @YaseenHQ! - Keep the managed Kimi subscription labeled as Kimi Code inside the Echadron host, identify ChatGPT Codex requests as Echadron, make device OAuth network waits and polling sleeps cancel immediately, expose model capabilities when agents choose between primary and secondary subagent models, and add v2 config deprecation guidance without breaking existing Echadron config files.

• #28 155adc4 Thanks @YaseenHQ! - Subagent lifecycle events and background task info now carry the subagent's bound model and thinking effort.

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

• #26 a9b1250 Thanks @YaseenHQ! - Complete v2 session parity for historical forks, session-only workspace directories, and non-blocking MCP startup.

@moonshot-ai/kimi-code-oauth@0.3.0

Minor Changes

• #13 f2830f8 Thanks @YaseenHQ! - Rework the host identity type: rename userAgentProduct to productName and add a required platform field, so every host explicitly declares the X-Msh-Platform value it reports instead of silently inheriting the CLI's. OAuth requests now also send the product User-Agent (with the optional runtime suffix), so the OAuth host can tell client families and surfaces apart.

• #1 02606aa Thanks @YaseenHQ! - Add unified account (OAuth) and API-key login routes with Kimi Code, xAI, OpenAI Codex, known catalog providers, and custom registries. Browser and device-code login methods are available for the supported OAuth providers. /logout supports individual and clearly described credential bundles plus separately confirmed provider-configuration removal; the redundant /provider slash command is removed.

Patch Changes

• #26 a9b1250 Thanks @YaseenHQ! - Run Echadron's interactive TUI, print mode, doctor, ACP, export, and provider commands on the native agent-core-v2 engine by default. Set ECHADRON_LEGACY_FLAG=1 (or KIMI_CODE_LEGACY_FLAG=1) to use the v1 compatibility path. Remove the dead v1 micro-compaction implementation while preserving historical replay records.

• #1 d92ccad Thanks @YaseenHQ! - Suppress the Skipped refreshing managed:kimi-code: ... requires login warning when switching models. The refresh orchestrator now treats an unauthenticated managed provider as not-yet-logged-in rather than a refresh failure.

• #1 1889925 Thanks @YaseenHQ! - Listen for OAuth browser callbacks on both IPv4 (127.0.0.1) and IPv6 (::1) loopback addresses while keeping the localhost redirect URI. This fixes browser login failures on systems where localhost resolves to ::1 first (e.g., Codex and Anthropic PKCE flows).

• #30 07b4780 Thanks @YaseenHQ! - Keep the managed Kimi subscription labeled as Kimi Code inside the Echadron host, identify ChatGPT Codex requests as Echadron, make device OAuth network waits and polling sleeps cancel immediately, expose model capabilities when agents choose between primary and secondary subagent models, and add v2 config deprecation guidance without breaking existing Echadron config files.

• #7 cf99ad9 Thanks @YaseenHQ! - Derive the /usage plan usage window labels and reset hints from structured usage data instead of preformatted text.

@moonshot-ai/pi-tui@0.82.0

Minor Changes

• #26 a9b1250 Thanks @YaseenHQ! - Port the compatible pi-tui improvements through upstream 0.84.1: Unicode LaTeX Markdown rendering with width-aware transforms, terminal-safe image fallbacks and iTerm2 payload metadata, OSC 8-safe truncation, Unicode grapheme width fixes, latency-sensitive input rendering, and the corrected OSC 9;4 progress-clear sequence. The alternate-screen renderer/layout stack remains intentionally deferred because Echadron still uses its existing main-screen renderer.

Patch Changes

• #1 02606aa Thanks @YaseenHQ! - Synchronize terminal input, paste restoration, line-ending handling, tab rendering, and cursor cleanup fixes from upstream.

• #1 02606aa Thanks @YaseenHQ! - Sync the vendored fork version string to the upstream release it tracks.

• #7 2419cc3 Thanks @YaseenHQ! - Keep editor scroll indicators within narrow terminal widths while preserving their border color.

@moonshot-ai/protocol@0.6.0

Minor Changes

• #1 02606aa Thanks @YaseenHQ! - Show transient request and compaction retries in the activity pane, report aggregate token/cache usage in the footer, and make file-tool paths clickable in supporting terminals.

Patch Changes

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

echadron-code@0.6.6

Patch Changes

• Updated dependencies [a9b1250, 02606aa, a9b1250, 21369e3, aef303d, 02606aa, b17ea1f, 07b4780, f2830f8, 02606aa, 155adc4, a9b1250, a9b1250]:
  • @moonshot-ai/kimi-code-sdk@0.15.0
  • @moonshot-ai/migration-legacy@0.1.17

@moonshot-ai/acp-adapter@0.3.6

Patch Changes

• #7 aef303d Thanks @YaseenHQ! - Align ACP with Echadron's live provider/model configuration: forward additional workspace roots, reload configuration after authentication and catalog changes, expose safe model context metadata, and stream context-window usage updates. Rename the terminal OAuth method and ACP identity to Echadron.

• Updated dependencies [f2830f8, a9b1250, 02606aa, 84680a4, a9b1250, 39934ed, 21369e3, aef303d, 55e7b53, 02606aa, b17ea1f, 07b4780, 8ee637d, f2830f8, 02606aa, 155adc4, a9b1250, a9b1250]:

  • @moonshot-ai/agent-core@0.16.0
  • @moonshot-ai/kimi-code-sdk@0.15.0

@moonshot-ai/acp-server@0.0.1

Patch Changes

• Updated dependencies [a9b1250, 02606aa, a9b1250, aef303d, 21369e3, aef303d, 02606aa, b17ea1f, 07b4780, f2830f8, 02606aa, 155adc4, a9b1250, a9b1250]:
  • @moonshot-ai/kimi-code-sdk@0.15.0
  • @moonshot-ai/acp-adapter@0.3.6

@moonshot-ai/klient@0.1.1

Patch Changes

• #26 a9b1250 Thanks @YaseenHQ! - Port compatible upstream agent-core fixes for durable cold-session turn outcomes, unknown context-limit reporting, and MCP removal tombstones.

• Updated dependencies [02606aa, 39934ed, 8b64c43, 0f934f1, a77ee03, aef303d, 55e7b53, 02606aa, 07b4780, 8ee637d, cf99ad9, a9b1250, f2830f8, 2512022, a9b1250, a9b1250]:

  • @moonshot-ai/agent-core-v2@0.3.0

@moonshot-ai/kosong@0.5.6

Patch Changes

• #3 55e7b53 Thanks @YaseenHQ! - Preserve opaque OpenAI Responses compaction state across turns and automatically
  use /responses/compact when the active provider exposes that capability,
  falling back to Kimi's existing local summarizer when it does not.

• #1 02606aa Thanks @YaseenHQ! - Add unified account (OAuth) and API-key login routes with Kimi Code, xAI, OpenAI Codex, known catalog providers, and custom registries. Browser and device-code login methods are available for the supported OAuth providers. /logout supports individual and clearly described credential bundles plus separately confirmed provider-configuration removal; the redundant /provider slash command is removed.

@moonshot-ai/migration-legacy@0.1.17

Patch Changes

• Updated dependencies [f2830f8, 02606aa, 84680a4, a9b1250, 39934ed, aef303d, 55e7b53, 02606aa, 8ee637d, 02606aa, a9b1250]:
  • @moonshot-ai/agent-core@0.16.0