Mirrored registry status from pre-wire session run-state transitions and required live session corroboration before a peer can sustain bare hub waits.
Fixes#8634
The online difficulty and unexpected-stop classifiers call the tiny/smol model with disableReasoning plus maxTokens=1024. On the openai-completions transport (LiteLLM), disableReasoning on a reasoning model is downgraded to the lowest reasoning effort, so omp still emits reasoning_effort. LiteLLM/Vertex translates that to an Anthropic thinking.budget_tokens of at least 1024, and max_tokens=1024 is not greater than the budget, so every classifier call 400s.
Give the online classifiers 4096 output tokens so the request clears a proxy-injected minimum thinking budget (and leaves room for the keyword). Local reasoning budgets are unchanged.
Fixes#8610
Only the post-commit end/session_compact fan-out is detached.
The start emit still waits so input during that yield lands
in the compaction queue, matching the existing comment.
Rolled partial file appends back to their pre-write size and marked malformed resumed sessions for an atomic rewrite.
Retried transient persistence failures from in-memory state and surfaced the first failure in the interactive TUI.
Fixes#8596
Skipped same-provider cross-model fallback candidates when the latest assistant turn contains signed or redacted Anthropic thinking.
Kept same-model retry available for transient failures and added regression coverage.
Fixes#8558
Prevented deterministic 400 errors for immutable thinking blocks from entering same-model retries or configured model fallback.
Added a signed-thinking session regression covering the terminal error and retry UI lifecycle.
Fixes#8558
A user-invoked /skill:<name> reaches the session as a user-attributed skill custom message (role custom, attribution user) whose expanded SKILL.md body is the task prompt. The auto-thinking gate in #promptWithMessage only accepted role === "user", so these turns skipped classifyDifficulty/applyAutoThinkingLevel and the effort stayed stuck on pending auto.
Broaden the gate to also accept user-invoked skill prompts via the now-exported isUserInvokedSkillPrompt helper; agent-originated and autoload skill injections stay excluded.
Fixes#8554
Address review: #checkpointState is now assigned in the pre-await section alongside the reminder append, so a model that calls rewind immediately after seeing the notice finds an active checkpoint instead of 'No active checkpoint'. The entry id is backfilled post-await once the checkpoint toolResult entry is persisted; #applyRewind runs on a later rewind turn, never before the backfill.
Move the per-request date/cwd line out of the system prompt into a
first-turn system-reminder so open-weight providers keep their tool-schema
prefix cache; the reminder refreshes itself at midnight. Closes#7404.
Generated with Codebuff 🤖
Co-Authored-By: Codebuff <noreply@codebuff.com>
Address review: move the reminder text into a static prompt file rendered via prompt.render (no inline prompt strings); append it to agent.state synchronously before any await in the message_end handler so the next provider call within the same tool loop always sees it; persist it via the interrupted-thinking pattern; locate the checkpoint entry by message identity since the reminder entry now follows it; collapse the rewind tool description to one line (the MUST-rewind rule lives in the notice).
Move the rewind instruction out of the permanent checkpoint tool result into a transient post-checkpoint notice that is branch-cut away on rewind; rewrite the rewind-report prompt forward-looking (no negation); collapse the rewind tool description to one line. Closes#8499.
A thinking-only stop still routes through the empty-stop handler
(isEmptyAssistantStop treats a non-actionable stop as empty) and bills
output tokens for its reasoning, so the billed-output branch would falsely
report that content was generated and dropped by a provider filter. Gate the
branch on a truly zero-block stop (content.length === 0) so thinking-only
stops keep the context/`/shake images` recovery hint.
Fixes#8511
An empty assistant `stop` that exhausts the retry cap always reported the
context/`/shake images` hint, even when the provider billed output tokens
for the turn. Billed output on a zero-block stop means content was generated
and then dropped downstream (a filter/refusal flattened to
`finish_reason: "stop"` by a proxy, or a lossy API translation), so the
images/context advice is actively misleading there.
Branch the capped `finalError` in `#handleEmptyAssistantStop` on
`assistantMessage.usage.output`: keep the context hint when nothing was
generated, and otherwise name the billed output-token count and point at a
provider-side filter/translation. Also log `outputTokens` alongside the
existing warning fields.
Fixes#8511
Reparent service-tier metadata before removing discarded empty turns. Persist a branch marker for content-bearing children so reload never selects the discarded subtree.
Fixes#5179
Empty assistant stops were reparented off the active leaf in memory only. The session loader rebuilds the active branch from the last physical journal entry, so a capped empty stop (or one killed mid-retry) resurfaced as the active leaf on reload. Added SessionManager.dropLeafEntry to physically remove the entry and rewrite the file, and route every empty-stop discard through it.
Fixes#5179
Kept the Bun event loop live across subagent yield drains and delayed parent result flushes. Added a timer-lifecycle regression for the idle flush.
Fixes#8462
can1357