{"categories":[{"label":"Software Development","url":"https://skillfed.io/packages/category/software-development/7"},{"label":"Utilities","url":"https://skillfed.io/packages/category/utilities/5"},{"label":"Terminals","url":"https://skillfed.io/packages/category/terminals"}],"enrichment":{"capability":"pairmux lets AI agents control interactive terminal programs through a real PTY while allowing humans to watch, take over, or hand back control when needed.","skillfed_tags":["ai-agent-tools","tmux-wrapper","human-in-the-loop"],"use_cases":["Let an LLM agent run interactive Python REPLs, SQL shells, or other stateful interpreters while maintaining session context across multiple commands","Enable agents to drive credential-protected workflows (ssh, database login, git operations) by pausing and notifying a human when a secret prompt appears","Orchestrate multi-step terminal tasks (build, test, deploy) where agents can watch live output and humans can intervene without restarting the session","Provide MCP clients with typed terminal operations (new, run, wait, send, log) as first-class tools for agentic terminal automation","Debug agent-driven terminal interactions by capturing full output with exit codes and timestamps in a queryable journal"],"what_it_does":"pairmux is a thin Agent-Computer Interface layer over tmux that gives AI agents a real pseudo-terminal with persistent shell state, allowing them to drive interactive programs like REPLs, psql, ssh, docker exec, git rebase, pagers, and TUIs. Unlike exec-style agent tools that run commands without a terminal, pairmux maintains a live tmux session where agents can run commands, wait for completion or recognized prompts, and inspect output \u2014 all while a human retains normal access to the same terminal and can take over at any time.\n\nThe package handles credential and confirmation prompts by recognizing patterns (passwords, passphrases, pager markers, Python prompts) and refusing to auto-answer; instead it can notify a human or wait for manual input via a live tmux pane. Agents interact through a CLI with JSON output, a blocking `run` command that waits for completion or timeout, and read-only `peek` and `log` commands that let multiple agents inspect the same terminal without lock contention. It ships as a prebuilt native Go binary in platform-specific wheels, requires no separate service, and integrates with MCP clients via a stdio server.","worth_installing":"Yes, if you need agents to drive interactive terminal programs with real PTY semantics and human handoff. The active maintenance, zero security issues, and permissive license are favorable. Install friction is moderate (platform and tmux version constraints) and the alpha status suggests early-stage stability \u2014 suitable for projects that can tolerate rapid iteration but not for production systems requiring long-term stability guarantees."},"id":"pairmux","links":{"html":"https://skillfed.io/packages/pairmux","md":"https://skillfed.io/packages/pairmux.md","pypi":"https://pypi.org/project/pairmux/"},"maintenance":{"status":"active"},"meta":{"latest_release":"2026-08-10","license_spdx":null,"license_treatment":"permissive","name":"pairmux","python_support":"supports_current","summary":"Blocking tmux terminal control for AI agents, with captured logs and live human handoff"},"popularity":{"monthly_downloads":654478,"position":5524,"tier":"top_15000"},"security":{"n_vulnerabilities":0},"version":"0.4.0"}