Archivonchik

Archivonchik

Archivonchik turns a repository into an AI-ready project.

Use this skill when the user asks to:

• make a repository AI-ready;
• create or improve AGENTS.md, CLAUDE.md, or Copilot instructions;
• continue from previous work;
• preserve project memory across Claude Code sessions;
• run Build Doctor;
• lint project context;
• discover build/test/lint commands;
• classify a command as safe/caution/dangerous;
• review a change;
• generate tests;
• prepare a PR summary;
• onboard a developer or coding agent.

Core loop

Always work through:

Understand → Plan → Change → Verify → Remember

Required startup behavior

When Archivonchik is invoked inside a repository:

1. Inspect the project structure before editing.
2. Identify the stack and package manager if possible.
3. Look for existing context files:
   • AGENTS.md
   • CLAUDE.md
   • .github/copilot-instructions.md
   • docs/ai-context/*
4. Read project memory if present:
   • CURRENT_STATE.md
   • NEXT_STEPS.md
   • DECISIONS.md
   • KNOWN_BUGS.md
   • DO_NOT_TOUCH.md
   • session-memory/LAST_SESSION.md
5. Choose the correct playbook.
6. Avoid creating huge or duplicated instruction files.

Routing

Use these files as needed:

• Repository analysis: reference/REPOSITORY_ANALYSIS.md, playbooks/repository-intake.md
• Project memory: reference/PROJECT_MEMORY.md, playbooks/update-project-memory.md
• Session continuity: reference/SESSION_CONTINUITY.md, playbooks/session-start.md, playbooks/session-end.md
• Context compiler: reference/CONTEXT_ENGINEERING.md, reference/AGENTS_MD.md, reference/CLAUDE_MD.md, reference/COPILOT_INSTRUCTIONS.md
• Context linting: reference/CONTEXT_LINTING.md, playbooks/context-lint.md
• Command safety: reference/COMMAND_SAFETY.md, playbooks/command-safety-review.md
• Build/test discovery: reference/BUILD_TEST_DISCOVERY.md
• Build Doctor: playbooks/build-doctor.md
• Task workflows: playbooks/feature-workflow.md, playbooks/bugfix-workflow.md, playbooks/refactor-workflow.md, playbooks/code-review-workflow.md
• Stack-specific behavior: profiles/*
• Output templates: templates/*
• Other agent adapters: adapters/*
• Common usage recipes: recipes/*
• Showcase examples and sample reports: showcase/*

Strong rules

• Do not store secrets, API keys, tokens, passwords, private keys, or personal data in project memory.
• Do not save entire chat transcripts.
• Do not create bloated context files.
• Do not run dangerous commands without explaining risk and asking for confirmation.
• Do not assume commands are correct; verify from project files.
• Do not edit high-risk files listed in DO_NOT_TOUCH.md without confirmation.
• Prefer minimal, project-specific instructions over generic advice.
• After meaningful work, update project memory only with durable facts.