| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Add deprecation notices across hexai-mcp-server codebase and documentation.
The MCP server is now considered experimental and not actively maintained,
as prompts are managed through slash commands and meta-commands instead.
Changes:
- Add runtime deprecation warning to stderr on binary startup
- Mark feature as deprecated in README.md
- Add deprecation notices to all 8 MCP documentation files
- Add deprecation log message to run.go
- Update Magefile.go build comment
The code remains fully functional but warns users of its experimental status.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Amp-Thread-ID: https://ampcode.com/threads/T-019c50bc-2906-77db-a31e-0d553338d99b
Co-authored-by: Amp <amp@ampcode.com>
|
|
Adds optional syncing of MCP prompts to Markdown slash command files
for AI agents that don't yet support MCP prompts (e.g., Cursor IDE).
Features:
- Syncs prompts on create/update/delete operations
- Configurable via TOML config, environment vars, or CLI flags
- Backfill support with --sync-all flag
- Thread-safe atomic file writes
- Non-fatal sync failures (logged but don't break operations)
- Comprehensive test coverage (81.1% total)
Configuration:
- Config: [mcp] slashcommand_sync = true, slashcommand_dir = "~/.cursor/commands"
- Env: HEXAI_MCP_SLASHCOMMAND_SYNC, HEXAI_MCP_SLASHCOMMAND_DIR
- CLI: --slashcommand-sync, --slashcommand-dir, --sync-all
Fixes config merging bug where project config would reset global MCP settings.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
|
|
Update update_prompt and delete_prompt meta-prompts to work without requiring
Claude to programmatically fetch prompts via prompts/get (which isn't available
as a callable tool in the MCP client).
Changes:
- delete_prompt: Simplified to just ask for confirmation without fetching
- update_prompt: Ask what changes to make without fetching current version
- Removed duplicate/confusing instructions
- More direct workflow: confirm → execute tool
This fixes the issue where Claude would try to use listMcpResources or other
incorrect APIs when trying to follow the meta-prompt instructions.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Update update_prompt and delete_prompt meta-prompts to explicitly instruct
Claude to use "prompts/get" instead of vague "access via prompts capability".
This prevents Claude from incorrectly trying to use the resources API.
Changes:
- update_prompt: "use prompts/get to retrieve the current prompt"
- delete_prompt: "use prompts/get to retrieve the prompt and show details"
Fixes issue where Claude would try to access "plugin:hexai-prompts" resources
instead of using the correct prompts/get method on the hexai-prompts server.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Adds a third built-in meta-prompt for interactively deleting custom prompts
with confirmation. This completes the prompt management trilogy (save, update,
delete).
Changes:
- Add delete_prompt meta-prompt with prompt_name argument
- Interactive workflow: show prompt → confirm → delete via delete_prompt tool
- Update test to expect 3 built-in prompts (was 2)
- Includes safety notes about built-in prompts and backups
The meta-prompt ensures users see what they're deleting and must explicitly
confirm before the delete_prompt tool is called.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Implements tools/list and tools/call endpoints to expose prompt management
operations (create, update, delete) as callable MCP tools. This enables Claude
to use the meta-prompts (save_prompt, update_prompt) to actually create and
modify prompts.
Key changes:
- Add Tool type definitions (Tool, ListToolsRequest, CallToolRequest, etc.)
- Implement handleToolsList() returning 3 tools with JSON Schemas
- Implement handleToolsCall() with tool wrappers for create/update/delete
- Add 17 comprehensive unit tests (82.8% coverage maintained)
- Update meta-prompts to reference tools instead of JSON-RPC methods
- Enable listChanged notifications for immediate prompt availability
- Refactor large functions into helpers to stay under 50-line limit
Tools advertised alongside Prompts capability. All functions under 50 lines.
Backward compatible - existing prompts/* JSON-RPC methods unchanged.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
This change moves built-in meta-prompts (save_prompt, update_prompt) from
external JSONL files into compiled Go code, making them always available
and version-controlled with the binary.
Changes:
- Add default_prompts.go with built-in meta-prompt definitions
- Update store to load built-ins from code, not files
- Add protection: built-ins cannot be updated/deleted
- Handle name conflicts: built-ins take precedence with warnings
- Update docs to reflect new architecture (no default.jsonl needed)
- Add comprehensive tests for built-in protection
- Add hexai-mcp-server binary to .gitignore
Benefits:
- Built-ins always in sync with binary version
- No setup required (no default.jsonl to manage)
- Clear separation between built-in and user prompts
- Protection prevents accidental modification of meta-prompts
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Move all cache and state files under ~/.local/hexai/ with subdirectories:
- ~/.local/hexai/cache/ (was ~/.cache/hexai/)
- ~/.local/hexai/state/ (was ~/.local/state/hexai/)
- ~/.local/hexai/data/ (was ~/.local/share/hexai/)
This centralizes all non-config hexai files under a single .local/hexai directory,
making it easier to manage and back up user data.
Amp-Thread-ID: https://ampcode.com/threads/T-019c4e03-73db-70a2-ae27-3e1cc31d59c3
Co-authored-by: Amp <amp@ampcode.com>
|
|
Two issues prevented Claude Code CLI from discovering MCP prompts:
1. Protocol version mismatch: Server always returned "2025-06-18" but
Claude Code sends "2025-11-25". Per MCP spec, server must echo
back the client's version if supported. Now supports all known
versions (2024-11-05 through 2025-11-25).
2. Null prompts array: Go nil slices marshal as JSON null, but Claude
Code expects an empty array []. Initialize prompts slice with make()
to ensure [] in JSON output.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
The MCP stdio protocol uses newline-delimited JSON (JSONL), not LSP-style
Content-Length headers. This was discovered during integration testing with
Claude Code CLI which could not connect to the server.
Also updates docs/mcp-setup.md with correct Claude Code CLI configuration
(use `claude mcp add` instead of editing JSON files) and adds integration
test runbook for testing against real Claude Code CLI instances.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
Ensures all code complies with AGENTS.md standards:
- Refactored functions to be under 50 lines each
- Removed built-in prompts - all prompts now served from database only
- Split handlePromptsCreate (72→37 lines) with helper functions
- Split handlePromptsUpdate (76→44 lines) with helper function
- Deleted internal/promptstore/builtin.go (no longer needed)
- Updated tests to create prompts dynamically instead of relying on built-ins
All tests pass with 81.5% coverage maintained.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Implements a full Model Context Protocol (MCP) server for managing and serving prompts
to LLM applications. The server provides CRUD operations for prompts with automatic
backups and template rendering support.
Key additions:
- cmd/hexai-mcp-server: Main MCP server binary entrypoint
- internal/hexaimcp: Server orchestrator with configuration and setup
- internal/mcp: Core MCP protocol implementation (JSON-RPC 2.0)
- internal/promptstore: Prompt storage with JSONL backend and automatic backups
- Comprehensive test suites achieving 80%+ coverage for all MCP packages
- Magefile targets for building and installing the MCP server
- Complete documentation for setup, API, prompts, and backups
Test coverage:
- internal/hexaimcp: 84.3%
- internal/mcp: 80.3%
- internal/promptstore: 81.2%
- Overall project: 81.5%
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
- Add StateDir() helper function respecting XDG_STATE_HOME (~/.local/state/hexai/)
- Implement JSONL-based history storage for tmux popup submissions
- New history.go with AppendHistory() and GetHistory() functions
- Store timestamp, agent name, cwd, and submitted text
- Comprehensive unit tests for history functionality
- Integrate history append into tmux edit workflow after successful submission
- Move logs from /tmp/ to persistent state directory:
- hexai-lsp.log: ~/.local/state/hexai/hexai-lsp.log
- hexai-tmux-edit.log: ~/.local/state/hexai/hexai-tmux-edit.log
- Update README.md with File Locations section documenting XDG directories
- Fix pre-existing test failures by isolating project config in unit tests
- Panic on state directory creation failure instead of silent fallback
All unit tests pass. Follows XDG Base Directory Specification for proper
state file management with persistence across reboots.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
hexai-tmux-edit was opening in the wrong directory because the tmux
popup wasn't being told which directory to use. This fix:
- Updates bind-key example to cd into pane's current path
- Passes working directory to popup via -d flag using os.Getwd()
- Updates all documentation with corrected bind-key examples
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
|
|
Implement comprehensive unit tests for critical internal packages
to increase overall test coverage from 80.9% to 81.8%, providing
a buffer above the 80% threshold specified in project guidelines.
Changes:
- Add config parsing tests (parseTemperatureValue, decodeModelEntry, resolvedModel, parseSurfaceEntries)
- Add HumanBytes utility tests with edge cases and boundary values
- Fix HumanBytes bug: corrected suffix array indexing (off-by-one error)
- Fix HumanBytes to handle negative values correctly
- Add comprehensive shellJoin and isSafeBare tests for shell escaping
- Update comments to reflect actual behavior and implementation details
Coverage impact:
- internal/appconfig: Improved config parsing function coverage
- internal/textutil: HumanBytes now at 100% coverage (fixed bug in process)
- internal/tmux: shellJoin and isSafeBare now at 100% coverage
- Overall project: 80.9% → 81.8% (+0.9%)
All tests pass. Code formatted with gofumpt.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
|
|
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
Claude Code's prompt input field does not support vim commands for
clearing. The previous sequence 'Escape gg C-v G d i' resulted in
literal text 'gGdijo' appearing in the prompt instead of clearing it.
Changed to 'C-a C-k' (Emacs/readline style):
- C-a: Move cursor to start of line
- C-k: Kill (delete) from cursor to end of line
Also added 150ms delay after Escape key in sendClearSequence to ensure
mode transitions complete before subsequent keys are sent (though Claude
uses readline, this helps other potential vim-based agents).
Changes:
- internal/tmuxedit/claude_agent.go: clearKeys "C-a C-k" instead of vim
- internal/tmuxedit/claude_agent_test.go: Update test expectations
- internal/tmuxedit/agentutil.go: Add time import and Escape delay
- docs/usage.md: Update claude clear method documentation
- docs/tmux.md: Clarify claude uses readline not vim
- config.toml.example: Update claude description
Integration tested:
- Extracted: "final verification test"
- Sent: "FINAL REPLACED TEXT NO VIM COMMANDS"
- Result: Clean replacement with NO vim command artifacts
- All unit tests pass (67/67)
- Coverage: 80.9%
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
Amp CLI runs in TUI mode with box-drawing UI (│ text │) similar to
Cursor, not shell-style (> prompt). Updated prompt pattern from
`(?m)>\s*(.+)$` to `(?m)│\s*(.+?)\s*│\s*$` to correctly extract
text from amp's box UI.
Changes:
- internal/tmuxedit/config_agent.go: Update amp promptPat to box pattern
- internal/tmuxedit/config_agent_test.go: Update test to use box format
- docs/usage.md: Document detection order and clear methods for all agents
- docs/tmux.md: Clarify input mode handling (Vim vs Emacs/readline)
- config.toml.example: Add detailed agent descriptions and patterns
- prompts/tmux-edit-integration-tests.md: Add test status, mock editor best practices
Integration tests verified:
- Amp detection: amp/sourcegraph keywords
- Prompt extraction: "hello world test" correctly captured
- End-to-end workflow: text modification and sending works
- Multi-line support: all lines delivered correctly
- All unit tests pass (67/67)
- Coverage: 80.9% (meets requirement)
- Cursor and Claude implementations unchanged
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
Replace monolithic AgentConfig struct with an Agent interface backed by
baseAgent defaults and separate implementations for cursor (box-drawing
extraction, bulk backspace clearing) and claude (section-scoped extraction
with continuation lines, vim clearing). Simple agents (amp, aider) and
user-defined agents use configAgent with baseAgent defaults.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
- Reorder agents: cursor first to avoid false positive from "Claude 4.5
Sonnet" model name appearing in cursor panes
- Change cursor detect pattern to box-drawing UI elements instead of
name-based matching
- Change claude detect pattern to ❯ prompt instead of generic "claude"
- Support multi-line prompt extraction using last-contiguous-block
algorithm to ignore command-review and dialog boxes
- Fix deduplicateText to always return full edited text (clear + resend)
instead of stripping original prefix which caused double-removal
- Replace vim-style clear (Escape gg dG i) with universal End+BSpace*200
since cursor's prompt is not actually vim
- Add Key*N repeat syntax in ClearKeys (parsed by parseKeyRepeat, sent
via tmux send-keys -N)
- Add 300ms sleep after clear to let TUI drain queued backspaces before
new text arrives
- Add debug logging to /tmp/hexai-tmux-edit.log
- Add strip pattern for "ctrl+c to stop" noise in cursor prompt
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
New tool that opens $EDITOR in a tmux popup for composing longer prompts
when working with AI CLI agents (Claude Code, Cursor, Amp, Aider, etc.).
Captures existing prompt text from the target pane, pre-fills the editor,
and sends edited text back via tmux send-keys. Config-driven agent
detection via regex patterns in [tmux_edit] config section.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
Files matching .gitignore patterns or user-configured extra patterns are
now skipped for completions and code actions. Configurable via [ignore]
section in config.toml with gitignore, extra_patterns, and
lsp_notify_ignored options. Includes hot-reload support and env var
overrides (HEXAI_IGNORE_*).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
to 80%
Introduce support for a .hexaiconfig.toml file at the git repository root
that selectively overrides the global config. Precedence order:
defaults → global config → project config → env vars.
Also lower the coverage threshold from 85% to 80%.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
|
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
Remove all GitHub Copilot integration from the codebase to streamline
the supported provider set to OpenAI, OpenRouter, Anthropic, and Ollama.
Changes:
- Delete core Copilot implementation (copilot.go) and all related tests
- Remove Copilot configuration fields from App struct and Config
- Remove Copilot from provider factory and API key handling
- Update all test files to replace Copilot references with other providers
- Remove Copilot documentation from README, configuration guide, and examples
- Remove Copilot section from config.toml.example
All tests pass successfully after removal.
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
- Add Anthropic field merging in config.mergeProviderFields() to properly load AnthropicModel/BaseURL/Temperature from config.toml
- Add anthropic and openrouter cases in lsp.resolveDefaultModel() to prevent fallback to OpenAI defaults
- Update default Anthropic model to valid claude-3-5-sonnet-20240620 (previous 20241022 doesn't exist)
- Update test expectation to match new default model
This fixes the issue where Anthropic provider configuration was ignored, causing LSP to return dummy completions instead of real LLM responses.
Co-authored-by: Cursor <cursoragent@cursor.com>
|
|
Add RequestTimeout field to llm.Config construction in:
- internal/hexailsp/run.go (buildClientIfNil)
- internal/llmutils/client.go (NewClientFromApp)
- internal/lsp/server.go (newClientForProvider)
This ensures the configured timeout is used in all code paths,
not just the ones already updated.
|
|
Local LLMs (LM Studio, Ollama, etc.) often need more than the default
30-second timeout. Added request_timeout config option (in seconds)
to [general] section and HEXAI_REQUEST_TIMEOUT env var.
Original constructor signatures preserved via *WithTimeout variants,
so no test changes required.
|
|
Major features:
- Native Anthropic API provider support
- Configurable completion_wait_all option for multi-backend completions
|
|
- Add completion_wait_all config option (default true) to wait for all
backends before returning results, or return first result immediately
- Fix Anthropic API: extract system messages to top-level system field
as required by Messages API (was causing 400 errors)
- Add anthropic case to server.go clientFor() for model overrides
|
|
- Reorganize anthropic.go: types, interface checks, constructor, public methods, private methods
- Extract helper methods from Chat() and ChatStream() to keep functions under 50 lines
- Add resolveOptions(), sendRequest(), extractContent() private methods
- Add explicit interface satisfaction check for Client and Streamer
- Add documentation comments to all public methods (Chat, Name, DefaultModel, ChatStream)
- Apply gofmt and gofumpt formatting
- Apply goimports for import ordering
All 51 tests in llm package pass
All code adheres to best practices from go-projects.md:
- Value semantics (value receivers only)
- Constructors before methods
- Public before private
- Functions under 50 lines
- Explicit interface satisfaction
- Documentation on all public identifiers
- Proper error handling
- Context as first parameter for I/O functions
Amp-Thread-ID: https://ampcode.com/threads/T-019c0af1-f215-72cf-9940-b014b1a9576b
Co-authored-by: Amp <amp@ampcode.com>
|
