v0.1.10 (bugfix for v0.1.9)

Version 0.1.910

Features

• [✅] 1: HOOKS

  • [✅] 1.1: Rename prePrompt to preLLMResponse
  • [✅] 1.2: Add postUserInput hook
  • [✅] 1.3: Add postLLMResponse hook
  • [✅] 1.4: Add preToolCall hook
  • [✅] 1.5: Add postToolCall hook

• [✅] 2: HOOK COMMANDS

  • [✅] 2.1: Add MAXCONTEXT command for setting maximum context length

• [✅] 3: CODE IMPROVEMENTS

  • [✅] 3.1: Move sessionState to top of context.json file
  • [✅] 3.2: added /context reload command to reload context verbatim

• [✅] 4: Tool enhancements

  • [✅] 4.1: Add encode-speech tool
  • [✅] 4.2: Add getLoraDetails to ImageTool
  • [✅] 4.3: Add compareImageToPrompt to ImageTool
  • [✅] 4.4: Add joycaption.sh helper script to repository

• [✅] 5: TECHNICAL DEBT

  • [✅] 5.1: Reduce duplicate code in llm-agent.ts
  • [✅] 5.2: Refactor llm-agent.ts to move tasks to helpers

• [✅] 6: DOCUMENTATION IMPROVEMENTS

  • [✅] 6.1: Update the morph tool instructions
  • [✅] 6.2: Update the README
  • [✅] 6.3: Documented classes in src/agent
  • [✅] 6.4: Add AGENTS.md

• [✅] 7: CLI USABILITY IMPROVEMENTS

  • [✅] 7.1: Add /? as shorter alias for /introspect command
  • [✅] 7.2: Enhanced /? (introspect) output formatting
    • Added color-coded output (cyan for values, yellow for properties, magenta for templates, dim for comments)
    • Improved YAML-like tree structure for variable definitions
    • Show explicit/implicit indicator for variable definitions
    • Display current values in def: output when available
    • Better error messages suggesting def: when var: not found

• [✅] 8: FILE LOCATION MIGRATION

  • [✅] 8.1: Move settings from ~/.grok/ to ~/.zds-ai/
    • User settings: ~/.grok/user-settings.json → ~/.zds-ai/cli-settings.json
    • MCP config: ~/.grok/mcp.json → ~/.zds-ai/mcp.json
    • Project settings: .grok/settings.json → .zds-ai/project-settings.json
    • Chat history: ~/.grok/chat-history.json → ~/.zds-ai/context.json

• [✅] 9: PROMPT VARIABLE SYSTEM IMPROVEMENTS

  • [✅] 9.1: Add SESSION variables for session state tracking
    • SESSION:BACKEND:MODEL (weight 10, persistent)
    • SESSION:BACKEND:SERVICE (weight 20, persistent)
    • SESSION:FRONTEND (weight 30, persistent)
    • SESSION:STDIN_IS_TTY (weight 31, persistent)
    • SESSION:STDOUT_IS_TTY (weight 31, persistent)
  • [✅] 9.2: Dynamic system message rendering
    • System prompt now renders from variables before each LLM call
    • getSystemPrompt() automatically renders current variable state
    • setSystemPrompt() deprecated - always renders from variables
    • Ensures instance hook variables always included in fresh sessions
    • renderSystemMessage() method added to LLMAgent and HookManager

Bug Fixes

• [✅] 1: Fixed backend/model switching regression (dependency injection using getters)
• [✅] 2: Fixed session state persistence (persona, apiKeyEnvVar now save correctly)
• [✅] 3: Fixed hook prompt variable commands (SET, SET_FILE, SET_TEMP_FILE) not being applied
• [✅] 4: Fixed hook ENV commands not being applied in llm-agent.ts hooks
• [✅] 5: Save permanent prompt vars with context.json
• [✅] 6: Fixed /introspect and /? commands not showing user input in chat history
• [✅] 7: Fixed SET/SET_FILE/SET_TEMP_FILE regex to allow underscores in variable names
• [✅] 8: Fixed XML wrapping logic for prompt variables (now checks child's template, not parent's)
• [✅] 9: Fixed findBirthChildren to only return immediate children, not grandchildren
• [✅] 10: Fixed var: output showing "Values (0)" - now shows "No direct values (renders from children/getter)"
• [✅] 11: Fixed var: output not showing children (now displays children list matching def: output)
• [✅] 12: Fixed empty child wrapper tags being rendered - now only creates XML wrappers when child has content
• [✅] 13: Fixed orphaned grandchildren not appearing in parent render - automatically creates intermediate parents (e.g., MESSAGE:ACL when MESSAGE:ACL:CURRENT exists)
• [✅] 14: Persona is not being saved in context.json when personas are changed (PR #53)
• [✅] 15: Fixed Variable.set() semantics to replace values instead of appending (Issue #28)
• [✅] 16: Fixed APP:TOOLS duplication by tracking seen variables and using append() for duplicates (Issue #28)
• [✅] 17: Migrated Grok API from deprecated search_parameters to Agent Tools API (Issue #25)
  • Fixed 410 error "Live search is deprecated"
  • Added WebSearchTool interface with full filter support
  • Updated chat() and chatStream() to use tools array with {type: "web_search"}
  • Maintains backward compatibility -- old search_parameters auto-convert to new format
• [✅] 18: Disabled Grok web_search tool due to API inconsistencies (PR #49)
  • Temporarily disabled web_search tool creation in chat() and chatStream() methods
  • Agents should use Tavily for web search instead
  • Code preserved for future re-enable when Grok API behavior is clarified
• [✅] 19: Fixed persona/mood/task not being restored in session resume (Issue #27)
  • hookManager.setPersona/setMood/startActiveTask now call deps methods to update agent properties
  • Previously only set environment variables, causing getPersona/getMood/getActiveTask to return empty strings
  • Fixes bug where persona was saved to context.json but not restored on session resume
• [✅] 20: Fixed OpenAI API error with tools array exceeding 128 tool limit (Issue #55)
  • OpenAI has hard limit of 128 tools per request
  • Added tool count check in chat() and chatStream() methods
  • Trims tools array to 128 when using OpenAI backend (base tools prioritized over MCP tools)
  • Logs warning when trimming occurs
• [✅] 21: Fixed OpenAI API 400 error with newer models requiring max_completion_tokens (Issue #57)
  • OpenAI models from gpt-5.x, o1, o3, o4-mini onwards require max_completion_tokens instead of max_tokens
  • Added model detection logic to use correct parameter based on model name
  • Applied to both chat() and chatStream() methods in src/grok/client.ts
  • Maintains backward compatibility for older models and non-OpenAI backends
• [✅] 22: Fixed Dependabot security vulnerabilities blocking v0.1.9 release (Issue #59)
  • Updated @modelcontextprotocol/sdk to 1.26.0 (fixes ReDoS and cross-client data leak)
  • Updated ink to 6.6.0 (resolves lodash prototype pollution)
  • Added qs override to 6.14.1 (fixes DoS vulnerability)
  • Resolved all 10 vulnerabilities (6 high, 4 moderate)

Full Changelog: v0.1.8...v0.1.10

v0.1.8

Full Changelog: v0.1.7...v0.1.8

v0.1.7

v0.1.6 ist kaput. v0.1.7 actually works!

Full Changelog: v0.1.5...v0.1.7

v0.1.5

Full Changelog: v0.1.4...v0.1.5

v0.1.4

Security fixes, bundled helpers for Excel/XSLX reading and SD/A1111 image generation, /rephrase support, System message cleanup, and Image Support!!!

Full Changelog: v0.1.3...v0.1.4

v0.1.3

What's Changed

• Potential fix for code scanning alert no. 1: Workflow does not contain permissions by @jcheek in #2
• Potential fix for code scanning alert no. 3: Clear-text logging of sensitive information by @jcheek in #4
• Potential fix for code scanning alert no. 4: Incomplete URL substring sanitization by @jcheek in #3
• Potential fix for code scanning alert no. 5: Incomplete string escaping or encoding by @jcheek in #5
• Fix incomplete temperature test suite by @jcheek in #6
• Fix trufflehog action input parameter names by @jcheek in #9
• Potential fix for 1 code quality finding by @jcheek in #10
• Refactor: rename ambiguous property, extract JSON utility functions by @Copilot in #7
• Remove unsupported base/head parameters from trufflehog action by @jcheek in #12
• Potential fix for 1 code quality finding by @jcheek in #14
• Extract magic numbers to named constants for better maintainability by @Copilot in #13
• Potential fixes for 5 code quality findings by @jcheek in #15
• Potential fix for 1 code quality finding by @jcheek in #11
• Potential fixes for 4 code quality findings by @jcheek in #16

New Contributors

• @jcheek made their first contribution in #2
• @Copilot made their first contribution in #7

Full Changelog: v0.1.2...v0.1.3

v0.1.3-alpha.1

What's Changed

• Potential fix for code scanning alert no. 1: Workflow does not contain permissions by @jcheek in #2
• Potential fix for code scanning alert no. 3: Clear-text logging of sensitive information by @jcheek in #4
• Potential fix for code scanning alert no. 4: Incomplete URL substring sanitization by @jcheek in #3
• Potential fix for code scanning alert no. 5: Incomplete string escaping or encoding by @jcheek in #5
• Fix incomplete temperature test suite by @jcheek in #6
• Fix trufflehog action input parameter names by @jcheek in #9
• Potential fix for 1 code quality finding by @jcheek in #10
• Refactor: rename ambiguous property, extract JSON utility functions by @Copilot in #7
• Remove unsupported base/head parameters from trufflehog action by @jcheek in #12
• Potential fix for 1 code quality finding by @jcheek in #14
• Extract magic numbers to named constants for better maintainability by @Copilot in #13
• Potential fixes for 5 code quality findings by @jcheek in #15
• Potential fix for 1 code quality finding by @jcheek in #11

New Contributors

• @jcheek made their first contribution in #2
• @Copilot made their first contribution in #7

Full Changelog: v0.1.2...v0.1.3-alpha.1

0.1.2

Full Changelog: https://github.com/ziquid/zds-ai-cli/commits/v0.1.2