菜单
Session start checklist. Runs inbox check, board state review, knowledge freshness audit, and intent declaration before pulling any work. The session counterpart to /lets-wrap.
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
| name | lets-start |
| description | Session start checklist. Runs inbox check, board state review, knowledge freshness audit, and intent declaration before pulling any work. The session counterpart to /lets-wrap. |
| disable-model-invocation | true |
Work through each step in order. Do not pull work from the board until step 6 is complete.
Read your project's operating instructions (e.g. a root CLAUDE.md or equivalent config file) and execute any session-start instructions it contains. Skipping this means missing operational steps that live outside this skill.
If you are the orchestrator agent, read the orchestrator operating model file and run any startup routines (inbox cron setup, communication conventions, etc.).
Check for agent-to-agent escalations waiting for you:
# Use your board tool to check your agent inbox
board-tool inbox <YOUR_INBOX_PREFIX>
For each inbox card found:
If inbox is empty, move on.
Review your open-actions.md file (if one exists). For each entry:
If the file has no entries or does not exist: move on.
# Use your board tool to check WIP and card aging
board-tool wip-age
Note:
Closed column purge check: If your board tool has a hard limit on closed cards per cell, check the count and archive old cards before hitting the limit.
Quick scan of project agent outputs since your last session. Purpose: catch coordination issues, conflicts, or shipped work that needs orchestrator attention.
Read the most recent daily log entry. For each project agent entry, check:
If conflicts or concerns found: note them for your intent declaration. If clean: move on.
This is a 2-minute scan, not deep analysis. The goal is to catch what the product owner would otherwise have to flag.
Check the age of each domain knowledge file against recent repository activity. For each stale domain (14+ days old, repository active):
Maintain a domain-to-repository mapping table so this step is repeatable across sessions.
State what you intend to do this session and why, based on:
Example format:
"Intent: Pull [card title] (#NNN) from Ready; it is the highest-priority unblocked card under initiative #NNN. Also reviewing #NNN in Doing which has been there Xd without a comment."
Then proceed. Do not wait for the product owner to confirm.
End-of-session wrap-up checklist. Prompts the agent to follow through on knowledge capture, board hygiene, and memory updates before signing off. Use when ending a working session.
Security penetration test + GDPR/SOC 2/ISO 27001 compliance audit. Targets a single project or the estate. Produces categorised findings, board cards, and regression tests. Invocable by any agent or the product owner.
Agent-only skill. After feature tests pass, walks the agent through regression test vectors for any new or modified API routes. Produces regression tests and documents coverage on the card. Invoked by the regression-gate hook, not by humans.
Set up a polling loop on a card where you are waiting for another agent to respond. Auto-blocks the card after a timeout with no response.
Exploratory testing from a user's perspective. AI-reasoned edge cases, happy path validation, non-happy-path scenarios, regression quality assessment. The closest thing to having a human test team. Invocable by any agent or the product owner.