Closed
Conversation
…chain docs, README overhaul
… call uses 4096 default
…vation; debug logs to stderr
…nside Claude Code
Based on Claude Code source review: - Grep/Glob: output relative paths instead of absolute (saves tokens) - Grep: add --max-columns 500 to prevent base64/minified line bloat Glob task tokens: 2,191 → 761 (-65%).
…cking (v2.5.8) Based on Claude Code source review: - context.ts: memoize assembleInstructions per workingDir; cap git log at 2KB - webfetch.ts: 15-min session cache (50 entries) — no re-fetching same URL - edit.ts: normalize curly quotes before failing (reduces false edit errors) - read.ts: track partial reads; edit warns when file was only partially read
…e fix + e2e tests (v2.5.11)
…ost unit (v2.5.14)
…e summaries (v2.5.16)
…mitted on turn done (v2.5.17)
…fresh, E2E tests (v2.5.18) - Scrollback: commit full responses to Static immediately; show last-5-line preview in dynamic area so terminal scrollback buffer captures all output - User-Agent: centralize USER_AGENT in config.ts, apply to LLM client and WebFetch tool - Balance: re-fetch after every turn_done so status bar reflects real spend - Session cost: live spend shown in status bar (-$X.XXXX in yellow) - Type-ahead: queue input while agent is busy, auto-submit on turn_done - E2E test suite: 13 tests covering all core tools + session cost tracking (node:test, no extra deps)
- Live balance: compute display balance as fetchedBalance - sessionCost in real-time, no stale on-chain lag - zai/glm-5 pricing: switch from per-token ($1.0/$3.2M) to per-call ($0.001/call) matching actual BlockRun billing - Track LLM call count per turn, emit in usage event, accumulate in session stats - Show call count in session footer: "109,795 in / 300 calls · $0.30 session" - estimateCost: add perCall support for flat-rate models
- Prompt caching: add cache_control ephemeral markers on system prompt, tools, and penultimate message for anthropic/* models — cuts input token costs ~90% on cache hits - Add anthropic-beta: prompt-caching-2024-07-31 header automatically for Anthropic models - README: comprehensive Solana/Base payment docs — setup, funding sources, chain switching, USDC acquisition - README: add `runcode base` / `runcode solana` commands to reference table
- GLM-specific request tuning: temperature=0.8 default per official zai spec
- Thinking mode: auto-enable {"thinking": {"type": "enabled"}} for -thinking- model variants
- Add zai/glm-5-turbo to model picker with shortcut glm-turbo
- Add glm5/glm-turbo shortcuts to model resolver
…EADME with GLM-5/caching docs (v2.5.23)
…call count (v2.5.24)
…mmand-aware git/test/build filters (v2.5.26)
…ing permission prompt (v2.5.28)
- Updated .gitignore to include VS Code and distribution files. - Added exports for VS Code integration in package.json. - Expanded README with installation instructions for the VS Code extension. - Introduced new functions for plain-text banner lines and footer for non-terminal UIs. - Enhanced model management commands to support listing models and switching via index or name. - Updated StreamEvent types to include status updates for UI hosts.
1bcMax
reviewed
Apr 9, 2026
Member
1bcMax
left a comment
1bcMax
left a comment
There was a problem hiding this comment.
Thanks for the PR — good feature choices. /history, /delete, and the numbered model picker are all real user pain points. The headless session API is the right abstraction for multi-host support.
A few things before I can review further:
-
Remove
dist/from the PR. Compiled output shouldn't be committed — it'll be generated by CI. Please also removevscode-extension/package-lock.jsonandvscode-extension/out/if present. -
Split
extension.ts(782 lines) into smaller modules. Suggestion:webview-provider.ts,session-manager.ts, andui-renderer.ts. One file doing webview setup + session management + HTML rendering + balance tracking is too much to review or maintain. -
Version bump should be a separate commit (2.5.28 → 2.5.29), not mixed into the feature.
-
Can you explain the design decision behind the headless session API (
vscode-session.ts) — why decouple viagetUserInput+onEventcallbacks rather than importing the agent loop directly? Want to make sure we're aligned on the architecture before this ships.
Looking forward to the updated PR.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add a VS Code sidebar extension (vscode-extension/) that provides the full RunCode agent experience
inside VS Code — chat panel, tool execution, model switching, and all slash commands
event using local cost estimation (same approach as the Ink terminal UI), with RPC sync on turn
completion
I/O, enabling any host (VS Code, future web UI, etc.) to drive sessions via getUserInput + onEvent
via console.error in TTY), allowing VS Code and other non-terminal hosts to display the picker
stderr
Test plan