Agent Session Manager

The README discloses the principal local configuration and history paths, tmux sessions, arbitrary custom commands, self-update behavior, and deletion effects. Updates, stops, and most deletions include confirmation, while the Windows installer is described as per-user, supporting ordinary least-privilege and confirmation handling. Deductions apply because quick tab close has no confirmation, YOLO mode weakens downstream agent safeguards, arbitrary commands may be launched, and potentially sensitive Claude, Aider, OpenCode, and shell histories are read without an access-control, redaction, retention, or privacy model. Dependencies are versioned, but the supplied evidence lacks go.sum, vulnerability scanning, an update policy, or release signing, and the workflow selects the latest GoReleaser. Forking and diff views offer limited recovery assistance, but there is no comprehensive undo for deletion, self-update, or agent file changes. Licensing, inspiration, and related tools are attributed, although publisher identity and maintenance ownership are not independently verified.

The README, go.mod, and release workflow are broadly consistent about Go 1.24, cross-platform builds, and principal dependencies, and the tmux or psmux and agent-CLI prerequisites are explicit. A deduction applies for visible shortcut inconsistencies: t is described both as creating a tab and toggling status lines, while another section assigns status lines to o; several capabilities are supported only by README assertions. Runtime availability depends on external package managers, GitHub releases, agent CLIs, and a multiplexer, although useful platform alternatives are documented. Failure-message evidence is thin: the single-instance PID error, missing-psmux explanation, and a few installation cautions are concrete, but systematic diagnostics, retries, and troubleshooting guidance are absent.

The intended audience and scenarios are covered thoroughly through workspaces, parallel agents, comparison, resume, history search, and conversation forks. Six named agent families, Amazon Q, and arbitrary custom commands are supported, with concrete Linux, macOS, and native Windows accommodation, earning strong scenario and environment scores. Boundaries are partly explicit: Aider and custom commands cannot resume, forking is Claude-only, and history search is restricted to ASMGR project directories. Deductions apply because there is no consolidated compatibility matrix or complete security boundary. Keyboard triggers, modal contexts, and confirmations are detailed, but shortcut conflicts and broad YOLO/custom-command entry points reduce trigger precision.

The README is well organized around features, installation, updating, operation, configuration, and architecture, with substantial navigation and examples. Homebrew, Windows, script, system-package, and source builds are documented, and the complete MIT license matches the metadata. Naming is mostly stable, but Agent Session Manager, ASMGR, and the separate desktop product coexist, while shortcut labels conflict. Numerous walkthroughs partly substitute for examples and FAQ, but there is no dedicated FAQ or troubleshooting section. Limitations are disclosed only across scattered sections. Versions and a release workflow are visible, but no changelog, compatibility policy, or explicit semantic-versioning commitment is supplied. Maintenance responsibility can only be inferred from the repository namespace, copyright line, and automation; no maintainer roster, support policy, or security-reporting path is shown.

The TUI combines multi-agent sessions, live previews, tabs, groups, notes, search, diffs, and forks into directly actionable output, with extensive shortcuts and state indicators. This provides strong usability and clear marginal value over manually coordinating many terminals. The cost-benefit score is reduced because users must install and maintain a multiplexer, at least one agent CLI, and the relevant account environment; self-updating and multiple history parsers add operational and privacy costs. No measurements of performance, resource use, or scaling limits are provided.

Core claims map to concrete README features, configuration paths, and named architecture files. go.mod corroborates the language and TUI dependencies, while the release workflow corroborates cross-platform release work and checksum generation, providing moderate traceability and cross-source support. Deductions apply because the supplied evidence omits implementation files, tests, go.sum, release configuration, and change history, leaving many detailed behaviors supported only by the README. The documentation generally distinguishes functions, prerequisites, and limitations, but promotional assertions such as “powerful” and “improved reliability” have no metrics or supporting evidence, and inferred statements are not explicitly labeled.