Trailing empty assistant 'stop' arriving after a successful 'yield'
revived the already-yielded subagent. AgentSession.agent_end maintenance
compared #assistantEndedWithSuccessfulYield(msg) against the trailing
empty-stop message — not the yield-bearing one — so the empty-stop
recovery path appended a retry reminder and scheduled agent.continue().
Track a sticky #yieldTerminationPending flag set when the yield tool
finishes without error and cleared on the next #promptWithMessage. The
agent_end routing extends the existing successful-yield branch: when the
flag is set, or the current message ended with yield, short-circuit
empty-stop / unexpected-stop / compaction continuations for the rest of
the run, so a successful yield is terminal regardless of trailing stops.
Fixes#3389
GitHub Copilot's /models response advertises supports.vision = true for
Claude/GPT chat models on every host, but only the canonical personal
endpoint (https://api.githubcopilot.com) actually accepts image inputs;
the business (api.business.githubcopilot.com) and enterprise
(copilot-api.{domain}) hosts respond '400 vision is not supported'.
snapcompact then injected rasterized transcript frames after compaction
and permanently broke every business-Copilot session.
- Catalog discovery (githubCopilotModelManagerOptions.mapModel) now
forces input=['text'] whenever the resolved baseUrl is not the
canonical personal-Copilot host, so the upstream's vision flag is
honoured only where it actually works.
- mergeDynamicModel honours the dynamic input value (instead of
OR-upgrading with the bundled reference) when the merged baseUrl
differs from the bundled one, so a bundled spec pinned to the
personal host can no longer taint a business-resolved merge.
- snapcompact-inline's canSendImages helper short-circuits the
rasterizer for any github-copilot model whose baseUrl is non-personal,
catching stale cached specs that still advertise vision.
- Helper isPersonalGitHubCopilotBaseUrl exported from
pi-catalog/wire/github-copilot so catalog and coding-agent share one
canonical check.
Regression coverage in github-copilot-model-limits.test.ts (vision
endpoint policy + full merge) and snapcompact-inline.test.ts (#3387
business/enterprise case).
Fixes#3387
With collapseCompactedHistory the live display fell into the LLM compaction
branch, which skips the firstKeptEntryId..compaction turns whenever an OpenAI
remote-compaction replacementHistory payload is present. That payload feeds the
provider only and is not rendered, so a remotely-compacted session showed just
the summary plus post-compaction rows, hiding recent turns that were visible
before. Emit the kept SessionEntry rows in transcript mode regardless. Adds a
regression.
The append fast-path opened the session file for sentinel comparison and
recompute without a guard, so a file unlinked/rotated between #refresh's
statSync and the sentinel read threw out of the 250ms poll timer (no catch),
risking a TUI crash. Treat sentinel read/recompute failures as non-appendable
and fall back to the guarded full reload. Adds a fake-timer regression.
The "skips snapcompact entirely" case asserted .rejects.toThrow() on
session.compact(), but after the maxFrames<1 skip the manual /compact path
falls through to the LLM summarizer whose outcome is provider/network
dependent (resolves when a summary lands, rejects only without
credentials/network). In a sandbox it resolves and blows the 5s default
timeout. Tolerate either outcome and pin only the deterministic skip
contract (snapcompact.compact not invoked + the kept-history notice), with
a generous explicit timeout. Addresses chatgpt-codex P2 review on #3249.
renderUsageReports (command-controller) carried the #3268 dedup contract
with no regression test; the PR's added CLI test asserts the opposite
(per-limit CLI rendering shows the note twice). Export renderUsageReports
and add a real regression through it: two accounts sharing one window group
render a provider-wide UsageReport.note once and an identical per-limit note
once. Verified failing on the pre-fix flatMap form (0 and 2 occurrences) and
passing on head (1 and 1).
When /settings (or the Extensions/Agents dashboard) is open and a tool
approval prompt fires, ExtensionUiController.showHookSelector swaps the
editor out of editorContainer for the HookSelectorComponent. On exit,
the overlay's done() called overlayHandle.hide() + setFocus(editor),
both pointing at the editor captured as preFocus when the overlay
opened — now no longer mounted. The visible approval prompt then sat
unreachable: Up/Down/Enter/Esc routed to the unmounted editor and only
Ctrl+C escaped (issue #3349).
SelectorController now exposes focusActiveEditorArea(), which restores
focus to editorContainer.children[0] (the live slot owner) or falls
back to the editor. Wired into showSettingsSelector, showExtensionsDashboard,
and showAgentsDashboard close paths after overlay.hide().
Tests: unit test verifying focusActiveEditorArea picks the live slot
owner; TUI overlay-focus regression pinning the post-fix contract plus
a 'pre-fix snapshot' test pinning the broken pre-fix behavior so the
restore-from-preFocus assumption can't silently change.
Fixes#3349
Converted bracketed non-image filesystem path pastes into session-local attachment references while preserving the existing image path flow.
Added regression coverage for editor routing and controller local file attachment behavior.
Fixes#3360
- Added an extractText override to pi-mnemopi remember paths so stored content and mined facts can use different text.
- Routed coding-agent mnemopi retention to store the full transcript while extracting only user-authored turns.
- Tightened deterministic Instruction extraction to require an explicit I/you subject.
Fixes#3372
Devin provider models (devin-agent) advertise reasoning: true but no
thinking.efforts metadata — Cascade selects effort by routing to sibling
model ids, not a wire param. getSupportedEfforts(model) therefore returns
[]. clampAutoThinkingEffort previously short-circuited that empty supported
list by returning the requested effort as-is, so the auto-thinking
classifier-resolved level (e.g. low) reached stream.ts:1163 where
requireSupportedEffort threw 'Thinking effort low is not supported by
devin/<id>. Supported efforts: '. In --print mode the user saw the error
text; in the TUI it was silently swallowed, producing the reported
'working then empty response' symptom.
Returns undefined when supported is empty so the result mirrors
clampThinkingLevelForModel's behavior on the same shape (the explicit
--thinking low / high paths already worked because of this). Updates
classifyDifficulty's return type to Effort | undefined and threads through
to the existing #applyAutoThinkingLevel undefined-effort early-return.
#applyAutoThinkingLevel also short-circuits the classifier call up front
for these models — there is no effort to pick.
Fixes#3356
The ask tool's "Other (type your own)" free-text input is a prompt-style
HookEditorComponent. The hook-style branch already called matchesAppFollowUp
(Ctrl+Q / Ctrl+Enter) so Windows Terminal users — which can't deliver a
distinct Ctrl+Enter (#1903 / fixed by #1905) — had a working chord on the
main editor, hook editors, and the agent dashboard. The prompt-style branch
did not, so Ctrl+Q was unbound and Ctrl+Enter fell through to Editor as a
newline that WT silently swallowed: pressing Ctrl+Enter did nothing.
#handlePromptStyleInput now checks matchesAppFollowUp before the rest of
the dispatch (mirroring #handleHookStyleInput), so plain Enter remains the
primary submit and the chord is a secondary submit for cross-terminal
muscle memory. The prompt-style hint now reads "enter or ctrl+q submit"
so the fallback is discoverable.
Fixes#3353
Stream fetched tool assets to disk under the existing download abort signal instead of passing the Response object to Bun.write. Remove partial files when a stalled body is aborted and cover completed plus stalled downloads with regression tests.
Fixes#3369
The welcome 'Tip:' line hardcoded #b48cff (label) and #9ccfff (body)
with an additional \x1b[2m dim on the body, ignoring the active theme.
On any light theme this dropped the body to ~1.5:1 contrast on a white
background (WCAG AA needs >=4.5:1), making the line effectively
invisible. Switching between light variants did not help because the
colors were not theme-derived.
renderWelcomeTip in packages/coding-agent/src/modes/components/welcome.ts
now paints the label through theme.fg('customMessageLabel', ...) and the
body through theme.fg('muted', ...), drops the manual dim, and wraps the
whole line with theme.italic(...). Both tokens are tuned per theme, so
the line stays vivid on dark backgrounds and readable on light ones
(e.g. light theme's customMessageLabel #7e57c2 = 5.21:1 on white).
A regression test pins the contract: dark/light themes must produce
different bytes for the same tip, and no manual \x1b[2m may remain.
Fixes#3337
Address P2 review: sanitizeText preserves \t which can create visual
holes in rendered output. Add .replace(/\t/g, ' ') to provider-wide
and per-limit notes in usage-report.ts and usage-cli.ts, matching the
TUI path which already uses replaceTabs().
Apply sanitizeText to provider notes in the shared usage-report
renderer and the CLI usage output, matching the sanitization added
to command-controller.ts. Prevents tabs/newlines/control characters
in provider notes from breaking terminal rendering.
can1357