菜单
Fetches Slack thread context, researches the topic using agent skills and MCP tools, and presents findings for the user to decide on a reply. Use when given a Slack message permalink to understand and respond to a conversation.
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
Deeply gather context from an Azure DevOps work item by following its relations, linked PRs, hyperlinks, and comments. Use whenever a command needs the full picture behind a work item -- planning features, reviewing PRs, triaging, or estimating.
Common steps for creating a new Azure DevOps work item in the FundGuard project, including iteration assignment, parent linking, and triage. Called by create-task, create-bug, and request-environment-access skills — not invoked directly by the user.
Mechanical estimate-and-transition of an Azure DevOps work item to the Triaged state. Called programmatically by skills that create or triage work items -- not invoked directly by the user. For user-initiated triage, use the triage skill instead.
Transitions an Azure DevOps work item to the Active state. Use when starting work on a work item, given an explicit ID or one inferred from the current feature branch.
Marks an Azure DevOps work item as Blocked and links it to its predecessor blocker. Use when a work item cannot proceed until another work item is completed.
Investigates bugs to find root cause, designs a TDD-based fix plan, and implements after approval. Use when given a bug ticket or description of defective behaviour that needs investigation and fixing in the codebase.
| name | answer-slack |
| description | Fetches Slack thread context, researches the topic using agent skills and MCP tools, and presents findings for the user to decide on a reply. Use when given a Slack message permalink to understand and respond to a conversation. |
| disable-model-invocation | true |
Given a Slack message link, fetch the conversation context, analyse the topic, research using relevant agent skills and MCP tools, and present findings to the user. The user digests the information and decides whether to ask for a Slack reply draft.
A Slack message permalink in the form:
https://<workspace>.slack.com/archives/<channel_id>/p<ts_without_dot>
Require Ask mode following the mode-gate skill. This command is primarily informational -- it researches a Slack conversation and presents findings without making changes. If the user later wants to post a Slack reply (step 7), they should switch to Agent mode for that.
Extract channel_id and reconstruct thread_ts from the URL:
channel_id is the path segment after /archives/.thread_ts is derived from the p-prefixed timestamp by inserting a dot before the last 6 digits (e.g. p1234567890123456 becomes 1234567890.123456).If the link is malformed or missing, ask the user and stop.
slack_get_thread_replies with the extracted channel_id and thread_ts to retrieve the full thread.slack_get_channel_history scoped around the timestamp to capture surrounding messages.slack_get_user_profile for each unique user ID that appears in the messages.Scan the messages for links to external resources and fetch each one to build a complete picture:
| Link type | How to resolve |
|---|---|
Slack message permalink (/archives/<channel>/p<ts>) | Parse the link (same logic as step 1) and call slack_get_thread_replies to pull that thread |
Azure DevOps work item (_workitems/edit/<id>) | Extract the ID and fetch via get_work_item |
Azure DevOps pull request (_git/<repo>/pullrequest/<id>) | Extract the ID and fetch via get_pr_details |
Azure DevOps build / pipeline (_build/results?buildId=<id>) | Extract the ID and fetch via get_build_details |
| Datadog incident or monitor URL | Extract the ID and fetch via the Datadog MCP tools |
| GitHub PR or issue | Extract the owner/repo/number and fetch via gh CLI or GitHub MCP |
| Generic URL | Fetch with WebFetch if the content is likely to add useful context |
Stop recursing when a linked resource doesn't itself contain further actionable links, or when the same resource has already been fetched.
Read the thread and determine:
Based on the domain identified in step 4, use the appropriate agent skills and MCP tools to research. Explain which skills were selected and why.
| Domain | Artifacts / tools to consider |
|---|---|
| Code questions | Search the codebase; apply the functional-typescript skill for TypeScript questions |
| Incidents | Follow the incident-response skill's investigation structure; query Datadog via MCP |
| Build failures | Follow the triage-build skill's investigation pattern; query Azure DevOps and Currents via MCP |
| Work items | Fetch details from Azure DevOps via MCP |
| Process / tooling | Search existing commands and skills for relevant guidance |
| General | Research using available MCP tools and codebase search |
When a topic spans multiple domains, investigate each relevant angle.
Present a structured summary to the user (not as a Slack message draft):
## Slack Thread: <channel name> -- <brief topic>
**From**: <who asked> | **Domain**: <domain> | **Response expected**: yes / no
### Context
<concise summary of the conversation so far>
### Research
<findings from step 5 -- evidence, code references, data points>
### Suggested Answer Direction
<bullet points outlining what a good reply would cover, without composing the reply itself>
After presenting findings, offer to compose a Slack reply. If the user accepts:
slack_reply_to_thread only after explicit approval.Wait for user confirmation before taking any action.
Follow the continuous-improvement skill.