Jobs plugin (jobs)

Scheduled job management: cron and one-time jobs, drafts, and AI-assisted creation and revision.

Command: /jobs (alias for the jobs plugin)

Demo

Interactive Demo

Commands

In the web Jobs widget, use Show logs from a job's ⋮ menu to open its complete execution log in a modal. Newest runs appear first, while events inside each run remain chronological. OpenCode backend status, tool commands, completion output, and backend errors are recorded while a run is active; use Refresh to load new events. Messages and structured details are not truncated.

Drafts

Job creation uses a draft/confirm flow:

• Use /jobs ai <prompt> to create a draft (e.g. "send me a brief every day at 8am").
• The bot returns a preview and a Draft ID.
• Use /jobs confirm <draft_id> to create the job, or /jobs revise <draft_id> <corrections> to have the AI adjust the draft, or /jobs discard <draft_id> to cancel.

Structured scheduler capability requests already contain a validated schedule and task, so they create jobs directly without the AI draft flow. scheduler:v2 additionally supports plugin-tool tasks, which execute a validated plugin tool directly without asking an agent to infer or issue the tool call.

OpenCode tools

When using the OpenCode backend, the agent can use:

• jobs__list — List jobs (optionally enabled only). Use this to resolve job IDs before other operations.
• jobs__show — Show full details for a job by ID.

Creation and mutation (confirm, revise, discard, enable, disable, delete) are done via the bot commands above, not via tools.

Engine and scheduling

• Engine: The plugin runs a small scheduler every 60 seconds. Agent-prompt jobs use the configured agent; scheduler:v2 plugin-tool jobs invoke the target tool directly and have no agent or shell-tool timeout in their execution path.
• Model routing: A job may pin a workspace, model source, and model. Unset options inherit the workspace and current source/model on every run. Core and PPQ jobs can run concurrently in separate sessions; different models of the same source share its OpenCode server. Source servers stay open until AppWeaver exits. scheduler:v3 exposes these options to other plugins; V1/V2 keep their existing task shapes.
• Sessions: Jobs start a fresh session on every run by default. /jobs update <id> --sticky-session true reuses the session on later runs. Changing the resolved source or workspace starts a fresh session instead of resuming on the wrong server.
• Existing jobs keep their saved data. The plugin adds optional/defaulted SQLite columns; historical model values are not treated as new model overrides unless explicitly configured.
• Long-running jobs: A job can have only one active run. Later scheduler ticks skip it while it is running, and manual execution reports that it is already active. Missed cron intervals are skipped; the next run is calculated from completion time.
• Running a job: The agent or plugin-tool output is stored in job_runs, and the result is sent by Nostr DM and Web Push. Use /jobs history <id> for run summaries and /jobs logs <id> for timestamped execution and delivery events.
• Recovery: Runs left in running state after AppWeaver stops are marked as interrupted when the plugin starts again.

Plugin data

• Database: The plugin uses its own SQLite DB at plugins/jobs/db.sqlite.
• Tables: jobs, job_runs, job_run_logs, and job_drafts are stored in that DB, separate from core bot job state. Runs and logs are deleted automatically when their job is deleted.

Architecture

See local architecture and source map.

Local modules

• types
• drafts
• db
• commands