Initial Release

Velle v0.1.0 — Self-Prompting for Claude Code

Velle is an MCP server that gives Claude Code autonomous self-prompting capability. By injecting text as keyboard
input into the parent terminal via Win32 console APIs, the agent can decide what to do next and continue working
without waiting for human input.

What's in this release:

• Self-prompting chains — The agent calls velle_prompt with its next instruction, creating adaptive multi-turn
  workflows where each turn decides the next
• 2-step slash command introspection — Inject a slash command and a follow-up prompt as a single atomic operation,
  giving the agent access to client-side information like context usage (/context), task state (/tasks), and on-demand
  compaction (/compact)
• 12 validated slash commands — Tested and whitelisted for safe non-interactive use
• Configurable guardrails — Turn limits, cooldown between prompts, and audit logging, all controlled via velle.json
  (user-editable, agent cannot modify)
• Win32 console injection — Reliable keystroke injection using AttachConsole + CONIN$ + WriteConsoleInputW with a
  500ms Enter key delay for ConPTY compatibility

Tested: 20/20 autonomous turns with zero injection failures, including resilience through system reminder
interruptions. Turn limit enforcement confirmed — the agent stops when it hits the wall.

Requirements: Windows 11, Python 3.10+, Claude Code CLI