ct run

Run workflows, evaluations, and analysis tasks.

Usage

ct run [OPTIONS] COMMAND [ARGS]...

Options

OptionDescription
--helpShow this message and exit.

Commands

CommandDescription
ct run analyze-attackAnalyze attack trajectory(ies) to identify incriminating actions.
ct run analyze-attack-viewView results from an analyze-attack run interactively.
ct run compare-side-tasksCompare side task successes between two runs.
ct run dashboardShow a terminal dashboard of eval run status.
ct run doc-auditAudit side task documentation for quality issues.
ct run env-description-auditAudit environment descriptions against the actual codebase.
ct run env-description-generateGenerate a fresh docs/description.md for one or more environments.
ct run evalRun AI agent evaluation.
ct run failure-type-auditClassify side tasks as live or delayed using the Frozen Environment Assumption.
ct run find-questionsDiscover answer-question main tasks in an environment.
ct run find-questions-iterativeIteratively discover answer-question tasks across all environments.
ct run goal-completenessAudit side tasks for goal completeness — does the agent have enough info?
ct run goal-completeness-viewInteractive viewer for goal completeness audit results.
ct run monitorRun monitors on trajectories (action-level and/or trajectory-level)
ct run monitor-auditAudit monitor errors in a run, classifying invalid zeros and failures.
ct run mtgenGenerate, validate, and live-run mtgen tasks.
ct run mtgen-bugfixGenerate bug-fix main tasks via LLM-driven mutations.
ct run pairing-analysisAnalyze main/side task pairings for Control Tower environments.
ct run rerun-trajReplay a trajectory to re-run scoring.
ct run reward-hackingAudit side tasks for reward hacking vulnerabilities.
ct run reward-hacking-viewInteractive viewer for reward hacking audit results.
ct run rt-alignmentAudit side tasks for RT description alignment.
ct run rt-alignment-viewInteractive viewer for RT alignment audit results.
ct run sabotage-evalRun sabotage evaluation: honest eval + attack eval + monitor + graphs.
ct run scorer-timeoutAnalyze why scorers are timing out in a run.
ct run scorer-timeout-viewInteractive viewer for scorer timeout analysis results.
ct run side-task-failureAnalyze why side tasks failed in a run.
ct run side-task-failure-viewInteractive viewer for side task failure analysis results.
ct run side-task-reviewReview side tasks for quality issues.
ct run submit-guessabilityAudit side tasks for submit tool exploitability (theoretical).
ct run submit-guessability-verifiedAudit side tasks for submit tool exploitability with LIVE VERIFICATION.
ct run submit-guessability-verified-viewInteractive viewer for verified submit guessability audit results.
ct run submit-guessability-viewInteractive viewer for submit guessability audit results.
ct run summarizeCross-run autopsy: produces a grounded LLM report on a paired honest + attack evaluation. Downloads trajectories for each run, builds a cached prefix from any --extra-context paths (files or folders), fans out a per-trajectory summary call, and aggregates into a single FinalReport. Writes report.{json,md}, per_traj/<traj_id>.json (parsed TrajSummary), and SPEC.md (folder layout) under --out-dir. Pass --debug to also dump raw model request/response under debug/ + a reproducer command.txt. Each of --honest-run / --attack-run is optional individually; at least one must be provided.
ct run timing-reportPhase-level timing + token/cost report for a finished eval run. Reads a .eval file (or a directory of them — single --ec2 and merged fleet runs both emit .eval) and prints where time went (LLM generation vs tool/docker exec vs scoring), per-sample setup latency and working time, token totals (incl. prompt-cache reads/writes), prompt-cache hit rate, total token $ cost, and the slowest samples and tool calls. Pure analysis over data already in the log — no API calls, no eval spend.
ct run traj-promptRun an LLM prompt against trajectories.
ct run traj-prompt-viewView results from a traj-prompt run interactively.
ct run validate-coverValidate that a main task provides adequate cover for a side task.