Glossary

• Account: A signed-in connection to Linear, Jira or GitHub (called a connection in the code). Sign-in verifies the credentials and records who they belong to. Triggers reference an account instead of storing their own credentials, so one account can feed many triggers. See Accounts.
• Agent step: A step that renders a prompt and runs it on a harness with a model. See Step types.
• Approval step: A step that pauses the run until someone clicks Approve or Reject. The approver’s comment is passed downstream. It can also let the approver pick from numbered options an earlier step wrote. See Step types and Approvals.
• Attempt: One execution of a step. Retries, loop-backs and run retries each add an attempt, with its own log. See Runs & statuses.
• Base branch: The branch a worktree or clone starts from and that PRs usually target. Default main. {{run.baseBranch}}.
• Branch name template: Template for the per-run git branch. Default jimothy/{{issue.key | slug}}. See Workspaces.
• Continue on error: Step option: if the step fails, its dependents still run and the run can still succeed. See Retries, timeouts & conditions.
• DAG: Directed acyclic graph. Steps and their Runs after dependencies form one; steps whose dependencies are done run in parallel. See DAG & parallelism.
• Feedback loop / loop back: When a step with On failure, loop back to fails, the target step and everything downstream of it run again, at most Max loops times. See Loops & verdicts.
• Harness: How an agent step is executed: a CLI (Claude Code, Codex, Gemini CLI, Aider, a custom command), an API (Claude API, OpenAI-compatible), or the Simulator. See Harnesses.
• Inbox: The in-app list of notification events (run started/succeeded/failed, approvals, trigger errors). See Notifications.
• Issue: The unit of work a run is about: from a tracker, a webhook, a schedule, or typed in manually. Exposed to templates as {{issue.*}}. A manual run without a key gets TASK-<n>.
• Iteration: How many times a step has been re-entered by a feedback loop. {{loop.iteration}}, 0 on the first pass.
• Pass / fail pattern (verdict): Regexes matched against a step’s output that turn a successful process into a pass or fail. See Loops & verdicts.
• Pipeline: A named list of steps plus workspace settings, variables, subagents and a concurrency limit. See Pipelines overview.
• Project: A group of pipelines sharing a workspace, commands, variables and env. See Projects.
• Run: One execution of a pipeline for one issue. See Runs & statuses.
• Run only if (runIf): A template on a step; the step is skipped unless it renders truthy. See Retries, timeouts & conditions.
• Shell step: A step that runs a command with the system shell in the workspace. See Step types.
• Simulator: A built-in fake harness for demos and dry runs. Models fast, realistic and flaky control its behavior.
• Step: One node in a pipeline: agent, shell or approval. Identified by its Step id, which templates use as {{steps.<id>.output}}.
• Subagent: A specialised helper an agent step can delegate to mid-task, defined per pipeline. Passed to Claude Code via --agents. See Subagents.
• Template: Text with {{ }} placeholders, filters and {{#if}} blocks, used in prompts, commands, approval messages, conditions and branch names. See Templates.
• Trigger: Watches a source (Linear, Jira, GitHub, a cron schedule, a webhook) and starts runs of a pipeline. See Triggers.
• Variables: Named values available as {{vars.NAME}}, from project commands, project variables, pipeline variables and per-run overrides. See Variables & precedence.
• Workspace: The directory a run’s steps execute in: existing folder, git worktree, fresh clone or empty folder. See Workspaces.
• Write-back: Commenting on and transitioning the source issue when a run starts, succeeds or fails. See Write-back.