Dev - #111
Closed
bobbyjohnstx wants to merge 19 commits into
Closed
Conversation
Some models (e.g., ornith:9b) append trailing newlines to tool call string arguments. The read tool rejected these with "File not found" because the path included a \n character. Adding .trim() to filePath parameters in read, edit, and write tools fixes this for all models. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Regenerate openapi.json from current API definitions (114 endpoints, version 1.16.3). Adds new fields from recent changes (mask_observations, model size config, server capacity settings). Add two new routes: - GET /api/openapi.json — serves the live OpenAPI 3.1.0 spec - GET /api/docs — serves Scalar API reference UI (single HTML page loading from CDN, pointing at /api/openapi.json) Scalar is a modern, lightweight API docs viewer that supports OpenAPI 3.1 natively with search, try-it-out, and dark mode. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
user-guide.md (894 lines): Complete beginner-to-power-user manual covering TUI shortcuts, sessions, 18 agents with tool scopes, skills with parameter substitution, model selection/compatibility, config, web/desktop UI, advanced multi-agent workflows, plugin system, performance tuning, and troubleshooting. Cross-references existing docs without replacing them. deployment-verification.md (880 lines): Ops-focused companion to /tc-doctor. Explains each of the 14 diagnostic checks in depth with exact commands, expected good/bad output, and fix procedures. Covers Ollama, ramalama, vLLM, container/K8s deployment, Mac-specific checks, performance baselines, and log locations. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add four copy-paste-ready workflows (debug a failing test, plan a feature, review a PR, explore a codebase) from the website's "First 5 Minutes" section. Placed between Quick Start and TUI deep-dive as the bridge from install to productive use. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Base system prompt (tool-pacing, bash format, anti-fabrication) is now always included, with agent prompts layered on top. Previously, persona agents replaced the entire base prompt, dropping critical tool-calling infrastructure that local models rely on. Closes: #85 Constraint: local 9B models need "use ONE tool per response" from base prompt Rejected: duplicating tool-pacing into each agent .md | fragile, 24 files to sync Confidence: high Scope-risk: narrow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…mpts
Cloud base prompts (default, anthropic, gpt, beast, gemini, kimi, codex,
trinity) stripped to tool infrastructure only (~13 lines each). Personality
and style rules (brevity, conventions, proactiveness) extracted into new
build.txt agent prompt used by build and plan agents.
Fixes conflict where base prompt brevity rules ("fewer than 4 lines",
"one word answers best") suppressed thorough output from agents like
architect and critic after the composition fix in 2e1c931.
Constraint: build/plan agents had no prompt field — relied on base prompt for personality
Rejected: per-provider personality variants | unnecessary maintenance burden
Confidence: high
Scope-risk: moderate
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Session child/parent navigation now uses ctrl+x + hjkl consistently, removing bare arrow key bindings that conflicted with macOS window management and terminal cursor movement. Closes: #88 Constraint: ctrl+x j (first child) already used leader-key pattern Rejected: keeping arrows as alternates | still conflicts with OS Confidence: high Scope-risk: narrow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Command palette (ctrl+p) now searches across commands, agents, sessions, and skills with frecency ranking. Fix crash when opening palette during model processing — FrecencyProvider was nested inside DialogProvider, so dialog content couldn't access the context. Also fix ErrorComponent double-fault from invalid URL constructor. Constraint: SolidJS context only flows to descendants, not siblings Rejected: defensive try/catch in useFrecency | masks real provider bugs Confidence: high Scope-risk: moderate Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Regex now matches decimal parameter counts like "3.5b" and "1.5b" in model IDs. Changed parseInt to parseFloat so values like 3.5 are preserved instead of truncated. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ARG TARGETARCH was declared after the build stages that reference it,
so the dynamic FROM build-${TARGETARCH} stage selection failed during
multi-platform buildx.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The event consumption loop was fire-and-forget — the function returned after the prompt/command API call completed but before SSE events were fully delivered. Now awaits the loop so all events (tool output, text responses, errors) are processed before exit. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The Task component now renders the subagent's final text as a collapsible block below the task header. Defaults to 3-line preview, click to expand. Navigate-to-child-session on header click preserved. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The publish script runs docker buildx with --platform linux/amd64,linux/arm64 but the default Docker driver doesn't support multi-platform builds. Add QEMU, buildx, and GHCR login steps before the publish step. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Update help, cheatsheet, user guide, CLAUDE.md, changelog, and UI-UX review to reflect: unified command palette (ctrl+p), vim-style hjkl session navigation, and reassigned keybindings (<leader>o for session list, <leader>; for code concealment). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…scovery - Add OpenRouter to bundled catalog (models-local.json) so it activates via OPENROUTER_API_KEY even without remote models.dev catalog - Add probeOpenRouter() in local-discovery.ts that fetches models from OpenRouter API, filtering to tool-capable non-free models with full capability/pricing/limit mapping (319 models discovered) - Add OpenRouter to /connect TUI dialog with priority and description - Fix 7 invalid JSON Schema enum definitions in omt tools.ts — nested arrays [["a","b"]] replaced with flat ["a","b"] (broke OpenAI strict schema validation) Constraint: env var activation only — /connect stores credential but discovery requires OPENROUTER_API_KEY in environment Rejected: custom() discoverModels hook | dead code, never wired into list() Confidence: high Scope-risk: moderate Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Removes OmtPlugin wrapper and registers omt tools directly in the tool registry. System prompt injection (notepad priority, project memory) now runs as native Effect code in both the agent layer and LLM request prep, eliminating the plugin hook indirection. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Contributor
Dependency Audit FailedPlease review and address HIGH/CRITICAL vulnerabilities. |
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
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.
Summary
Brief description of the changes.
Type
Checklist
bun run lintpassescd packages/tinycode && bun typecheckpassescd packages/tinycode && bun testpassesconsole.logor debug statementsTest Plan
How were these changes tested?