Agentlog skill

Treat Agentlog as the task record. The starting prompt contains the task ID. The agentlog CLI is a local command that reads and writes tasks in the Agentlog database.

Start

Run agentlog task show <task-id> --json before editing.
Use the returned title, description, and checklist as the work scope.
Keep the task in running while work can continue.

While working

Record meaningful milestones with agentlog task activity <task-id> --kind <kind> --message "<message>".
Use short kinds such as file, test, build, or note.
Mark a checklist item complete with agentlog task check <task-id> <item-id>.
Read the task again after every write: agentlog task show <task-id> --json.

Tasks

Full CRUD of tasks is available. New tasks start in backlog.

Create

agentlog task create --project <project-id|name> --title "<title>" [--description "<text>"] [--checklist "<item>"]

List

agentlog task list [--project <project-id|name>] [--status <status>] [--json]

Show

agentlog task show <task-id> [--json]

Update

agentlog task update <task-id> [--title "<title>"] [--description "<text>"] [--status <status>] [--branch <branch>] [--worktree-path <path>] [--checklist-add "<item>"] [--checklist-remove <item-id>]

Delete

agentlog task delete <task-id> --yes

--project is optional on create and list; when omitted it is inferred from the current directory. Run agentlog project list to see available projects.

Automations

Automations run a prompt or a saved workflow when a task moves to a status. Provide --prompt and/or --workflow-id.

backlog running waiting review done

Create

agentlog automation create --project <project-id|name> --name "<name>" --trigger-status <status> [--prompt "<text>"] [--workflow-id <id>] [--enabled|--disabled] [--model-provider <provider>] [--model-id <id>]

List

agentlog automation list [--project <project-id|name>] [--json]

Show

agentlog automation show <automation-id> [--json]

Update

agentlog automation update <automation-id> [--name "<name>"] [--trigger-status <status>] [--prompt "<text>"] [--workflow-id <id>] [--enabled|--disabled] [--model-provider <provider>] [--model-id <id>]

Delete

agentlog automation delete <automation-id> --yes

Status

When blocked on user input, approval, credentials, or an external service, run:
```
agentlog task status <task-id> waiting --note "<exact blocker>"
```

When the blocker clears, run:

agentlog task status <task-id> running --note "Work resumed"

When implementation and relevant checks are complete, run:

agentlog task status <task-id> review --note "<result and checks>"

Before stopping, run agentlog task show <task-id> --json again and update the task if its status, checklist, or notes are stale.

Rules

Never move a task to done. Only the user does that after review.

Do not change status merely because one command failed. Investigate first.
Do not claim tests or builds passed unless they ran successfully.
Keep notes factual and short.
If agentlog is unavailable, continue the work and say that task updates could not be recorded.