Show HN: sudocode – manage specs, tasks, and context-as-code for coding agents (github.com)

🤖 AI Summary
sudocode is a git-native "context-as-code" system for coding agents that lets you capture durable specs (user intent), agent tasks (issues), and agent executions alongside your codebase. Installing via npm and initializing a .sudocode project (and optionally running a local MCP server or Claude Code plugin) gives agents immediate access to long-lived, version-controlled context stored as human-editable Markdown/YAML and machine-optimized JSONL/SQLite. Agents can create and update issues, emit execution artifacts (diffs, reports), and backlink to specs—producing an auditable trail of decisions and work that lives in your repo. For AI/ML practitioners building multi-agent workflows, sudocode addresses agent amnesia and hallucination by persisting context under version control and providing structured relationships (blocks, parent-child, related) and graph traversal features like topological ordering, hierarchical handoffs, and automatic issue claiming for concurrency. Technical highlights: .sudocode/specs/, .sudocode/issues/, and .sudocode/agents/ storage; JSONL as source-of-truth with a local SQLite cache and 5s debounce auto-sync; bidirectional backlinks; support for multiple executions per issue; and git-based distribution with AI-assisted merge conflict handling. The result is a lightweight, auditable coordination layer that makes AI agents collaborative, persistent contributors to software projects rather than ephemeral chat sessions.
Loading comments...
loading comments...