max_ticks limit. Everything the agent does in that run is recorded under the session, and every executor it opens is tagged with the session’s ID.
To watch and control sessions in the Workspace or Telegram, see Agent Loops.
Session IDs
Each run gets an ID made from the agent, the loop and a number:N is one more than the highest number already used by that loop’s sessions (or, for _e{N}, its dry runs).
The ID is also the controller_id on every executor the run creates. The risk gate refuses an executor create that doesn’t carry this exact ID, so every position can be traced back to the session that opened it.
Executors left by a stopped session
If you stop a loop with executors still running and start it again, the new session adopts them. On every tick it looks for RUNNING executors tagged with an earlier session of the same agent and loop whose session is no longer running. They appear in the tick data asadopted_executors, count toward its risk limits, and the risk gate lets it stop them.
Adoption is deliberately narrow:
- Executors of a session that is still running, or of another loop, are never adopted.
stop_executoron them is refused. - Held positions (from an executor closed with
keep_position=true) stay on the earlier session’s ID. Each tick’s position state is fetched for the current session’s ID only. - Run Once executors are never adopted.
Bots are adopted too: a session of the same loop adopts the still-running bots it recognizes as its own, either by name or because an earlier session of that loop recorded deploying them. It checks on any tick while it owns no bots, and again when a new deploy is recorded. Bots listed in the loop’s
disowned_bots.json are skipped.Dry Run and Run Once
Dry Run and Run Once each run a single tick and then stop. They don’t create a session folder or a journal. Each writes one file toloops/<loop>/dry_runs/experiment_N.md, with the same sections as a tick snapshot (see below).
Neither mode keeps a journal, so neither one triggers a drawdown pause or an emergency shutdown.
What a session records
A loop session writes toloops/<loop>/sessions/session_N/. The folder layout is on the Architecture page; this section explains what’s inside.
journal.md
The journal is the session’s running record. It has four sections:
The sections are capped (20 lines for Decisions, 1,000 for Ticks and Snapshots). Older lines aren’t deleted. They move to
journal_archive.md beside the journal, and a marker line records what was moved.
The agent reads and writes its own journal with trading_agent_journal_read and trading_agent_journal_write.
Snapshots
snapshots/snapshot_N.md is the complete record of tick N:
Condor keeps the 100 most recent snapshots per session and deletes older ones. The Workspace’s Runs tab reads these files.
Restarts
If Condor restarts while a loop is running, the old session is closed. It is never resumed.- By default, the loop stays stopped. If Condor exited unexpectedly, the session is marked interrupted and the journal notes it.
- If the loop has
restart_on_bootset (Resume after Condor restarts in the Start dialog, or the toggle in the Playbook tab), Condor starts a fresh session with a new number, using the loop’s config as it stands at that moment.
Learnings carry across sessions
The journal belongs to one session.loops/<loop>/learnings.md belongs to the loop and outlives every session.
Each new learning is checked against the existing ones and skipped if it’s a near-duplicate. Each section keeps its 20 most recent entries. Every tick’s prompt includes the Market Observations and Execution Notes sections; Retired Insights is not sent to the agent.
You can edit learnings in the agent page’s Playbook tab or in the file itself.
Learn more
- Architecture: folder layout, the tick pipeline and the risk engine
- Agent Loops: watching and controlling sessions
- Inventory: how a session’s positions and PnL are tracked

