v0.6.2: A stop that worked is no longer reported as a failure

• Killing a session's children leaves them waiting to be collected by a parent
  that is itself about to be killed, and ps still lists them. They were
  counted as survivors, so stopping a Claude session with six MCP servers under
  it reported one of seven stopped when all seven had gone.

v0.6.1: Stopping a session now stops what it started

• k in the table stops the whole process tree, as agent-ps stop already
  did. It was reading the agent rows rather than the process table, so a
  session's MCP servers and helpers were left orphaned.
• A stop is no longer reported as failed when it worked. SIGKILL is delivered
  after os.kill returns, and the check ran before the kernel had finished.
• When something does survive, the message names the PIDs instead of only
  counting them.

v0.6.0: Reclaiming what ended sessions leave behind

• agent-ps prune removes the parts of an ended session that are not the
  conversation: subagent transcripts, file history, task records, shell
  snapshots, request dumps. On the machine it was written on that is 96M across
  43 sessions, against 33M for deleting every session older than a month,
  because the sessions holding the space are recent rather than old.
• It reports by default and removes only with --apply. Transcripts are never
  removed, a session with a running process is never touched whatever its age,
  and agents that keep sessions in a database are left alone entirely.
• p in the table does the same for the selected row, after a confirmation
  naming what goes. On a running session it refuses and says why.
• The detail panel marks which parts of a session can be pruned, and the line
  above the keys reports the total when there is one.
• One confirmation path now serves every destructive key, rather than each
  growing its own.
• Release pages take their title and notes from the changelog instead of the
  commit list.

v0.5.0: The token counts four agents were already writing

• Token counts for Claude Code, Codex, Pi and CommandCode, which all record
  what a turn spent and were simply never read. Seven of the eight agents now
  report what a session cost.
• The counts are summed over a whole log, which is the one thing here that
  reads a file end to end, so it happens only for the session whose detail
  panel is open. Logs are append only, so a session that grew costs the bytes
  it just wrote rather than all of it again.
• agent-ps list now says when ps or lsof failed even where the table came
  back with rows, rather than only when it came back empty.
• The tests run in CI, across every supported interpreter.

v0.4.0

Full Changelog: v0.3.0...v0.4.0

v0.3.0

Full Changelog: v0.2.0...v0.3.0

v0.2.0

Full Changelog: v0.1.0...v0.2.0

v0.1.0

First public release.

Seven coding agents in one process table: Claude Code, Pi, CommandCode, Codex CLI, OpenCode, Hermes and GitHub Copilot. Each session shows its model, working directory, time since the last turn and disk footprint, and one key stops a process tree.

curl -fsSL https://raw.githubusercontent.com/mkhuda/agent-ps/main/install.sh | sh

A single zipapp with no dependencies, built by the standard library and tested on Python 3.8 through 3.14. The build is reproducible, so agent-ps below is exactly what ./build.sh produces from this tag, and CI checks that on every push.

See the changelog for what it does in more detail.