# Agterm Archive

> Park a whole agterm workspace on disk and bring it back later — snapshot its sessions, order, cwds, splits and each pane's claude session id, close it, and recreate it with every claude resumed where it left off. For workspaces that are done for now but not done for good.

- Skill: `isachivka/agterm-archive` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add isachivka/agterm-archive`
- Raw SKILL.md: https://api.skillmd.com/api/skills/isachivka/agterm-archive/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: isachivka (https://skillmd.com/u/isachivka)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/isachivka/agterm-archive

---


# agterm-archive

Only inside agterm. Session ids come from the `agterm-backup` capture hook — without it an
agent pane comes back as a plain shell. Claude and Codex panes are both handled; the
restore types `claude --resume <id>` or `codex resume <id>` accordingly.

```bash
agterm-archive archive [workspace]   # snapshot + close a whole workspace (default: active)
agterm-archive session [target]      # snapshot + close ONE session (default: active)
agterm-archive list                  # what is parked, both kinds
agterm-archive restore <name>        # recreate it, resume the agents
agterm-archive drop <name>           # forget an archive
agterm-archive install               # add the palette entries (cmd+ctrl+a>a / cmd+ctrl+a>r)
agterm-archive uninstall             # remove them
```

`install` writes `~/.local/bin/agterm-archive` and a managed block in
`~/.config/agterm/keymap.conf` with two custom commands: **Archive workspace**
(`cmd+ctrl+a>a`, archives the workspace the cursor is in, passing `$AGT_WORKSPACE_ID`) and
**Archive session** (`cmd+ctrl+a>s`, archives just the session under the cursor) and
**Restore archive** (`cmd+ctrl+a>r`, lists both kinds in agterm's native fuzzy picker — rows
read `workspace: name` / `session: workspace / name` — and restores the pick). The wrapper resolves the newest installed copy of the skill at call
time, because the plugin cache path carries a commit sha.

The plugin's `bin/` is on `PATH` inside a Claude Code session, so run it by name.
Workspace archives live in `~/.agterm-backup/archives/<name>.json`, single sessions in
`archives/sessions/<workspace>--<session>.json`, the run log in
`~/.agterm-backup/archive.log` — palette commands print nowhere else.

- `archive` records every session's name, cwd, title, order, split (axis + ratio) and, per pane,
  the claude session id plus its launch flags. Then `agtermctl workspace delete`. It refuses
  to archive the workspace it is running in — that would kill the caller — and warns when a
  claude pane has no captured id. `--keep` snapshots without closing, `--force` overwrites.
- `session` archives one session out of a workspace the same way, and remembers which
  workspace it came from. `restore` puts it back there, creating the workspace if it is
  gone. Same refusal to close the session it is running in, same `--keep`/`--force`.
- `restore` recreates the sessions in order (`session new --workspace-name … --create-workspace`),
  re-opens each split at its old ratio, and types `claude --resume <id> <flags>` into each
  pane that had one. `--no-boot` leaves plain shells; `--keep` keeps the archive file
  (otherwise a successful restore consumes it); `--force` restores into an already-open
  workspace.
- After booting a claude, `restore` waits for its "resume full session as-is / from
  summary" question and answers **2** (full session) — the same reason you restored it.
  `--answer-choice 1` / `--no-answer` change that.
- Non-claude panes come back as shells in the right cwd. Whatever process they ran is gone —
  a snapshot is not a checkpoint.

This is a separate concern from `agterm-backup`: that one covers what is currently open
across a restart, this one covers what you deliberately closed. A restored workspace is
live again, so the hook re-records it and backup picks it up from there.