65 Commits

• Enable more biome lints and fix things
  Mario Zechner · 2025-12-21 22:56:20 +01:00
  d5fd685901
• feat(coding-agent): add --skills CLI flag for filtering skills

  Adds glob pattern support for skill filtering:
  - --skills <patterns> CLI flag (comma-separated glob patterns)
  - includeSkills setting in settings.json
  - ignoredSkills now supports glob patterns
  - ignoredSkills takes precedence over includeSkills and --skills
  
  Closes #268

  Mario Zechner · 2025-12-21 20:58:54 +01:00
  d95a5c4186
• feat(coding-agent): add configurable skills directories (#269)
  Nico Bailon · 2025-12-21 20:48:40 +01:00
  70440f7591
• Add configurable OAuth storage backend and respect --models in model selector

  - Add setOAuthStorage() and resetOAuthStorage() to pi-ai for custom storage backends
  - Configure coding-agent to use its own configurable OAuth path via getOAuthPath()
  - Model selector (/model command) now only shows models from --models scope when set
  - Rewrite OAuth documentation in pi-ai README with examples
  
  Fixes #255

  Mario Zechner · 2025-12-20 22:00:53 +01:00
  a81dc5eaca
• Add Google Gemini CLI and Antigravity OAuth providers

  - Add google-gemini-cli provider: free Gemini 2.0/2.5 via Cloud Code Assist
  - Add google-antigravity provider: free Gemini 3, Claude, GPT-OSS via sandbox
  - Move OAuth infrastructure from coding-agent to ai package
  - Fix thinking signature handling for cross-model handoff
  - Fix OpenAI message ID length limit (max 64 chars)
  - Add GitHub Copilot overflow pattern detection
  - Add OAuth provider tests for context overflow and streaming

  Mario Zechner · 2025-12-20 21:34:18 +01:00
  c359023c3f
• fix(coding-agent): fix Google Cloud project discovery

  Use correct API response format (cloudaicompanionProject) and proper
  onboarding flow with tierId and metadata. Add retry logic for project
  provisioning which may take time.

  Mario Zechner · 2025-12-20 10:29:09 +01:00
  1294239a7a
• feat(coding-agent): add Google Cloud Code Assist OAuth flow

  - Add OAuth handler with PKCE flow and local callback server
  - Automatic project discovery via loadCodeAssist/onboardUser endpoints
  - Store credentials with projectId for API calls
  - Encode token+projectId as JSON for provider to decode
  - Register as 'google-cloud-code-assist' OAuth provider

  Mario Zechner · 2025-12-20 10:27:07 +01:00
  b6fe07b618
• fix(coding-agent): add connection error to retryable errors

  Fixes #252

  Mario Zechner · 2025-12-20 09:55:03 +01:00
  c1382818cc
• fix(coding-agent): clamp thinking level to model capabilities

  - setThinkingLevel() now clamps xhigh to high when model doesn't support it
  - Model changes automatically re-clamp the current thinking level
  - Fixed /model command to use session.setModel() instead of agent.setModel()
  - Footer and editor border color update after model/thinking changes
  
  Closes #253

  Mario Zechner · 2025-12-20 09:52:57 +01:00
  b7c3cf9436
• feat: add xhigh thinking level support for gpt-5.2 models

  - Add supportsXhigh() function to ai package for checking xhigh support
  - Clamp xhigh to high for OpenAI models that don't support it
  - Update coding-agent to use centralized supportsXhigh()
  - gpt-5.2, gpt-5.2-codex now show xhigh in thinking selector
  
  Closes #236

  Mario Zechner · 2025-12-19 20:07:24 +01:00
  7e38897673
• feat: add xhigh thinking level support for gpt-5.2 and gpt-5.2-codex

  - Add XHIGH_MODELS constant and getAvailableThinkingLevels() to AgentSession
  - Update ThinkingSelectorComponent to accept availableLevels parameter
  - Both shift+tab cycling and /thinking command now show xhigh for supported models
  - Update types.ts documentation to list supported models

  theBucky · 2025-12-19 19:59:13 +01:00
  4f981d8ebc
• fix(coding-agent): handle OpenRouter models with colons in IDs (#242)
  Aliou Diallo · 2025-12-19 15:26:12 +01:00
  9a7863fc98
• fix(coding-agent): prevent global AGENTS.md from being loaded twice (#239)

  When cwd is `~/.pi/agent/` or a subdirectory, the global AGENTS.md file
  was included twice in the system prompt.
  
  `loadProjectContextFiles()` loads context files in two steps:
  1. Explicitly loads from `getAgentDir()` (`~/.pi/agent/`)
  2. Walks from cwd up to root, collecting all AGENTS.md/CLAUDE.md files
  
  There was no deduplication. When cwd is at or below `~/.pi/agent/`, the
  ancestor walk includes that directory again.
  
  Fix: track seen paths with a Set and skip duplicates.

  Aliou Diallo · 2025-12-19 15:24:49 +01:00
  a9a1a62b14
• Forward onUpdate callback through hook wrapper (#238)
  Nico Bailon · 2025-12-19 15:23:41 +01:00
  a54c70bbed
• Refactor subagent tool, fix custom tool discovery, fix JSON mode stdout flush

  Breaking changes:
  - Custom tools now require index.ts entry point in subdirectory
    (e.g., tools/mytool/index.ts instead of tools/mytool.ts)
  
  Subagent tool improvements:
  - Refactored to use Message[] from ai package instead of custom types
  - Extracted agent discovery to separate agents.ts module
  - Added parallel mode streaming (shows progress from all tasks)
  - Added turn count to usage stats footer
  - Removed redundant Query section from scout output
  
  Fixes:
  - JSON mode stdout flush: Fixed race condition where pi --mode json
    could exit before all output was written, causing consumers to
    miss final events
  
  Also:
  - Added signal/timeout support to pi.exec() for custom tools and hooks
  - Renamed pi-pods bin to avoid conflict with pi

  Mario Zechner · 2025-12-19 04:54:02 +01:00
  4fb3af93fb
• Merge pull request #232 from aliou/fix/symlinked-slash-commands

  fix(coding-agent): support symlinked slash commands in discovery

  Mario Zechner · 2025-12-19 00:57:11 +01:00
  c1f2541639
• Add type guards for tool_result event narrowing

  - Export isBashToolResult, isReadToolResult, etc. type guards
  - Update hooks.md with type guard usage examples
  - Document custom tool handling in hooks.md

  Mario Zechner · 2025-12-19 00:48:03 +01:00
  d353e5e219
• Expose full tool result content and details in hook tool_result event

  Breaking change: ToolResultEvent now exposes content and typed details
  instead of just a result string. Hook handlers returning { result: ... }
  must change to { content: [...] }.
  
  - ToolResultEvent is now a discriminated union based on toolName
  - Each built-in tool has typed details (BashToolDetails, etc.)
  - Export tool details types and TruncationResult
  - Update hooks.md documentation
  
  Closes #233

  Mario Zechner · 2025-12-19 00:42:08 +01:00
  3d9bad8fb6
• Skills standard compliance

  Implement Agent Skills standard (https://agentskills.io/specification):
  - Validate name (must match parent dir, lowercase, max 64 chars)
  - Validate description (required, max 1024 chars)
  - Warn on unknown frontmatter fields
  - Warn on name collisions (keep first)
  - Change prompt format to XML structure
  - Remove {baseDir} placeholder (use relative paths)
  - Add tests and update documentation
  
  fixes #231

  Mario Zechner · 2025-12-19 00:11:39 +01:00
  05b7b81338
• fix(coding-agent): support symlinked slash commands in discovery
  Aliou Diallo · 2025-12-18 23:57:54 +01:00
  b9fd912ecf
• fix(coding-agent): support symlinked tools and hooks in discovery (#219)
  Aliou Diallo · 2025-12-18 13:07:30 +01:00
  de2de851c8
• Check for compaction before prompt submission, not just after agent turn
  Mario Zechner · 2025-12-17 22:25:20 +01:00
  d5dde00d28
• Improve documentation: README settings table, philosophy section, custom-tools intro, rpc hook_error event, hooks import aliases
  Mario Zechner · 2025-12-17 22:04:28 +01:00
  3424550d21
• Improve system prompt docs, clean up theme/skills/hooks docs, fix toolResults type

  - System prompt: clearer pointers to specific doc files
  - theme.md: added thinkingXhigh, bashMode tokens, fixed Theme class methods
  - skills.md: rewrote with better framing, examples, and skill repositories
  - hooks.md: fixed timeout/error handling docs, added custom tool interception note
  - Breaking: turn_end event toolResults changed from AppMessage[] to ToolResultMessage[]

  Mario Zechner · 2025-12-17 21:27:28 +01:00
  5e5bdadbf9
• fix(coding-agent): detect image MIME via file-type (#205)

  Co-authored-by: Mario Zechner <badlogicgames@gmail.com>

  Peter Steinberger · 2025-12-17 17:11:56 +01:00
  d70edf571e
• Fix TUI performance regression: add caching to Box, use Text directly for built-in tools
  Mario Zechner · 2025-12-17 16:39:23 +01:00
  909989066a
• Custom tools with session lifecycle, examples for hooks and tools

  - Custom tools: TypeScript modules that extend pi with new tools
    - Custom TUI rendering via renderCall/renderResult
    - User interaction via pi.ui (select, confirm, input, notify)
    - Session lifecycle via onSession callback for state reconstruction
    - Examples: todo.ts, question.ts, hello.ts
  
  - Hook examples: permission-gate, git-checkpoint, protected-paths
  
  - Session lifecycle centralized in AgentSession
    - Works across all modes (interactive, print, RPC)
    - Unified session event for hooks (replaces session_start/session_switch)
  
  - Box component added to pi-tui
  
  - Examples bundled in npm and binary releases
  
  Fixes #190

  Mario Zechner · 2025-12-17 16:03:23 +01:00
  e7097d911a
• Release v0.22.5
  Mario Zechner · 2025-12-17 01:22:13 +01:00
  295f51b53f
• Add tool result streaming

  - Add AgentToolUpdateCallback type and optional onUpdate callback to AgentTool.execute()
  - Add tool_execution_update event with toolCallId, toolName, args, partialResult
  - Normalize tool_execution_end to always use AgentToolResult (no more string fallback)
  - Bash tool streams truncated rolling buffer output during execution
  - ToolExecutionComponent shows last N lines when collapsed (not first N)
  - Interactive mode handles tool_execution_update events
  - Update RPC docs and ai/agent READMEs
  
  fixes #44

  Mario Zechner · 2025-12-16 14:53:17 +01:00
  7ac832586f
• GitHub Copilot: auto-enable models, fix gpt-5 API, normalize tool call IDs

  - Auto-enable all models after /login via POST /models/{model}/policy
  - Use openai-responses API for gpt-5/o3/o4 models (not accessible via completions)
  - Normalize tool call IDs when switching between github-copilot models with different APIs
    (fixes #198: openai-responses generates 450+ char IDs with special chars that break other models)
  - Update README with streamlined GitHub Copilot docs

  Mario Zechner · 2025-12-15 20:06:11 +01:00
  c5543f7586
• Add GitHub Copilot support (#191)

  - OAuth login for GitHub Copilot via /login command
  - Support for github.com and GitHub Enterprise
  - Models sourced from models.dev (Claude, GPT, Gemini, Grok, etc.)
  - Dynamic base URL from token's proxy-ep field
  - Use vscode-chat integration ID for API compatibility
  - Documentation for model enablement at github.com/settings/copilot/features
  
  Co-authored-by: cau1k <cau1k@users.noreply.github.com>

  Mario Zechner · 2025-12-15 19:05:17 +01:00
  b66157c649
• clean up
  cau1k · 2025-12-15 00:34:38 -05:00
  fb648bb158
• feat: initial impl

  - add GitHub Copilot model discovery (env token fallback, headers,
  compat) plus fallback list and quoted provider keys in generated map
  - surface Copilot provider end-to-end (KnownProvider/default, env+OAuth
  token refresh/save, enterprise base URL swap, available only when
  creds/env exist)
  - tweak interactive OAuth UI to render instruction text and prompt
  placeholders
  
  gpt-5.2-high took about 35 minutes. It had a lot of trouble with `npm
  check`  and went off on a "let's adjust every tsconfig" side quest.
  Device code flow works, but the ai/scripts/generate-models.ts impl is
  wrong as models from months ago are missing and only those deprecated
  are accessible in the /models picker.

  cau1k · 2025-12-14 17:18:13 -05:00
  ccae7a4e0e
• Merge feature/tui-inline-images: Inline image rendering for supported terminals
  Mario Zechner · 2025-12-13 23:22:31 +01:00
  27e68d856e
• Add /show-images command to toggle inline image display

  - Add terminal.showImages setting to settings-manager.ts
  - Add /show-images slash command (only visible if terminal supports images)
  - ToolExecutionComponent checks both terminal support and user setting
  - Shows text fallback when inline images are disabled

  Mario Zechner · 2025-12-13 23:14:46 +01:00
  f68a933d2c
• coding-agent: fix macOS screenshot filenames with unicode spaces (#181)
  Nico Bailon · 2025-12-13 22:04:01 +01:00
  9a7bbb2839
• coding-agent, mom: add skills API export and mom skills auto-discovery

  coding-agent:
  - Export loadSkillsFromDir, formatSkillsForPrompt, and related types
  - Refactor skills.ts to expose public API
  
  mom:
  - Add skills auto-discovery from workspace/skills and channel/skills
  - Fix skill loading to use host paths (not Docker container paths)
  - Update README and system prompt with SKILL.md format docs

  Mario Zechner · 2025-12-13 00:56:10 +01:00
  e707ac4cd0
• coding-agent: change Pi skills to use SKILL.md convention

  Breaking change: Pi skills must now be named SKILL.md inside a directory,
  matching Codex CLI format. Previously any *.md file was treated as a skill.
  
  Migrate by renaming ~/.pi/agent/skills/foo.md to ~/.pi/agent/skills/foo/SKILL.md

  Mario Zechner · 2025-12-12 23:23:17 +01:00
  3b2b9abffc
• Add skills system with Claude Code compatibility (#171)

  * Add skills system with Claude Code compatibility
  
  * consolidate skills into single module, merge loaders, add <available_skills> XML tags
  
  * add Codex CLI skills compatibility, skip hidden/symlinks

  Nico Bailon · 2025-12-12 18:24:52 +01:00
  09bca9672f
• Auto-retry on transient provider errors (overloaded, rate limit, 5xx)

  - Add retry logic with exponential backoff (2s, 4s, 8s) in AgentSession
  - Disable Anthropic SDK built-in retries (maxRetries: 0) to allow app-level handling
  - TUI shows retry status with Escape to cancel
  - RPC mode: add set_auto_retry, abort_retry commands and auto_retry_start/end events
  - Configurable via settings.json: retry.enabled, retry.maxRetries, retry.baseDelayMs
  - Exclude context overflow errors from retry (handled by compaction)
  
  fixes #157

  Mario Zechner · 2025-12-10 23:36:46 +01:00
  bb445d24f1
• Fix branch selector for single message and --no-session mode

  - Allow branch selector to open with single user message (changed <= 1 to === 0 check)
  - Support in-memory branching for --no-session mode (no files created)
  - Add isEnabled() getter to SessionManager
  - Update sessionFile getter to return null when sessions disabled
  - Update SessionSwitchEvent types to allow null session files
  - Add branching tests for single message and --no-session scenarios
  
  fixes #163

  Mario Zechner · 2025-12-10 22:41:32 +01:00
  3d35e7c469
• feat: add line number display to HTML export for read tool calls

  Implements line number display in HTML exports for read tool calls, matching the TUI format.
  When offset/limit parameters are provided, displays path:startLine-endLine in yellow color.
  
  Fixes #166

  Mario Zechner · 2025-12-10 22:05:14 +01:00
  09a48fd1c3
• Add Mistral as AI provider

  - Add Mistral to KnownProvider type and model generation
  - Implement Mistral-specific compat handling in openai-completions:
    - requiresToolResultName: tool results need name field
    - requiresAssistantAfterToolResult: synthetic assistant message between tool/user
    - requiresThinkingAsText: thinking blocks as <thinking> text
    - requiresMistralToolIds: tool IDs must be exactly 9 alphanumeric chars
  - Add MISTRAL_API_KEY environment variable support
  - Add Mistral tests across all test files
  - Update documentation (README, CHANGELOG) for both ai and coding-agent packages
  - Remove client IDs from gemini.md, reference upstream source instead
  
  Closes #165

  Mario Zechner · 2025-12-10 20:36:19 +01:00
  99b4b1aca0
• Add hooks system with pi.send() for external message injection

  - Hook discovery from ~/.pi/agent/hooks/, .pi/hooks/, --hook flag
  - Events: session_start, session_switch, agent_start/end, turn_start/end, tool_call, tool_result, branch
  - tool_call can block execution, tool_result can modify results
  - pi.send(text, attachments?) to inject messages from external sources
  - UI primitives: ctx.ui.select/confirm/input/notify
  - Context: ctx.exec(), ctx.cwd, ctx.sessionFile, ctx.hasUI
  - Docs shipped with npm package and binary builds
  - System prompt references docs folder

  Mario Zechner · 2025-12-10 00:50:30 +01:00
  7c553acd1e
• Fix queued messages.
  Mario Zechner · 2025-12-09 23:51:22 +01:00
  942d8d3c95
• feat(coding-agent): implement hooks system

  - Add hooks infrastructure in core/hooks/ (loader, runner, types)
  - HookUIContext interface with mode-specific implementations
  - Interactive mode: TUI-based selector/input/confirm dialogs
  - RPC mode: JSON protocol for hook UI requests/responses
  - Print mode: no-op UI context (hooks run but can't prompt)
  - AgentSession.branch() now async, returns { selectedText, skipped }
  - Settings: hooks[] and hookTimeout configuration
  - Export hook types from package for hook authors
  
  Based on PR #147 proposal, adapted for new architecture.

  Mario Zechner · 2025-12-09 22:17:12 +01:00
  04d59f31ea
• Simplify compaction: remove proactive abort, use Agent.continue() for retry

  - Add agentLoopContinue() to pi-ai for resuming from existing context
  - Add Agent.continue() method and transport.continue() interface
  - Simplify AgentSession compaction to two cases: overflow (auto-retry) and threshold (no retry)
  - Remove proactive mid-turn compaction abort
  - Merge turn prefix summary into main summary
  - Add isCompacting property to AgentSession and RPC state
  - Block input during compaction in interactive mode
  - Show compaction count on session resume
  - Rename RPC.md to rpc.md for consistency
  
  Related to #128

  Mario Zechner · 2025-12-09 21:43:49 +01:00
  5a9d844f9a
• feat(coding-agent): implement new compaction system with overflow recovery

  Phase 1: Updated compaction.ts
  - findCutPoint now returns CutPointResult with isSplitTurn and turnStartIndex
  - Can cut at user, assistant, or bashExecution messages (never tool results)
  - Added turnPrefixSummary support for split turns (parallel summarization)
  - estimateTokens helper for context size estimation
  
  Phase 2: Updated session-manager.ts
  - CompactionEntry now has optional turnPrefixSummary field
  - loadSessionFromEntries injects both summaries when turn was split
  
  Phase 3: Updated agent-session.ts
  - Overflow detection via isContextOverflow after agent_end
  - Proactive compaction check on turn_end before next LLM call
  - _abortingForCompaction flag to skip saving aborted messages
  - Auto-retry after overflow recovery or proactive compaction
  - New event fields: reason (overflow/threshold), willRetry
  
  Phase 4: Updated interactive-mode.ts
  - Shows reason in compaction status (Context overflow detected...)
  - Shows retry status after compaction
  
  Tests updated for new CutPointResult return type.

  Mario Zechner · 2025-12-09 17:18:53 +01:00
  a38e619095
• Fix hardcoded truncation limits in tool output display

  - Add maxLines and maxBytes fields to TruncationResult to track actual limits used
  - Update tool-execution.ts to use actual limits from truncation result
  - Add fallbacks to DEFAULT_MAX_* for backward compatibility with old sessions
  - Fix outdated comments that said 30KB when default is 50KB

  Mario Zechner · 2025-12-09 17:06:29 +01:00
  ee9acdb49d
• feat(coding-agent): maintain in-memory session entries for --no-session compaction support

  SessionManager now tracks all entries in memory regardless of whether file
  persistence is enabled. This allows compaction to work in --no-session mode
  where no session file is created.
  
  - Added inMemoryEntries array to store session entries
  - All save methods now push to inMemoryEntries
  - loadEntries() returns in-memory entries when file persistence is disabled
  - File persistence only writes when enabled flag is true
  - Session loading (constructor, setSessionFile) populates in-memory entries

  Mario Zechner · 2025-12-09 16:55:42 +01:00
  8226975080