chatgpt-codex second-pass review on #3249: the previous helper folded
the 4k SUMMARY_TEXT_RESERVE into both the maxFrames cap math AND the
skip decision (return 0 when frameBudget < 0). That made any residual
headroom below 4k fall negative and force the LLM-summarizer fallback,
even though a text-only snapcompact archive (the 'text.length <= 2 *
edgeCap' short-circuit in planArchive) typically costs only a few
hundred tokens of summary lead-in and would have fit cleanly.
The two reserves now serve their own jobs:
- Skip iff 'baseTokens >= totalBudget' (kept-recent + non-message
already eats the entire window − reserve envelope). No reserve
fudge here; positive residual is always worth attempting.
- Cap reserve (4k) is applied ONLY to the maxFrames calculation so
the projection still passes once frames land. When the frame budget
goes negative under that reserve but residual headroom is positive,
the helper now returns maxFrames=1 instead of 0 so snapcompact's
frame-less planArchive branch can still produce a valid archive.
Updated regression test to pin the new contract directly: kept-recent
tuned for 1500 tokens of headroom (well below the 4k cap reserve), the
old helper returned 0 and skipped to the LLM summarizer, the new helper
invokes snapcompact with maxFrames=1.
chatgpt-codex review on #3249: the helper returned 0 when frameBudget
< FRAME_TOKEN_ESTIMATE, causing the caller to skip snapcompact entirely.
But snapcompact.planArchive has a 'text.length <= 2 * edgeCap' short-
circuit that produces a valid frames:[] archive when the discarded
history is small enough — and the projection charges 0 for that. Hard
return-0 blocked that opportunity, forcing the LLM summarizer fallback
in offline/no-credential sessions where the text-only path would have
landed cleanly.
#computeSnapcompactMaxFrames now distinguishes two near-full cases:
- frameBudget < 0 → return 0 (kept-recent already exhausted budget;
no text-only summary can fit either) → caller still skips outright.
- 0 ≤ frameBudget < FRAME_TOKEN_ESTIMATE → return 1 → snapcompact runs
and picks the frame-less planArchive branch automatically for small
discarded histories; the projection guard rejects any actual
frame-bearing archive that overflows.
Added regression test pinning maxFrames=1 (not 0) in the near-full
window case.
Snapcompact's bundled MAX_FRAMES_DEFAULT (80) × FRAME_TOKEN_ESTIMATE (5024)
≈ 402k tokens worth of frames. AgentSession was calling snapcompact.compact()
with no maxFrames override, so the post-render projection inside #runAuto
Compaction / compact() always overflowed the budget on any sub-1M-token
window (Claude Sonnet 4.5's 200k = 170k usable, the 80-frame projection
alone clears that 2.4×), looping the 'snapcompact could not bring the
context under the limit — using an LLM summary instead' warning on every
threshold tick.
AgentSession.#computeSnapcompactMaxFrames now sizes the frame cap from
the resolved budget — (window − reserve − non-message − kept-recent −
summary-text reserve) / FRAME_TOKEN_ESTIMATE, clamped to MAX_FRAMES_DEFAULT
— and threads it into snapcompact.compact() in both the auto-compaction
and manual /compact paths. When the kept-recent slice already exceeds the
budget, snapcompact is skipped outright instead of running just to be
rejected: the projection guard remains as a defensive check.
Fixes#3247
- Extracted common kernel and executor logic into `BaseKernel` and `executor-base` to eliminate duplicated implementations for Julia, Python, and Ruby.
- Migrated shared operational workflows--including session namespacing, environment filtering, and result mapping--to centralized backend helpers.
- Consolidated runtime discovery and resolution logic into a unified `runtime-env` utility module.
- Simplified language-specific modules by delegating subprocess lifecycle, IPC, and configuration management to the newly established base classes.
- Centralized Parallel API utilities and parsing logic into a single module.
- Exported constants and helper functions from `parallel.ts` to replace duplicated definitions in the search provider.
- Updated the search provider to leverage the unified `parseParallelSearchPayload` function with metadata parsing toggled off.
- Changed default evaluation backend configuration to only enable Python and JavaScript by default.
- Implemented dynamic tool parameter generation to hide Ruby and Julia from the model's schema when they are disabled in settings.
- Updated tool summary and field descriptions to reflect the currently enabled runtime backends.
- Implemented persistent execution backends for Ruby and Julia using dedicated kernel processes and NDJSON-based IPC.
- Integrated language-specific prelude environments, runtime path resolution, and security-focused environment variable filtering.
- Exposed configuration options, tool schema updates, and lifecycle management for seamless agent interaction with both languages.
- Added comprehensive integration tests and updated prompt documentation to support the new evaluation capabilities.
- Centralized draft state and image management by migrating fields from context to the CustomEditor component.
- Standardized transcript row construction by introducing shared helpers for background jobs, IRC traffic, and file mentions.
- Refactored redundant UI logic and helper functions into reusable utility modules to streamline message submission and component rendering.
- Standardized event handler types by consolidating lifecycle definitions into a shared module while maintaining public API stability.
- Removed the `readHashLines` setting to consolidate hashline display logic.
- Simplified `resolveFileDisplayMode` to derive hashline visibility solely from the active edit mode.
- Added automatic cleanup of the `readHashLines` key from existing configuration files.
- Implement `handleWheelAt` to restrict wheel scrolling to the settings pane.
- Update selection movement to support clamping when scrolling at boundaries.
- Add tests to verify boundary behavior and spatial constraints.
- Established shared `subprocess` infrastructure to standardize worker IPC, environment resolution, and error handling.
- Migrated mnemopi, speech-to-text, tiny-model, and TTS clients to utilize consolidated worker-client utilities.
- Implemented common runtime helpers for ONNX model loading, logging, and process readiness probing.
- Eliminated redundant local spawn logic and environment mapping across all inference worker clients.
- Updated system and tool prompts to present dedicated tools as preferred defaults rather than absolute prohibitions.
- Relaxed the hard-forbidding of shell equivalents for file operations, searching, and editing.
- Retained guidance on prioritizing tools for their gitignore semantics, structure, and line-anchoring capabilities.
- Introduced `selector-helpers.ts` to centralize reusable list, dashboard, and selection utilities.
- Refactored `AgentDashboard`, `ExtensionList`, `HistoryResultsList`, and `TreeList` to use standardized rendering and navigation helpers.
- Encapsulated viewport padding, scrolling logic, and key matching to reduce code duplication across TUI components.
- Maintained consistent scrollbar themes and keyboard behaviors while simplifying component-specific implementations.
- Extracted incremental byte-stream framing logic into a shared `MessageFramer` class.
- Replaced redundant, per-client implementations in `dap/client.ts` and `lsp/client.ts`.
- Centralized buffer management to avoid O(n^2) allocation patterns during large message bursts.
- Refined obfuscation logic to use granular, typed transformations instead of generic object traversal.
- Enforced an 8-character minimum for secret patterns and restricted redaction to user-authored content to prevent false positives.
- Preserved system prompts, tool schemas, and opaque remote replay data to maintain provider context and data integrity.
- Integrated protected snapshot exports with targeted redaction to safeguard sensitive information in shared sessions.
Re-added the isConfigured guard inside applyDefaultSettingOverrides so the
runtime override applied at RPC/ACP startup only fills holes — explicit
embedder/project/--config/global values now survive for task.isolation.{mode,
merge,commits}, task.{eager,batch,maxConcurrency,maxRecursionDepth,disabledAgents,
agentModelOverrides}, memory.backend, memories.enabled, advisor.{enabled,
subagents,syncBacklog,immuneTurns}, plus the RPC-only async.{enabled,maxJobs}
and bash.autoBackground.{enabled,thresholdMs}. The guard was originally added
for #2598 and had quietly regressed, so the override re-asserted the schema
default and clobbered every explicit value — matching the contract problem
#2828 already fixed for the todo paths.
Replaced the prior 'default-disables advisor' test (which codified the
clobbering as the contract) with a comprehensive 'honors explicit
host-defaulted settings' regression covering every contested path across
rpc/rpc-ui/acp.
Fixes#3207
- Introduced `web-palette.ts` to implement the collab-web pink/purple brand identity for HTML exports.
- Updated `generateThemeVars` to support a `palette` option, allowing users to choose between the brand-web aesthetic and a specific TUI theme.
- Configured public exports and the share-viewer script to default to the brand-web palette rather than inheriting the user's terminal theme.
- Refactored `AgentSession.exportToHtml` to align with the new branding defaults while allowing for per-export theme overrides.
- Adjusted `dark.json` background values to ensure better visual consistency across internal surfaces.
- Implemented `waitForSelector` and `waitForNavigation` methods in the browser tool API.
- Introduced per-operation fail-fast budget management with dynamic timeout clamping.
- Added validation for selector engines to reject unsupported Playwright-only platform features.
- Standardized error handling to provide descriptive, named timeouts for stalled browser operations.
- Updated model detection to exclude WebP format for Codex-based providers, which do not support it.
- Forced the image resize pipeline to encode to PNG or JPEG for incompatible models to resolve transmission errors.
- Added test coverage to verify that WebP images are re-encoded when using the Codex Responses backend.
Skipped optimistic replacement when a user message_start matches another recorded local submission, preserving the pending prompt bubble until its own expanded event arrives.
Added coverage for the queued-message drain race between startPendingSubmission and prompt dispatch.
Fixes#3199
Updated optimistic replay to track the replacement component handles created during transcript rebuilds, so expanded slash prompts still replace the raw replayed message.
Extended the regression test to cover the rebuild window called out in review.
Fixes#3199
Replaced raw optimistic slash-command transcript entries with the canonical user message emitted by AgentSession when prompt expansion changes the text.
Added coverage for prompt-template expansion reconciliation so the transcript keeps one expanded user message.
Fixes#3199
When the user set `providers.tinyModel` to a local key, `generateSessionTitle`
still raced local against the online `smol` path with a 10 s timeout and
silently fired the online request whenever the local worker returned `null`
(unknown key, model not downloaded, transformers.js failure). The online path
resolves the `smol` role through `priority.json` (haiku → flash → mini → …);
with an `OPENROUTER_API_KEY` picked up from env, that silently billed
OpenRouter without consent.
Drop the race entirely for local choices: honor the user's setting, log a
warning on local failure, leave the session untitled. The `raceFirstNonNull`
helper and `TITLE_LOCAL_FALLBACK_DELAY_MS` had no other consumer and are
removed; the obsolete \"silently bills online when local fails\" tests are
flipped into regressions that lock the no-fallback contract, including the
unknown-key path (e.g. \"ollama:gpt-oss\") which previously also leaked
straight through to the online billing path.
Fixes#3187
Mapped OpenCode MCP array commands to stdio command plus args and accepted environment as the provider-native env key.\n\nAdded regression coverage for array command normalization, environment mapping, env fallback, and empty args omission.\n\nFixes #3180
- Inlined the temporary model status formatting logic directly into the controller.
- Removed the unused `formatTemporaryModelStatus` utility function and its associated test.
- Update `shouldRetry` to treat `EISDIR` and `ENOTDIR` as terminal errors, preventing unnecessary retries when encountering Git reference directory conflicts.
- Add a test suite to verify graceful resolution of branches in scenarios where a packed ref conflicts with a directory path in the filesystem.
- Added `includeWorkspaceTree` configuration setting to optionally render the workspace directory tree.
- Configured the system prompt template and SDK to respect this toggle, allowing users to disable the tree to prevent prompt cache invalidation.
- Promoted `write` and `find` tools to `essential` status to ensure they are always available regardless of discovery mode.
- Updated `DEFAULT_ESSENTIAL_TOOL_NAMES` to include these tools by default.
- Updated documentation and tests to reflect the change in default essential tool availability.
Fixes#3165
- Synchronize tool arguments with the component state upon receipt of `tool_execution_start` to ensure visual consistency when final update events are missed.
- Terminate active argument reveal streams to prevent late ticks from overwriting valid, fully-materialized tool arguments with stale partial data.
- Add test coverage to verify that tool UI components render finalized arguments even in the absence of intermediate streaming updates.
- Implemented `tab.ariaSnapshot()` to capture and represent page structures as ARIA-tree YAML.
- Introduced `tab.ref()` and ref-based selector parsing to enable precise element interaction via unique ARIA identifiers.
- Integrated automated script bundling for cross-environment evaluation of ARIA snapshot logic.
- Updated browser action methods to resolve and target elements using ARIA-ref handles.
Tracked current-registry built-in provenance through AgentSession so plan mode
only force-activates the built-in write implementation. Extension or SDK tools
that shadow the name `write` stay inactive, preserving plan mode's read-only
contract through the built-in write/edit guard.
Added a regression that registers a shadowing write tool without built-in
provenance and verifies plan mode does not activate it.
Plan-mode entry only added `resolve` to the active toolset, so when
`tools.discoveryMode: "all"` left `write` hidden behind
`search_tool_bm25` the agent was stuck with `edit` to create the plan
file — which fails on a non-existent path and stalls the planning
turn. `#enterPlanMode` now augments the active set with both `resolve`
and `write` whenever the registry built them, matching what
`plan-mode-active.md` instructs the model to use, and `#exitPlanMode`
still restores the pre-plan toolset verbatim.
Fixes#3165
- Overhauled stealth spoofing mechanisms for WebGL, screen dimensions, Web workers, and iframe contexts using prototype-aware injection.
- Centralized function string representation patching to improve mimicry of native browser behavior across global objects.
- Patched puppeteer-core to remove detectable evaluation markers and implement lazy, pull-style execution context management.
- Enabled support for capturing LLM request JSON dumps and adjusted launcher flags to improve organic request patterns.
roboomp