一键导入
cli-task-lifecycle
Use the `at` CLI to create, execute, monitor, and complete tasks on the Auto-Tundra daemon. Covers the full bead lifecycle from sling to done.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use the `at` CLI to create, execute, monitor, and complete tasks on the Auto-Tundra daemon. Covers the full bead lifecycle from sling to done.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when troubleshooting build failures, test failures, daemon connectivity issues, or integration errors in Auto-Tundra.
Use the Auto-Tundra HTTP API directly (via curl) for operations not exposed in the CLI, such as task updates, terminal management, GitLab/Linear queries, and Kanban board manipulation.
Use when adding features, fixing bugs, or refactoring code in any Auto-Tundra crate. Covers build, test, lint, and validation workflows.
Use for GitHub/GitLab/Linear wiring in rust-harness to remove stub-token paths, enforce env-driven credential resolution, and add regression tests for real-client behavior with safe fallbacks.
Use for local Ollama-first execution with queued multi-agent CLI workflows. Configures auto-tundra for one-human-overseer and many-agent-subscriber code refinery mode.
Use when developing or refactoring at-cli. Runs deterministic smoke checks for run/agent/skill/doctor commands, including --dry-run and --out artifact paths.
| name | cli-task-lifecycle |
| description | Use the `at` CLI to create, execute, monitor, and complete tasks on the Auto-Tundra daemon. Covers the full bead lifecycle from sling to done. |
| allowed_tools | ["Bash","Read"] |
| references | ["/Users/studio/rust-harness/docs/PROJECT_HANDBOOK.md"] |
Use this skill when you need to create a task, execute it, check status, or complete work via the Auto-Tundra CLI.
cargo run --bin at-daemon (or check with at doctor -p .).ANTHROPIC_API_KEY, OPENROUTER_API_KEY, or OPENAI_API_KEY. Alternatively, have Ollama running locally — the daemon will automatically use it as the local-runtime AI profile.claude, codex, opencode) must be installed and in $PATH if using specific agent roles.at doctor -p /Users/studio/rust-harness -j
at status
# Simple
at sling "Fix login bug"
# With priority lane
at sling "Security patch" --lane critical
# Create + execute
at run \
-t "Wire GitLab env handling" \
-s integration-hardening \
-p /Users/studio/rust-harness
# Create without executing
at run \
-t "Wire GitLab env handling" \
-s integration-hardening \
-p /Users/studio/rust-harness \
--no-execute
at run \
-t "Plan integration wave" \
-s wave-execution \
-p /Users/studio/rust-harness \
--dry-run --emit-prompt -j
at agent run \
-r qa-reviewer \
-t "Audit auth edge cases" \
-s integration-hardening \
-p /Users/studio/rust-harness \
-m sonnet -n 2 -j
at status # Check all beads and active agents
at hook <bead-id> # Move from Backlog -> Hooked
at sling <bead-id> # Move from Hooked -> Slung (Active processing)
at review <bead-id> # Move from Slung -> Review
at done <bead-id> # Move from Review -> Done
at nudge <agent-id> # Restart stuck agent (use sparingly)
Note on State Transitions: The event bus enforces strict transitions. A task must flow:
Backlog -> Hooked -> Slung -> Review -> Done. You cannot jump states (e.g., Hooked directly to Done will throw a 400 Bad Request).
-j flag for machine-readable JSON output.-o /tmp/result.json to write output to a file for later reading.at doctor will report it cleanly.at run or at agent run without -p (project path) — it defaults to . which may be wrong.at doctor to verify connectivity.at agent run -r must map to valid task categories (e.g., qa-reviewer → QA, builder → Build).