selfmod Records tmux Sessions and Replays Episodes
Per the GitHub README, the open-source project selfmod records tmux terminal panes, processes captured frames with Claude Code, and replays named episodes interactively. The tool captures a pane by default every 50ms with smart filtering that stores frames just before a change, 400ms after changes settle, and periodic snapshots every 5 seconds if different, according to the repository documentation. selfmod groups described frames into hyphenated-named episodes with titles and consolidated summaries, stores data in a SQLite database, and supports commands to record, process, consolidate, search, show, and replay episodes. Replay can run an episode in an interactive Claude Code session or execute in a tmux pane, with a --dry-run option noted in the README.
What happened
Per the GitHub README for the repository selfmod (user BerryBaronBonanza), the project records tmux terminal panes and converts those recordings into searchable, replayable episodes. The README documents a three-step flow: Record (capture frames), Process (send frames to Claude Code for frame descriptions and episode grouping), and Replay (feed an episode summary to an interactive Claude Code session to re-execute the task).
Technical details
Per the README, capture defaults to every 50ms with "smart filtering" that stores the frame just before content changes, the frame 400ms after changes settle, and a periodic snapshot every 5 seconds if different from the last stored frame. The tool writes to a SQLite database (default ~/.local/share/selfmod/selfmod.db) and exposes CLI commands: record, process, consolidate, search, show, and replay. Episode summaries are consolidated from batches so they read as self-contained instructions for replay.
Editorial analysis - technical context
Tools that convert terminal UI recordings into structured episodes pair lightweight capture with LLM-based description to make tacit workflows reproducible. Using periodic snapshots plus change-based frames is a common technique to balance fidelity and storage cost.
Context and significance
Industry context
For practitioners, the repo exemplifies an emerging pattern where developer-facing automation is built by combining screen/terminal capture with LLM summarization and instruction-driven replay. This reduces manual documentation friction but raises operational questions around model cost, reproducibility, and safety when re-executing commands.
What to watch
Monitor privacy and safety controls around replayed commands, the cost and latency of using Claude Code for batch processing, and how episode consolidation handles incomplete or noisy captures. The README does not provide a public rationale beyond usage examples.
Scoring Rationale
This is a solid open-source developer tool that demonstrates LLM-driven terminal automation, relevant to practitioners exploring reproducibility and automation workflows. Its scope is niche and experimental, so impact is notable but not industry-shifting.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
