Adds internal/provider/subprocess — a provider.Provider that spawns CLI
agents (claude, gemini, vibe) as subprocesses and streams their output.
- FormatParser interface + three parsers for claude-stream-json,
gemini-stream-json, and vibe-streaming formats; fixtures captured from
real binaries
- subprocessStream: pull-based stream.Stream over subprocess stdout with
bounded stderr capture (8KB) and guarded reap() to prevent double-Wait
- DiscoverCLIAgents: parallel PATH scan with 10s timeout, stable ordering
- Provider: only the last user message is passed as --prompt; all other
request fields (history, tools, system prompt) are intentionally ignored
(see package doc)
- main.go: discover and register CLI arms at startup; TODO(P0c) for
tier-based routing to enforce preference order explicitly
Three compounding bugs prevented tool calling with llama.cpp:
- Stream parser set argsComplete on partial JSON (e.g. "{"), dropping
subsequent argument deltas — fix: use json.Valid to detect completeness
- Missing tool_choice default — llama.cpp needs explicit "auto" to
activate its GBNF grammar constraint; now set when tools are present
- Tool names in history used internal format (fs.ls) while definitions
used API format (fs_ls) — now re-sanitized in translateMessage
Additional changes:
- Disable SDK retries for local providers (500s are deterministic)
- Dynamic capability probing via /props (llama.cpp) and /api/show
(Ollama), replacing hardcoded model prefix list
- Engine respects forced arm ToolUse capability when router is active
- Bundled /init skill with Go template blocks, context-aware for local
vs cloud models, deduplication rules against CLAUDE.md
- Tool result compaction for local models — previous round results
replaced with size markers to stay within small context windows
- Text-only fallback when tool-parse errors occur on local models
- "text-only" TUI indicator when model lacks tool support
- Session ResetError for retry after stream failures
- AllowedTools per-turn filtering in engine buildRequest
The discovery loop's reconcileArms removed the CLI-forced arm
(llamacpp/default) because the llama.cpp server reports the real model
name (e.g. gemma-26b), creating a mismatch. After 30s the forced arm
disappeared and all subsequent requests failed.
Three-layer fix:
- Eager: query the specific provider at startup to resolve the real
model name before registering the forced arm
- Lazy: reconcileArms detects placeholder "default" arm names and
atomically renames them when discovery reveals the real identity,
with an onReconcile callback to update the session and TUI
- Guard: the forced arm is never garbage-collected by the removal loop
Also fixes misleading /init error messaging — failed inits now show
"loaded from disk (init failed)" instead of "AGENTS.md written to".
- Fix append footgun: allHooks/allMCPServers allocated fresh to avoid
mutating cfg's backing array (lines 391/413 in main.go)
- Fix pipe-mode permission prompt: detect no-TTY stdin and auto-deny
instead of blocking forever on fmt.Scanln EOF
- Tighten Mistral API key regex from bare [a-zA-Z0-9]{32} (matched
commit hashes, UUIDs) to context-gated pattern requiring "mistral"
keyword nearby. Added scanner test for positives and negatives.
- Remove README demo GIF TODO placeholder
- Unify version string: pass buildVersion from ldflags into tui.Config
instead of hardcoding "v0.1.0-dev"
- Populate benchmarks doc with actual Go benchmark results
Complete the remaining M8 extensibility deliverables:
- MCP client with JSON-RPC 2.0 over stdio transport, protocol
lifecycle (initialize/tools-list/tools-call), and process group
management for clean shutdown
- MCP tool adapter implementing tool.Tool with mcp__{server}__{tool}
naming convention and replace_default for swapping built-in tools
- MCP manager for multi-server orchestration with parallel startup,
tool discovery, and registry integration
- Plugin system with plugin.json manifest (name/version/capabilities),
directory-based discovery (global + project scopes with precedence),
loader that merges skills/hooks/MCP configs into existing registries,
and install/uninstall/list lifecycle manager
- Config additions: MCPServerConfig, PluginsSection with opt-in/opt-out
enabled/disabled resolution
- TUI /plugins command for listing installed plugins
- 54 tests across internal/mcp and internal/plugin packages
- store: validate session ID against store root to block path traversal in Load/Save
- local: seed turnCount from LocalConfig.TurnCount so resumed sessions keep correct turn count
- main: pass TurnCount from snapshot to LocalConfig on resume
- main: suppress quality.json save when --incognito is active
- main: handle UserConfigDir error in quality save defer instead of silently using wrong path
- test: add TestSessionStore_Load/Save_RejectsPathTraversal
- Add --resume/-r flags; empty = list sessions, ID = restore specific session
- Create SessionStore from config.ProjectRoot() and cfg.Session.MaxKeep
- Wire SessionID and Store into session.NewLocal
- Restore QualityTracker EMA data from ~/.config/gnoma/quality.json at startup
- Persist QualityTracker data to quality.json via defer on process exit
Refactor NewLocal to accept LocalConfig (matching engine/router patterns),
add persistence fields (SessionID, Store, Incognito, Logger), capture
finalState before releasing the lock to avoid data races, and auto-save
a Snapshot after each successful turn when a store is configured.
Add SessionID() to the Session interface and three new tests covering
auto-save, no-store no-panic, and SessionID accessors.
provider/openai:
- Fix doubled tool call args (argsComplete flag): Ollama sends complete
args in the first streaming chunk then repeats them as delta, causing
doubled JSON and 400 errors in elfs
- Handle fs: prefix (gemma4 uses fs:grep instead of fs.grep)
- Add Reasoning field support for Ollama thinking output
cmd/gnoma:
- Early TTY detection so logger is created with correct destination
before any component gets a reference to it (fixes slog WARN bleed
into TUI textarea)
permission:
- Exempt spawn_elfs and agent tools from safety scanner: elf prompt
text may legitimately mention .env/.ssh/credentials patterns and
should not be blocked
tui/app:
- /init retry chain: no-tool-calls → spawn_elfs nudge → write nudge
(ask for plain text output) → TUI fallback write from streamBuf
- looksLikeAgentsMD + extractMarkdownDoc: validate and clean fallback
content before writing (reject refusals, strip narrative preambles)
- Collapse thinking output to 3 lines; ctrl+o to expand (live stream
and committed messages)
- Stream-level filter for model pseudo-tool-call blocks: suppresses
<<tool_code>>...</tool_code>> and <<function_call>>...<tool_call|>
from entering streamBuf across chunk boundaries
- sanitizeAssistantText regex covers both block formats
- Reset streamFilterClose at every turn start
Gap 11 (M6): Fixed context prefix
- Window.PrefixMessages stores immutable docs (CLAUDE.md, .gnoma/GNOMA.md)
- Prefix stripped before compaction, prepended after — survives all compaction
- AllMessages() returns prefix + history for provider requests
- main.go loads CLAUDE.md and .gnoma/GNOMA.md at startup as prefix
Gap 12 (M6): Deferred tool loading
- DeferrableTool optional interface: ShouldDefer() bool
- buildRequest() skips deferred tools until activated
- Tools auto-activate on first model request (activatedTools map)
- agent + spawn_elfs marked as deferrable (large schemas, rarely needed early)
- Saves ~800 tokens per deferred tool per request
Gap 13 (M6): Pre/post compact hooks
- OnPreCompact/OnPostCompact callbacks in WindowConfig
- Called in doCompact() (shared by CompactIfNeeded + ForceCompact)
- M8 hooks system will extend these to full protocol
- Elf tool calls show as 🦉 [elf] <prompt> (not ⚙ [agent])
- Live 2-line progress beneath the elf label showing what the
elf is currently outputting (grey, auto-updated)
- Agent tool forwards elf streaming events via progress channel
- Progress cleared on turn completion
- elfProgressCh wired from agent tool → TUI
internal/elf/:
- BackgroundElf: runs on own goroutine with independent engine,
history, and provider. No shared mutable state.
- Manager: spawns elfs via router.Select() (picks best arm per
task type), tracks lifecycle, WaitAll(), CancelAll(), Cleanup().
internal/tool/agent/:
- Agent tool: LLM can call 'agent' to spawn sub-agents.
Supports task_type hint for routing, wait/background mode.
5-minute timeout, context cancellation propagated.
Concurrent tool execution:
- Read-only tools (fs.read, fs.grep, fs.glob, etc.) execute in
parallel via goroutines.
- Write tools (bash, fs.write, fs.edit) execute sequentially.
- Partition by tool.IsReadOnly().
TUI: /elf command explains how to use sub-agents.
5 elf tests. Exit criteria: parent spawns 3 background elfs on
different providers, collects and synthesizes results.
SummarizeStrategy: calls LLM to condense older messages into a
summary, preserving key decisions, file changes, tool outputs.
Falls back to truncation on failure. Keeps 6 recent messages.
Tool result persistence: outputs >50K chars saved to disk at
.gnoma/sessions/tool-results/{id}.txt with 2K preview inline.
TUI: /compact command for manual compaction, /clear now resets
engine history. Summarize strategy used by default (with
truncation fallback).
At startup, polls ollama (/api/tags) and llama.cpp (/v1/models) for
available models. Registers each as an arm in the router alongside
the CLI-specified provider.
Discovered: 7 ollama models + 1 llama.cpp model = 9 total arms.
Router can now select from multiple local models based on task type.
Discovery is non-blocking — failures logged and skipped.
- Default permission mode changed from bypass to default
- Removed mode info from status bar (shown on separator line instead)
- Ctrl+I toggles incognito mode
- Incognito mode: amber/yellow separator lines with 🔒 label
overrides permission mode color when active
- Shift+Tab cycles permission modes
- Shift+Tab cycles permission modes: bypass → default → plan →
accept_edits → auto → bypass
- /permission <mode> slash command to set specific mode
- Current mode shown in status bar (🛡 bypass)
- Permission checker wired into TUI config
Tools now go through permission.Checker before executing:
- plan mode: denies all writes (fs.write, bash), allows reads
- bypass mode: allows all (deny rules still enforced)
- default mode: prompts user (pipe: stdin prompt, TUI: auto-approve for now)
- accept_edits: auto-allows file ops, prompts for bash
- deny mode: denies all without allow rules
CLI flags: --permission <mode>, --incognito
Pipe mode: console Y/N prompt on stderr
TUI mode: auto-approve (proper overlay TODO)
Verified: plan mode correctly blocks fs.write, model sees error.
- Fixed: chat content no longer overflows past allocated height.
Lines are measured for physical width and hard-truncated to
exactly the chat area height. Input + status bar always visible.
- Header scrolls with chat (not pinned), only input/status fixed
- Git branch in status bar (green, via git rev-parse)
- Alt screen mode — terminal scrollback disabled
- Mouse wheel + PgUp/PgDown scroll within TUI
- New EventToolResult: tool output as dimmed indented block
- Separator lines above/below input, no status bar backgrounds
TUI launches when no piped input detected. Features:
- Chat panel with scrollable message history
- Streaming response with animated cursor
- User/assistant/tool/error message styling (purple theme)
- Status bar: provider, model, token count, turn count
- Input with basic editing
- Slash commands: /quit, /clear, /incognito (stub)
- Ctrl+C cancels current turn or exits
Built on charm.land/bubbletea/v2, charm.land/lipgloss/v2.
Session interface decouples TUI from engine via channels.
Pipe mode still works for non-interactive use.
System prompt gets a one-line summary (~200 chars): OS, CPU, RAM,
GPU, top runtimes, package count, PATH command count.
Full details available on demand via system_info tool with sections:
runtimes, packages, tools, hardware, all. LLM calls the tool when
it needs specifics — saves thousands of tokens per request.
Hardware detection: CPU model, core count, total RAM, GPU via lspci.
Package manager: pacman/apt/dnf/brew with dev package filtering.
PATH scan: 5541 executables. Runtime probing: 22 detected.
No hardcoded tool lists. Scans all $PATH directories for executables
(5541 on this system), then probes known runtime patterns for version
info (23 detected: Go, Python, Node, Rust, Ruby, Perl, Java, Dart,
Deno, Bun, Lua, LuaJIT, Guile, GCC, Clang, NASM + package managers).
System prompt includes: OS, shell, runtime versions, and notable
tools (git, docker, kubectl, fzf, rg, etc.) from the full PATH scan.
Total executable count reported so the LLM knows the full scope.
Milestones updated: M6 fixed context prefix, M12 multimodality.
Thin wrapper over OpenAI adapter with custom base URLs.
Ollama: localhost:11434/v1, llama.cpp: localhost:8080/v1.
No API key required for local providers.
Fixed: initial tool call args captured on first chunk
(Ollama sends complete args in one chunk, not as deltas).
Live verified: text + tool calling with qwen3:14b on Ollama.
Five providers now live: Mistral, Anthropic, OpenAI, Google, Ollama.
Streaming via goroutine+channel bridge (range-based iter.Seq2 → pull
iterator). Tool use with FunctionCall/FunctionResponse, tool name
sanitization, tool name map for FunctionResponse correlation.
Stop reason override (Google uses STOP for function calls).
Hardcoded model list (gemini-2.5-pro/flash, gemini-2.0-flash).
Wired into CLI with GOOGLE_API_KEY + GEMINI_API_KEY env support.
Live verified: text streaming + tool calling with gemini-2.5-flash.
Four providers now live: Mistral, Anthropic, OpenAI, Google.
Streaming, tool use (index-based delta accumulation), tool name
sanitization (fs.read → fs_read), StreamOptions.IncludeUsage for
token tracking. Hardcoded model list (gpt-4o, gpt-4o-mini, o3, o3-mini).
Wired into CLI with OPENAI_API_KEY env support.
Live verified: text streaming + tool calling with gpt-4o.
Streaming, tool use (with InputJSONDelta assembly), thinking blocks,
cache token tracking, system prompt separation. Tool name sanitization
(fs.read → fs_read) for Anthropic's naming constraints with reverse
translation on tool call responses.
Hardcoded model list with capabilities (Opus 4, Sonnet 4, Haiku 4.5).
Wired into CLI with ANTHROPIC_API_KEY + ANTHROPICS_API_KEY env support.
Also: migrated Mistral SDK to github.com/VikingOwl91/mistral-go-sdk.
Live verified: text streaming + tool calling with claude-sonnet-4.
126 tests across 9 packages.
Go 1.26 module (somegit.dev/Owlibou/gnoma), Makefile with
build/test/lint targets, CLAUDE.md with project conventions,
placeholder main.go, and .gitignore.
vikingowl