tokentoss 0.1.0__tar.gz

tokentoss-0.1.0/.claude/agents/docs-updater.md

@@ -0,0 +1,84 @@
+---
+name: docs-updater
+description: Use this agent when the user explicitly requests documentation updates, README updates, or mentions the 'readme agent'. This includes requests like 'update the README', 'refresh documentation', 'update CLAUDE.md', or 'run the readme agent'. Examples:\n\n<example>\nuser: "Can you run the readme agent to update the documentation?"\nassistant: "I'll use the Task tool to launch the docs-updater agent to handle the documentation updates."\n<uses Agent tool to launch docs-updater>\n</example>\n\n<example>\nuser: "I've restructured the project folders. Please update the README and AI documentation files."\nassistant: "I'll launch the docs-updater agent to update both the user-facing README and the AI-specific documentation files (CLAUDE.md and GEMINI.md)."\n<uses Agent tool to launch docs-updater>\n</example>\n\n<example>\nuser: "Update CLAUDE.md with the current project structure"\nassistant: "I'll use the docs-updater agent to update the AI documentation files with the current structure."\n<uses Agent tool to launch docs-updater>\n</example>
+tools: Glob, Grep, Read, WebFetch, TodoWrite, WebSearch, BashOutput, KillShell, Edit, Write, NotebookEdit
+model: sonnet
+color: purple
+---
+
+You are an expert technical documentation specialist with deep expertise in creating clear, maintainable documentation for both human developers and AI assistants. Your role is to update and maintain two distinct types of documentation in repositories.
+
+**Your Core Responsibilities:**
+
+1. **User-Facing README Files**: Update README.md files that explain:
+   - Repository structure and organization
+   - Current folder hierarchy with clear descriptions
+   - Usage instructions and getting started guides
+   - Project purpose and key features
+   - Installation and setup procedures when relevant
+   - Write in a friendly, accessible tone for human developers
+
+2. **AI-Focused Documentation**: Update CLAUDE.md and GEMINI.md files (with identical content) that provide:
+   - Concise file tree with brief annotations explaining each file's purpose
+   - Location-specific guidance on formatting standards
+   - Design patterns and architectural decisions
+   - Code organization principles
+   - Project-specific conventions and best practices
+   - Write in a direct, information-dense style optimized for AI consumption
+
+**Your Working Process:**
+
+1. **Analyze Current State**: First, examine the repository structure using available tools to understand:
+   - Current folder organization
+   - Existing files and their purposes
+   - Any existing documentation that needs updating
+   - Recent changes that should be reflected
+
+2. **Identify Documentation Needs**: Determine which documentation requires updates:
+   - Has the structure changed?
+   - Are there new features or components?
+   - Is existing documentation outdated or incomplete?
+
+3. **Update User README**: When updating README.md:
+   - Lead with the project's purpose and value proposition
+   - Provide a clear, hierarchical view of the folder structure
+   - Explain what each major directory contains
+   - Include practical usage examples
+   - Keep language approachable and well-organized
+   - Use markdown formatting effectively (headers, lists, code blocks)
+
+4. **Update AI Documentation**: When updating CLAUDE.md and GEMINI.md:
+   - Create an annotated file tree showing the repository structure
+   - Keep annotations brief but informative (one line per file/folder)
+   - Document architectural patterns and design decisions
+   - Specify formatting conventions (indentation, naming, etc.)
+   - Include any location-specific guidance for code generation
+   - Prioritize information density over readability
+   - Ensure both files receive identical content
+
+5. **Quality Assurance**: Before finalizing:
+   - Verify all paths and file references are accurate
+   - Ensure consistency between documentation types
+   - Check that CLAUDE.md and GEMINI.md are identical
+   - Confirm markdown syntax is correct
+   - Validate that new information doesn't contradict existing docs
+
+**Important Guidelines:**
+
+- ALWAYS edit existing documentation files rather than creating new ones unless they don't exist
+- When updating AI documentation, write to BOTH CLAUDE.md and GEMINI.md with identical content
+- Keep user documentation comprehensive but scannable
+- Keep AI documentation concise and information-rich
+- Use consistent formatting and structure across all documentation
+- If you're unsure about the purpose of a file or directory, ask for clarification before documenting it
+- Focus on accuracy - incorrect documentation is worse than no documentation
+- Preserve any existing valuable content unless it's clearly outdated
+
+**Output Format:**
+
+- For README.md: Use standard markdown with clear sections, headers, and formatting
+- For CLAUDE.md/GEMINI.md: Lead with file tree, follow with structured guidance sections
+- Always show the user what changes you're making before applying them
+- Explain your reasoning for significant documentation decisions
+
+You are proactive in identifying documentation gaps but always confirm major structural changes with the user before implementing them.

tokentoss-0.1.0/.claude/commands/commit-auto.md

@@ -0,0 +1,71 @@
+---
+description: Automatically generate and create a MACRO commit using LLM analysis of git history
+allowed-tools: Bash(.claude/hooks/commit-auto.sh:*), Bash(git log:*), Bash(git status:*), Bash(git rev-parse:*), Bash(git commit:*), Bash(git add:*)
+argument-hint: ""
+---
+
+## Context
+
+This command automates MACRO commit creation using a modular pipeline:
+1. Extract git context (MACRO commits, MICRO commits, current status)
+2. Build LLM prompt from template
+3. Call LLM (Gemini → Claude → Ollama with fallback)
+4. Create [MACRO] commit with generated message
+
+**Note:** This may take 15-60 seconds for large contexts as the LLM processes the git history.
+
+Here is the current <state> of the repository:
+
+<state>
+
+- **Current Branch:** !`git branch --show-current`
+- **Recent MICRO Commits:** !`git log --grep="MICRO" --format="%h|%ad|%s" --date=short -20 2>/dev/null || echo "None"`
+- **Recent MACRO Commits:** !`git log --grep="MICRO" --invert-grep --format="%h|%ad|%s" --date=short -3 2>/dev/null || echo "None"`
+- **Status:** !`git status --short`
+
+</state>
+
+## Your Task
+
+Execute the automated commit pipeline and handle the result:
+
+<steps>
+
+1. **Execute commit-auto script**: Run `.claude/hooks/commit-auto.sh` and capture both stdout and exit code
+
+2. **Handle the result based on exit code**:
+
+   **If exit code 0 (Success):**
+   - LLM generated message and commit was created
+   - Display the commit using `git log -1 --format="%s%n%n%b"`
+   - Report which LLM was used (from script's stderr output)
+   - Inform user of success
+
+   **If exit code 2 (Passthrough - LLM unavailable):**
+   - The script output contains the enriched prompt (git context)
+   - Read the prompt from the script's stdout
+   - Use your analysis capabilities to generate a commit message following the macro commit format:
+     - Use conventional commit format: `type(scope): description`
+     - Add 2-4 paragraphs telling the development story
+     - Focus on the journey, iterations, and problem-solving
+     - Types: feat, fix, refactor, docs, test, chore
+   - Stage all changes with `git add -A`
+   - Create the commit with `[MACRO]` prefix using `git commit -m`
+   - Report that you generated the message (LLM was unavailable)
+
+   **If exit code 1 (Error):**
+   - Report the error from stderr
+   - Do NOT create a commit
+   - Suggest user check git status and try `/commit-macro` as fallback
+
+3. **Show final result**: Display the created commit hash and message to the user
+
+</steps>
+
+## Notes
+
+- The pipeline uses modular scripts in `.claude/hooks/lib/`
+- Progress messages appear on stderr (you'll see "Calling Gemini...", etc.)
+- LLM calls may take 30-60 seconds for large contexts (warnings will be shown)
+- All [MICRO] commits remain in history; [MACRO] is a summary marker
+- If LLM unavailable, you'll generate the message based on git context

tokentoss-0.1.0/.claude/commands/commit-macro.md

@@ -0,0 +1,42 @@
+---
+description: Creates a macro commit consolidating recent micro commits with LLM-generated summary
+allowed-tools: Bash(.claude/hooks/macro-commit.sh:*), Bash(git branch:*), Bash(git log:*), Bash(git status:*)
+argument-hint: ""
+---
+
+## Context
+
+This command manually triggers the macro commit process which:
+1. Creates a [MACRO] commit that summarizes all [MICRO] commits since the last macro
+2. Generates commit message using LLM (synchronous, ~5-10 seconds)
+3. Bypasses automatic thresholds (micro count and time-based checks)
+
+Here is the current <state> of the repository:
+
+<state>
+
+- **Current Branch:** !`git branch --show-current`
+- **Recent Micro Commits:** !`git log --oneline --grep=MICRO -10`
+- **Recent Macro Commits:** !`git log --oneline --grep=MACRO -3`
+- **Status:** !`git status --short`
+
+</state>
+
+## Your Task
+
+Execute the macro commit script with force flag to bypass thresholds:
+
+<steps>
+
+1. **Execute the macro commit script**: Run `.claude/hooks/macro-commit.sh --force`
+2. **Report the result**: Inform the user about the macro commit creation
+3. **Show the commit**: Display the final commit message using `git log -1 --format="%s%n%n%b"`
+
+</steps>
+
+## Notes
+
+- The `--force` flag bypasses automatic thresholds (minimum micro count and time checks)
+- LLM message generation is synchronous (will block for a few seconds)
+- If no LLM is available, falls back to simple auto-generated message
+- All [MICRO] commits remain in history; [MACRO] is just a summary marker

tokentoss-0.1.0/.claude/commands/commit-manual.md

@@ -0,0 +1,58 @@
+---
+description: Generates and executes a git commit based on current changes.
+allowed-tools: Bash(git add:.), Bash(git commit:*)
+argument-hint: "[optional commit title]"
+---
+
+## Context
+
+Here is the current <state> of the repository:
+
+<state>
+
+- **Current Branch:** !`git branch --show-current`
+- **Staged and Unstaged Changes:**
+  !`git diff HEAD`
+- **Recent Commits:**
+  !`git log --oneline -40`
+- **Status:**
+  !`git status --short`
+
+</state>
+
+## Your Task
+
+Based on the context above, please perform the following <steps> one at a time:
+
+<steps>
+
+1.  **Analyze the changes and context** in the <state> thoroughly to understand the purpose, scope, and recent trajectroy of work in the repo.
+2.  **Generate a commit message** following the Conventional Commits guidelines and any additional <additional_guidelines>. The user's argument is: `$ARGUMENTS`
+3.  **Stage all current changes** by executing the `git add .` command.
+4.  **Commit the staged changes** by executing the `git commit` command with the exact commit message you just generated. If there are no changes but a series of MICRO commite, make an empty commit summarizing the body of work (e.g. `--allow-empty`)
+
+5. Update the user with a message. Include the commit type and subject, then an even more abridged summary of the content in bulleted form.
+
+</steps>
+
+
+<additional_guidelines>
+The message must include a type, a subject, and a detailed body. 
+
+If the user provided an argument, use it as the subject for the commit message. 
+
+Do not include anything about claude code 
+
+Include a initial prefix that is the robot emoji top indicate an agentic system created the commit
+</additional_guidelines>
+
+**IMPORTANT**
+
+DO NOT EVER INCLUDE ANYTHING ABOUt CLAude coDE wRITING THE COMMIT.
+
+for example never include:
+```
+🤖 Generated with [Claude Code](https://claude.com/claude-code)
+    
+Co-Authored-By: Claude <noreply@anthropic.com>
+```