When Anthropic Claude returns stopReason: "toolUse" but the assistant
message contains no actual toolCall, the session would append a spurious
toolResult entry. This creates tool_result blocks without matching
tool_use blocks — structurally invalid for Anthropic's API validator.
The symptom is overloaded_error on every subsequent request (in: 0 out: 0),
even though servers aren't overloaded. Other providers handle the same
corrupted history more leniently, making the problem appear Anthropic-specific.
The #isEmptyAssistantStop guard now checks for stopReason === "toolUse"
with no text and no toolCall. When the retry cap is hit, tool-use orphans
are still removed from active context (unlike regular empty stops) because
they corrupt message history. A regression test covers both the retry and
cap scenarios.
For affected sessions: run /compact to drop the corrupted history tail.
Fixes review feedback from chatgpt-codex-connector.
- Parsed zip metadata via central directory and lazy ranged reads.
- Inflated member contents only when a specific entry is read.
- Prevented large or corrupt zips from freezing directory reads.
- Added selectorLineRanges to extract ranges from raw/conflicts selectors.
- Routed internal URLs through URL-aware splitter in content search.
- Treated display-mode selectors as whole-resource searches instead of rejecting.
- Resolved @-mentions to exact existing paths only.
- Relied on the TUI @-selector to insert complete real paths.
- Dropped candidate scanning to avoid dragging in same-named files.
- Made `@`-mention resolution directory-aware so a mention ending in `/` or `\` matches only directory candidates.
- Stopped stripping the trailing separator and fuzzy-matching an arbitrary same-named file (e.g. npm scopes like `@scope/`).
- Left non-slash mentions unchanged.
- Skipped count/concurrency normalization when --bench is set.
- Errored when no OAuth accounts resolve for the provider.
- Updated flag docs to run one request per OAuth account.
- Added `getOAuthAccesses` to resolve each stored credential once.
- Sent one live request per account, reporting TTFT and TPS.
- Streamed per-account progress with interactive status lines.
If createAgentSession failed after installing a newly created AsyncJobManager but before AgentSession took ownership, the process-global singleton stayed installed. The new singleton guard then caused the next top-level session to skip constructing a scoped manager, disabling async bash/task support.
The startup-error cleanup now clears the singleton only when it still points at the newly created manager, disposes that manager, and then continues the existing registry/kernel cleanup. The regression test forces a startup failure after singleton installation and verifies the next top-level session can create and use its own async manager.
AgentSession now stores the same scoped AsyncJobManager reference that tools receive: owning top-level sessions use their constructed manager, subagents inherit the parent's manager, and secondary in-process top-level sessions get no manager when a singleton is already live.
getAsyncJobSnapshot and ACP delivery drains now use that scoped manager instead of AsyncJobManager.instance(), so secondary sessions cannot report or drain the primary session's background jobs. The regression test covers a secondary session created while the primary has a Main-owned running job.
- Appended newly sealed transcript blocks to native scrollback once.
- Deferred only the active live block during ED3-risk streaming.
- Hardened snapshot TTL parsing to handle whitespace-only env values.
- Added `NativeScrollbackLiveRegion` seam so components report the live suffix start.
- Stopped ED3-risk streaming from dropping sealed transcript rows above the live block.
- Appended newly sealed rows once while keeping the active tail deferred to checkpoint.
Per PR review on #1926: a secondary in-process top-level createAgentSession() that exposes bash/task/job tools would still call AsyncJobManager.instance() at execute time, register on the primary's manager, and have the primary's onJobComplete enqueue results into the primary's yieldQueue — corrupting the owning session's conversation.
ToolSession now carries an asyncJobManager reference scoped to its session: the constructed manager for top-level sessions, the inherited singleton for subagents (so their bash/task completions still flow into the spawning conversation as before), and undefined for secondary in-process top-level sessions that found a singleton already installed. bash, task, and job tools resolve the manager through ToolSession instead of the process-global singleton, so a secondary session whose tools attempt async work fails fast with the standard "Async job manager unavailable" error instead of contaminating the primary.
- Added AES-GCM cache for at-rest broker snapshots keyed on token, with URL as additional data.
- Added `onSnapshot` hook to RemoteAuthCredentialStore for persisting applied snapshots.
- Exposed cache read/write and TTL defaults through the coding-agent re-exports.
- Added `getAuthBrokerSnapshotCachePath` with `OMP_AUTH_BROKER_SNAPSHOT_CACHE` override.
Any in-process secondary createAgentSession() (e.g. the Agent Control Center's create flow in agent-dashboard.ts) was constructing its own AsyncJobManager, overwriting the process-global singleton, and then clearing it on its own dispose. The primary session still held its #ownedAsyncJobManager reference, but AsyncJobManager.instance() was undefined for the rest of the process — the task async path hard-failed with "Async execution is enabled but no async job manager is available" and only a full restart cleared it.
- sdk.ts: skip constructing/installing a second AsyncJobManager when a singleton is already live, so secondary top-level sessions share the owning session's manager instead of clobbering it.\n- agent-session.ts: scope #cancelOwnAsyncJobs so a secondary session inheriting the singleton with the default MAIN_AGENT_ID can no longer cancel the primary session's running bash/task jobs at dispose time. Subagents still reach the inherited singleton via their unique agent ids; the owning session still cancels its own jobs through #ownedAsyncJobManager.
Fixes#1923
- Added `omp dry-balance` command with model, count, concurrency, and JSON flags.
- Implemented random session-id sampling with bounded concurrency for OAuth access dry-run checks.
- Added success/failure summary generation with account and reason stats and optional JSON output.
- Set CLI exit status to 1 when any dry-balance attempt fails.
- Added the new dry-balance capability to the unreleased changelog notes.
When `async.enabled` is true but `AsyncJobManager.instance()` returns
`undefined` (orphaned-session state, host that never wired one up, etc.),
the `task` tool was returning a hard error and was unusable for the rest
of the session — even though the existing sync codepath (`#executeSync`,
which still parallelizes via `mapWithConcurrencyLimit`) was right there.
Fall back to `#executeSync` instead and emit a `logger.warn` so the
missing-manager state stays diagnosable. Background/job-poll semantics
are lost in this degraded mode, but the tool keeps working.
Fixes#1922
- Stopped leaking absolute home directory to the model in ttsr-interrupt and ttsr-tool-reminder blocks.
- Rendered rule paths as cwd-relative in-project, `~`-relative under home, else raw.
Added two regression tests covering the bidirectional reporter scenario:
clearing hindsight.bankId after a non-empty value rebuilds the live state
and routes subsequent retainBatch calls to the recomputed bank under both
per-project and global scoping. The PR's existing fix already handles
this direction; these lock the contract in place explicitly.
Fixes#1902
Reserved the first wrapped plain-text login URL rows above the manual-code prompt, while rendering the complete wrapped URL again below the prompt for terminals without OSC 8 support.
Fixes#1919
Hoisted an always-visible OSC8 'Browser login' row in the setup sign-in tab so wizard body clipping never hides both the clickable link and the focused manual-code prompt. The wrappable URL text still renders below for terminals without OSC 8 support.
Fixes#1919
Rendered manual-code prompts before wrapped OAuth status lines so wizard body clipping cannot hide the focused input behind a long login URL.
Fixes#1919
Kept explicit null package imports as exclusions so exact entries and active conditions do not fall through to wildcard or fallback targets.\n\nFixes #1889
Wrapped setup sign-in OAuth status lines instead of truncating them, and added a full OSC8 login link so narrow terminals still expose the complete URL.
Fixes#1919
The session-tree selector collapsed to "No entries found" whenever the
default filter rejected every entry. On a fresh session that is the
normal case: `sdk.ts` writes `model_change` + `thinking_level_change`
at startup so model + thinking state survive resumes, and the default
filter treats both as bookkeeping. `selector-controller.showTreeSelector`
guards on `tree.length === 0` so the selector still opens, and the user
sees an unexplained empty panel with `(0/0)` next to a "Recent sessions"
list that does contain data.
Split the empty-state branch into three shapes:
- `flatNodes.length === 0` → unchanged "No entries found".
- `searchQuery` non-empty → "No entries match search \"…\"" plus a
Backspace hint, with the real total in `(0/N)`.
- otherwise → "N entries hidden by the current filter [mode]" plus
"Press Alt+A to show all, Alt+D for default", with the real total in
`(0/N)`.
So the fresh-session case now explains why the panel is empty and how
to widen it instead of reading as "/tree is broken".
Fixes#1909
When an HTTP MCP server returns invalid_grant (or invalid_token / revoked /
plain 401 from the token endpoint) during OAuth refresh, MCPManager
previously logged "MCP OAuth refresh failed, using existing token" and
re-attached the stale access token as Authorization: Bearer on every
subsequent request. The next tool-load 401'd with invalid_token, future
sessions repeated the loop, and the only recovery was to hand-clear the
credential row in agent.db. Reported with Logfire as the trigger; any
remote HTTP MCP that rotates / revokes refresh tokens is affected.
#resolveAuthConfig now reuses pi-ai's isDefinitiveOAuthFailure classifier
(same one auth-broker and AuthStorage use for first-party providers): on
a definitive failure it calls AuthStorage.remove(credentialId), drops the
Bearer entirely, and the next request surfaces a clean auth error so the
user can /mcp reauth <server> (or /mcp unauth) to recover. Transient
failures (network/fetch failed/ECONNREFUSED) still fall back to the
existing token to ride out blips.
Verified with new mcp-manager-oauth-refresh.test.ts (invalid_grant, 401,
transient fallback, happy-path rotation). The full mcp-* test set
(45 tests across 5 files) still passes.
Fixes#1908
The github provider registered context-files (.github/copilot-instructions.md)
and instructions (.github/instructions/*.instructions.md), but no skills
capability — so .github/skills/<name>/SKILL.md, the layout GitHub documents
for Copilot Agent Skills, was silently never discovered. The skill://
URL resolved to 'Available: none' and nothing surfaced in the system prompt.
Register a skill capability on the github provider (priority 30, project-only)
pointing at .github/skills/ and reuse scanSkillsFromDir with
requireDescription: true to match the Agent Skills spec and the sibling
native/omp-plugins providers. Pin the wiring with a discovery test that
loads the skills capability scoped to the github provider against a temp
cwd containing a SKILL.md, and a negative case that drops a skill missing
a description.
Fixes#1906
Coalesced synchronous Hindsight routing setting hooks into one serialized
session rebuild so cwd reloads and multi-setting updates cannot have
multiple continuations capture the same old state and leak fresh
session listeners.
Also re-read the current state after awaiting the previous queue flush
before installing the replacement state, so an unexpected concurrent
owner cannot leave the actual current state undisposed.
Fixes#1902
DarkPhilosophy