tommy0103/obelisk

Obelisk is a skill for Claude Code that gives an AI agent queryable access to its own past work. Most session history tools are built for humans who want to scroll through old conversations and find a specific chat. Obelisk is built differently: it indexes past sessions, tool calls, subagents, and workflows into a local SQLite database so that the agent itself can write and run queries against that history and answer questions in plain language. The idea is that you ask a question like "which files did we change last time we worked on the authentication bug" or "what did the review workflow subagents each conclude" and the agent figures out the right query, runs it, and gives you a structured answer. You do not browse or tag anything manually. The index covers several layers of recorded activity. Sessions are stored with their title, timestamps, and git branch. Individual messages from both user and assistant turns are indexed with full text. Every tool call is captured with its name, inputs, and which file paths it touched. Subagent conversations and workflow runs are also indexed, including per-agent transcripts linked back to the workflow that spawned them. Full-text search runs across all of these through SQLite's built-in FTS5 engine. Installation is a single command using the Claude Code skills system. The first run builds the index from your existing session files, which the README says takes about five seconds for 100 sessions. After that, the index updates only when new or changed session files are detected. The runtime has no npm dependencies and is about 400 lines of JavaScript using Node 22's built-in SQLite module. The license is MIT.