blun-king-cli 9.1.536 → 9.1.561

package/CHANGELOG.md

@@ -1,321 +0,0 @@
-# Changelog
-
-## 9.1.536 - 2026-09-01
-
-- Keeps productive work alive across private Telegram DMs: direct work commands no longer stop at the reply boundary, and buffered DMs inject one bounded return step unless the user explicitly pauses or stops.
-- Preserves the active work turn when short status steers arrive.
-- Downgrades exhausted empty thinking retries instead of repeating the same effort.
-- Keeps more deferred tool schemas available and delays message offload until real history pressure.
-- Includes the verified TodoList progress policy and a one-line God Mode plus Swarm startup path.
-
-## 9.1.535 - 2026-09-01
-
-- Keeps the current real user instruction complete for its entire active turn instead of aging it out after a fixed number of history messages.
-- Pins all `Read` results produced during that turn so task files and source slices cannot be replaced by historical previews while the agent is still using them.
-- Preserves the existing bounded offload behavior for completed older turns and ignores injected system reminders when locating the active user turn.
-- Adds a packaged regression derived from the F35 live case where a complete 59-line task Read was archived after only four later history messages.
-
-## 9.1.534 - 2026-09-01
-
-- Separates DeepSeek V4's native `max` reasoning effort from the explicit BLUN Swarm action, so selecting Max no longer enables agents or injects the swarm context.
-- Keeps `/effort blun-swarm` as the deliberate session-only agent mode while showing native `max` as `Max` in the effort selector and session header.
-- Adds a mutation-backed packaged-bundle regression for the native Max path and preserved explicit Swarm path.
-
-## 9.1.533 - 2026-09-01
-
-- Adds a profile-gated, session-local model switch for Fredrik between King, GLM 5.3 Flash, and GLM 5.3 without changing the global default or any other King profile.
-- Pins the external provider to the Z.AI Coding Plan endpoint, reads its credential only from a validated profile-relative file, and sends always-on thinking with the explicit `low`, `high`, or `max` effort selected in the TUI.
-- Preserves the selected alias across normal prompts and resume while keeping new sessions on King, and rejects provider, alias, endpoint, effort, secret-path, and catalog widening mutations.
-- Adds focused provider, payload, picker, profile-isolation, secret-format, and mutation regressions to the package gate.
-
-## 9.1.532 - 2026-09-01
-
-- Makes automatic TodoList maintenance compare canonical visible state instead of treating every accepted rewrite as progress, while keeping truthful unchanged refreshes executable and the normal recovery tool catalogue available.
-- Preserves bounded work evidence across unchanged refreshes, asks for a concrete evidence-backed transition after two repeats, emits one non-blocking warning after three, and backs later maintenance off from 8 to at most 64 work calls.
-- Caps new and ordinary visible lists at 32 items, safely migrates oversized legacy lists only through explicit `done` pruning, and preserves bounded maintenance state across resume checkpoints.
-- Adds executable bundle, runtime, checkpoint, compatibility, and seven-mutation regressions derived from read-only fleet measurements of Fredrik, Werner, and Manfred.
-
-## 9.1.531 - 2026-09-01
-
-- Keeps nested `ProgrammaticTool` calls and results fully observable in the Wire while excluding their raw payloads from the next model context.
-- Applies the same isolation marker during replay so resume cannot reintroduce nested payloads into context.
-- Adds a mutation-backed regression for event marking, record preservation, live context exclusion, and resume reuse.
-
-## 9.1.530 - 2026-09-01
-
-- Packages the existing local semantic `CodebaseSearch` runtime and documentation instead of resolving to a source-tree-only path that does not exist after npm installation.
-- Resolves the bundled script relative to `blun.mjs`, preserves an explicit absolute `CODEBASE_INDEX_SCRIPT` override, and fails closed for missing or relative overrides.
-- Adds an executable package, resolver, Python-syntax, compatibility, and mutation regression without changing indexing, querying, permission, or model behavior.
-
-## 9.1.529 - 2026-09-01
-
-- Adds optional `response_format` support for foreground `Agent` calls, returning one canonical validated JSON envelope without changing the existing free-form result path.
-- Compiles a bounded object JSON Schema through BLUN's existing AJV validator before child creation, keeps the schema reminder inside that child, and grants no additional tools or permissions.
-- Allows exactly one cancellation-aware repair turn for invalid output, preserves the schema across automatic continuation, and rejects structured background or swarm execution until their persistence contracts are implemented separately.
-- Adds executable helper, bundled-agent, repair, cancellation, compatibility, and mutation regressions.
-
-## 9.1.528 - 2026-09-01
-
-- Adds a bounded QuickJS `ProgrammaticTool`, adapted from the verified DeepAgents interpreter pattern, so King can coordinate several sequential tool calls without returning every intermediate value to the model context.
-- Keeps every nested call inside BLUN's existing schema validation, preparation, authorization, permission, scheduler, event, and result pipeline; recursive interpreter calls and undeclared tools fail closed.
-- Separates the 120-second wall limit for real tools from the three-second guest-compute budget, propagates cancellation and turn-stop results, and exposes no host filesystem, network, process, package loader, clock, or dynamic evaluation API.
-- Adds 18 runtime tests plus executable bundle, pipeline, mutation, resource-limit, teardown, and host-isolation regressions.
-
-## 9.1.527 - 2026-09-01
-
-- Adds DeepAgents' `/copy` workflow to copy the latest completed, non-empty assistant response through BLUN's existing cross-platform clipboard path.
-- Keeps `/copy` available during active turns, where the existing command queue runs it after the response completes, and reports empty-history or clipboard failures in all six interface languages.
-- Adds a red-to-green regression for command registration, dispatch, response selection, byte-preserving clipboard content, and released UI text.
-
-## 9.1.526 - 2026-09-01
-
-- Materializes the AgentSpine version bundled with the current package before `/reload` rebuilds the plugin catalogue and MCP configuration.
-- Preserves unrelated plugin records and reuses an unchanged managed AgentSpine tree without rewriting `installed.json`.
-- Adds a red-to-green regression against 9.1.525 for bootstrap ordering, managed-plugin adoption, and idempotent reload behavior.
-
-## 9.1.525 - 2026-09-01
-
-- Starts every local shell command in its own process group on Windows as well as POSIX, so programs cannot attach to the BLUN TUI's controlling console and consume `Escape`, `Ctrl+S`, or other editor input.
-- Keeps stdin, stdout, and stderr on private pipes; interactive commands receive EOF and fail or time out without competing with the editor.
-- Adapts DeepAgents' local-shell terminal isolation to Windows and adds an executable regression for both platform branches plus a real captured-process probe.
-
-## 9.1.524 - 2026-09-01
-
-- Keeps the immediate prompt created by a Telegram `/loop` command inside the protected remote-command delivery path, so it cannot survive as a second plain queue item.
-- Shows each session's resume-checkpoint size, retained message and token counts, Wire size, and missing, invalid, or stale checkpoint state directly in the resume picker.
-- Updates the bundled AgentSpine runtime to the verified official 0.11.4 source at commit `8034487cdd24498076bac48d3f24400bca5e6ae4`, while preserving BLUN's 180-second background `SessionStart` completion budget.
-- Adds executable regressions for Telegram loop exactly-once delivery and all resume-metric states while preserving the existing local busy-loop queue behavior.
-
-## 9.1.523 - 2026-09-01
-
-- Makes `Escape` cancel an open slash-command menu and clear its slash draft in one action instead of leaving the command text behind.
-- Lets `Ctrl+S` pass a slash draft into the existing queue-and-drain path even when the surface currently reports idle, instead of silently returning before the flush.
-- Preserves ordinary drafts when `Escape` only dismisses non-slash autocomplete.
-- Adds red-to-green runtime regressions for both shortcuts against the exact 9.1.522 package behavior.
-
-## 9.1.522 - 2026-09-01
-
-- Writes `permission.approval_requested` to the session wire as soon as a tool approval opens, so an unattended King no longer looks like a silent model or provider stall while it is waiting for an operator.
-- Reuses the existing Telegram approval buttons automatically when the allowlist contains exactly one private chat plus any number of groups; ambiguous private recipients remain fail-closed unless `BLUN_TELEGRAM_APPROVAL_CHAT_ID` is set.
-- Adds a red-to-green regression for immediate wire visibility, unique-private-chat routing, explicit target precedence, and ambiguous-recipient rejection in both packaged relay copies.
-
-## 9.1.521 - 2026-09-01
-
-- Runs the AgentSpine `SessionStart` scan outside the interactive startup path, so a cold 123.6-second home-directory scan cannot hold the editor or prevent the first user prompt from reaching the wire.
-- Keeps successful SessionStart output automatic: the completed background hook still persists its scoped briefing into the session context.
-- Extends the cold-scan budget to 180 seconds and adds a red-to-green runtime regression proving that startup returns before a slow hook while the hook can still finish afterward.
-
-## 9.1.520 - 2026-08-31
-
-- Gives AgentSpine's `SessionStart` hook a 60-second cold-project budget, so startup and `/reload` can persist the automatic briefing even when the measured source scan exceeds the former 15-second limit.
-- Keeps prompt, tool, compact, stop, and subagent hook budgets unchanged; only the full SessionStart briefing receives the larger allowance.
-- Extends the SessionStart regression with the packaged AgentSpine manifests and a real hook-process output probe instead of relying only on a simulated result.
-
-## 9.1.519 - 2026-08-31
-
-- Persists successful `SessionStart` hook output before startup or reload returns, so AgentSpine readiness reaches the resumed model context automatically instead of waiting for the first user prompt.
-- Keeps a bounded head-and-tail preview in historical offloaded tool-result references, preserving recovery anchors without restoring the full result to every model request.
-- Adds red-to-green runtime regressions for SessionStart context delivery and historical preview retention while preserving the 9.1.518 Todo recovery catalogue.
-
-## 9.1.518 - 2026-08-31
-
-- Keeps `Read`, `Write`, `Edit`, and `Bash` available alongside `TodoList` during proactive task-list maintenance, preserving the normal recovery path for large file changes.
-- Retains automatic Todo maintenance and its bounded evidence prompt without switching the active model step to a TodoList-only tool catalogue.
-- Adds a red-to-green regression gate against the 9.1.499 catalogue change and verifies recovery-tool selection without duplicate schemas.
-
-## 9.1.517 - 2026-08-31
-
-- Keeps `Ctrl+S` and `Escape` queue controls available while an approval panel is open: `Ctrl+S` flushes all waiting messages, while `Escape` releases one waiting message before rejecting the approval when the queue is empty.
-- Observes asynchronous session cancellation in the keyboard, Telegram `/reload`, and slash-command paths, treating `session.not_found` as an already-stopped session instead of crashing the core with an unhandled rejection.
-- Adds red-to-green package gates for the raw Windows control bytes and the measured stale-session cancellation failure.
-
-## 9.1.516 - 2026-08-31
-
-- Caps the first tool-capable turn's wait for the initial MCP catalogue at 2.5 seconds, so a cold optional desktop server can finish in the background instead of blocking Read, Bash, Grep, TodoList, Telegram, and AgentSpine.
-- Verifies and installs AgentSpine only for the active profile at startup; dormant profiles are verified when they are launched, preventing startup time from growing with abandoned proof profiles.
-- Reuses the immutable packaged plugin source snapshot across managed-plugin checks while still reading and byte-verifying every active installed plugin tree, and logs each MCP server's startup status for reproducible timing evidence.
-- Retires an active-steer queue entry exactly once as soon as the core has durably accepted its `turn.steer`, instead of restoring an already delivered message after a 15-second UI confirmation timeout; genuine rejections remain queued.
-- Restarts an owned Telegram bridge after an unexpected exit or launch error with one deduplicated timer, while ownership loss and normal shutdown still prevent respawn.
-- Reconstructs a missing assistant step when a persisted `content.part` or tool call follows full compaction without its `step.begin`, preserving the continuation instead of making `king -c` fail during replay.
-- Allows only the first TodoList write in a model step, preventing parallel list replacements from racing while leaving read-only list queries untouched.
-- Extends the package gates for Ctrl+S, Escape, durable queue acknowledgement, rejection replay, stable urgent/direct/addressed/bot/normal priority, Todo lifecycle, visible pending markers, valid initial Todo state, and single-bridge recovery.
-
-## 9.1.515 - 2026-08-31
-
-- Accepts a truthful unchanged TodoList during mandatory maintenance instead of terminating the turn after five identical retries.
-- Lets an idle session loop resume unfinished Todo work instead of suppressing the loop because unfinished items still exist.
-- Adds a prepack regression gate for both failure modes while preserving the 9.1.514 Telegram hardening and the bundled AgentSpine 0.10.1 runtime.
-
-## 9.1.514 - 2026-08-31
-
-- Rejects nonnumeric Telegram message IDs and malformed precomputed chat hashes before they can enter the content-free delivery lifecycle journal.
-- Closes a fail-open diagnostic edge where a forged queue envelope could smuggle text through the field named `message_id`, even though normal Telegram message IDs are positive decimal numbers.
-- Adds a negative privacy regression and keeps the complete package suite green at 133 tests without changing Telegram routing, queueing, acknowledgement, or replay behavior.
-
-## 9.1.513 - 2026-08-31
-
-- Adds one bounded Telegram delivery lifecycle journal that connects bridge receipt, TUI injection, queue class, attempt, acceptance, commit or restore, quarantine, and durable acknowledgement by message ID.
-- Keeps the journal content-free and secret-safe: chat IDs are hashed, only allowlisted metadata is accepted, the active file rotates at 2 MiB, and one previous file is retained.
-- Removes full Telegram text, display text, origin, chat ID, and image path from quarantine diagnostics while preserving durable replay in the original append-only inbound queue.
-
-## 9.1.512 - 2026-08-31
-
-- Makes a proactive TodoList maintenance step set the same mandatory refresh state as the fallback policy gate, so an unchanged list can no longer reset the deadline and silently continue.
-- Gives the isolated Todo-only request at most eight compact tool-name/outcome facts since the last accepted refresh, without replaying prompts, arguments, commands, full paths, results, or work history.
-- Retains bounded evidence after a failed TodoList attempt, clears it after the accepted TodoList tool succeeds, and covers proactive enforcement, privacy, boundedness, retry, and tool restoration in the 127-test package suite.
-
-## 9.1.511 - 2026-08-31
-
-- Upgrades the managed core plugin from AgentSpine 0.8.0 to the cache-distinct 0.10.1 bundle with mandatory pre-answer receipts, Must-Remember, self-healing authenticated persona and relationship graph reconciliation, and a five-second local relationship deadline.
-- Race-safely verifies and identity-binds every active host instruction file without duplicating its already host-loaded text into the injected briefing; the real 15,519-byte user `CLAUDE.md` produced a 5,667-byte total injected context with no instruction content field.
-- Preserves the King-specific bounded Telegram-to-Mnemo recall rule, keeps all plugin state external to the package, and adds a package regression for compact BLUN delivery plus descriptor-only Codex delivery.
-
-## 9.1.510 - 2026-08-31
-
-- Adds `ReadBatch` for two to twelve independent known UTF-8 files or ranges in one concurrent tool call, reducing repeated model round trips over a large working context.
-- Reuses the exact `Read` path resolution, sensitive-file checks, permissions, UTF-8 validation, paging, and byte limits while preserving request order and reporting partial failures individually.
-- Keeps `ReadBatch` in the bounded core tool catalogue for main, coder, explore, and plan profiles, with focused concurrency, access, partial-failure, registration, and full-package regressions.
-
-## 9.1.509 - 2026-08-30
-
-- Requires every authorized bot-to-bot Telegram work message to mention its intended King explicitly.
-- Keeps unmentioned bot messages visible without buffering, queueing, delivering, or replaying them into that King's model context.
-- Prevents a priority bot message addressed to one King from becoming work for another King, with focused runtime and full-package regressions.
-
-## 9.1.508 - 2026-08-30
-
-- Keeps pending TodoList markers visible on reduced-color terminals and rejects a newly tracked multi-step list unless it has exactly one active item.
-- Captures accepted Telegram group and direct messages into Mnemo asynchronously with durable retry, exact deduplication, and per-agent private-chat scoping while excluding loop ticks and `/loop` controls.
-- Connects Telegram recall to AgentSpine's narrow scope only on demand with bounded result and byte budgets, and keeps every managed Telegram runtime import inside the installed plugin.
-
-## 9.1.507 - 2026-08-30
-
-- Instructs the Read tool to issue independent file and range reads together in one model response when several inputs may be useful.
-- Reduces avoidable model round trips over large work contexts without changing Read execution, paging, permissions, or result handling.
-- Adds a focused regression and keeps all TodoList, Telegram queue, bot-priority, tool-restoration, and package tests green.
-
-## 9.1.506 - 2026-08-30
-
-- Removes completed TodoList entries from the visible TUI immediately, matching the already-pruned persistent Todo store while retaining every unfinished item.
-- Runs automatic TodoList maintenance with a dedicated 546-character system prompt instead of appending the full work prompt measured at roughly 19,000 characters.
-- Preserves the full work prompt and ordinary tool catalogue on the following step, with regressions covering Todo lifecycle, maintenance isolation, and tool restoration.
-
-## 9.1.505 - 2026-08-30
-
-- Preserves Telegram messages that arrive after a TUI claims channel ownership but before the delayed channel activation finishes.
-- Captures the queue boundary at ownership start, keeps a valid durable checkpoint authoritative, and consumes a replaced queue from byte zero.
-- Adds a regression for the measured update handoff where message 63456 arrived after ownership began and was skipped one second later as historical backlog.
-
-## 9.1.504 - 2026-08-30
-
-- Starts an automatically queued Telegram message as a fresh model turn when King is idle instead of waiting forever for an already active turn.
-- Preserves one-at-a-time delivery, queue ordering, deferred acknowledgement, direct-message priority, and the existing active-turn steer path.
-- Adds a regression for the measured failure where an addressed group message reached the inbox and queue checkpoint but never entered the agent context.
-
-## 9.1.503 - 2026-08-30
-
-- Gives the isolated TodoList maintenance message the complete provider message shape with an explicit empty `toolCalls` array.
-- Prevents the real BLUN provider normalization path from aborting the maintenance turn with `TypeError: message.toolCalls is not iterable`.
-- Adds a regression that iterates the generated message exactly as the provider boundary does before accepting the Todo-only request.
-
-## 9.1.502 - 2026-08-30
-
-- Runs proactive TodoList maintenance against one isolated Todo-only request instead of replaying the old work-tool history into the restricted maintenance step.
-- Prevents the maintenance model from repeating a historical `Read`, `Bash`, or MCP call while only TodoList is exposed, eliminating the resulting red `Tool not found` detour.
-- Removes completed TodoList items from the persisted visible list immediately while retaining in-progress, pending, blocked, waiting, and aborted work.
-
-## 9.1.501 - 2026-08-30
-
-- Restores the ordinary work-tool catalogue immediately after a proactive TodoList-only maintenance step.
-- Prevents `Bash`, `Read`, and every other selected turn tool from remaining hidden after TodoList maintenance succeeds.
-- Adds a regression check for both normal and thinking-effort work steps after the Todo-only step.
-
-## 9.1.500 - 2026-08-30 - Golden Release
-
-- Marks the complete cross-platform BLUN King package as the Golden Release for Windows, Linux, and macOS.
-- Keeps the runtime behavior and verified TodoList, Telegram queue, session-loop, context-efficiency, AgentSpine, and managed vision changes from 9.1.499 byte-identical.
-- Gives the 9.1.500 heading a dedicated gold treatment in the public changelog without changing the styling of earlier releases.
-
-## 9.1.499 - 2026-08-30
-
-- Replaces image, audio, and video payloads older than the newest twelve conversation messages with short model-facing references while preserving the complete original session wire and every recent media attachment.
-- Reduces Fredrik's measured repeated model input by 85,748 characters, approximately 21,437 estimated tokens per subsequent work step, without changing the current image, message text, or persisted history.
-- Runs overdue TodoList maintenance as its own low-effort model step before the ninth work batch and exposes only TodoList during that step, preventing the normal workflow from hitting a visible red policy error first.
-- Keeps fresh user steers ahead of TodoList housekeeping while preserving the hard policy gate as a fallback for an actually invalid update.
-
-## 9.1.498 - 2026-08-30
-
-- Shows prompt-cache hit rate plus cache-read and cache-write tokens per model in `/usage`.
-- Adds `/tokens` as a short alias for the existing detailed context doctor without introducing another model call.
-- Adds `/offload` as an alias for the existing archive-backed context compaction command.
-- Preserves every distinct authorized Telegram bot message instead of dropping same-sender messages that arrive within a fixed 15-second window.
-- Preserves the exact TodoList through full compaction and adds a one-shot recovery reminder to read the archived history path before broad searches when an exact path, command, hash, or requirement is uncertain.
-- Keeps completed TodoList evidence append-only: later uncertainty becomes a separate in-progress step instead of silently reverting or removing completed work.
-- Defers recurring session-loop wakeups while the maintained TodoList still contains active or pending work, coalescing missed intervals until the assignment is genuinely idle without delaying unrelated scheduled reminders.
-
-## 9.1.497 - 2026-08-30
-
-- Prioritizes Telegram bot messages from shared groups behind direct and explicitly addressed traffic but ahead of ordinary queued context, without forcing a reply.
-- Retires an accepted queued Telegram message as soon as its steered model step starts, so delivered messages no longer remain visible until the whole run ends.
-- Allows due Telegram replies and message edits through TodoList housekeeping while keeping the next ordinary work action blocked until the TodoList is refreshed truthfully.
-
-## 9.1.496 - 2026-08-30
-
-- Delivers Telegram user messages automatically one at a time without depending on the manual Ctrl+S or Escape release counter.
-- Retires only the exact accepted or answered Telegram message ID, preserving newer messages from the same chat and removing delivered items from the visible queue.
-- Enforces truthful progress for every existing visible TodoList, including sessions without an active Goal, and rejects unchanged refreshes after extended work.
-- Hides successful internal hook payloads from visible output while keeping blocking hook failures visible.
-
-## 9.1.495 - 2026-08-30
-
-- Starts every new or replaced goal with a fresh visible TodoList and requires a truthful refresh after at most eight work batches.
-- Retires answered Telegram group focus immediately after a successful reply or edit instead of waiting for a long-running turn to end.
-- Defers normal inbound checkpoint acknowledgement until turn end, so restart replay cannot lose a message accepted during an early model step.
-- Adds real control-key decoding coverage for Ctrl+S and Ctrl+T, including Kitty keyboard sequences with Caps Lock active.
-
-## 9.1.494 - 2026-08-30
-
-- Removes the conflicting Ctrl+S fallback that opened or collapsed the Todo panel when no queue item was visible.
-- Keeps Todo expansion exclusively on Ctrl+T and queue-backlog release exclusively on Ctrl+S.
-
-## 9.1.493 - 2026-08-30
-
-- Automatically authorizes every new Telegram user message for delivery at the next safe handoff, including messages from groups, instead of requiring Ctrl+S or Escape first.
-- Keeps private DMs ahead of group traffic while preserving FIFO order inside each priority class.
-- Keeps Ctrl+S as an immediate full-backlog release and Escape as one-message-at-a-time release or active-run cancellation when no message remains.
-
-## 9.1.492 - 2026-08-30
-
-- Keeps an authorized Telegram group message pending when an in-flight delivery temporarily blocks Ctrl+S or Escape, then retries it without requiring another key press.
-- Prevents unreleased group arrivals from arming the delivery controller before the user authorizes them.
-- Makes a current changelog entry mandatory before npm can pack or publish the package.
-- Adds an external release check that compares npm latest, npm package integrity, the public update manifest, and the public changelog before a release may be reported complete.
-
-## 9.1.491 - 2026-08-30
-
-- Restores real-time private Telegram priority: a user DM now auto-authorizes delivery into active work instead of waiting behind the manual queue-release gate.
-- Keeps Ctrl+S batch release and one-message-per-Escape control for the remaining console and channel queue.
-
-## 9.1.490 - 2026-08-30
-
-- Corrects the packaged release history by recording the unpublished 9.1.487 build and replacing the incomplete 9.1.488 summary with the changes verified from the release commits.
-- Carries forward the TodoList, queue-control, Telegram-delivery, AgentSpine profile bootstrap, and managed RTX image-reading repairs from 9.1.489 without runtime changes.
-
-## 9.1.489 - 2026-08-30
-
-- Restores the visible TodoList as a core tool and requires active goals to keep it truthful from first action through completion.
-- Makes Ctrl+S release the current queue batch exactly once and Escape release one queued message per press before cancelling an empty-queue run.
-- Defers Telegram acknowledgement until real execution starts, restores unexpectedly cancelled pre-step work, and prevents duplicate queue execution or rendering.
-- Installs and enables the bundled AgentSpine plugin for every existing local BLUN profile, including Fredrik, Werner, and Manfred profiles on the same installation.
-- Routes image understanding through the managed BLUN media service on RTX infrastructure before any optional experimental local reader.
-
-## 9.1.488 - 2026-08-30
-
-- Published the managed AgentSpine core-plugin bundle prepared in the 9.1.487 build.
-- Advanced returning relationship status, kept telemetry disabled when no endpoint is configured, and preserved the selected reasoning effort across successful tool steps.
-- Cleared duplicate Telegram ghost turns and added regression coverage for Telegram intake, reload queues, session resume, and effort preservation.
-
-## 9.1.487 - 2026-08-30 (not published)
-
-- Prepared AgentSpine as a managed core plugin in the BLUN King package and installer.
-- This version was never published to npm; 9.1.488 was the next public package after 9.1.486.