/reply-and-close

/reply-and-close

After the lawyer has reviewed (and optionally edited) the AI draft in their Outlook AI Drafts folder, this command:

1. Sends the draft (now reviewed and edited by the lawyer)
2. Files the sent email + any new documents to SharePoint via the n8n Legal Copilot workflow
3. Captures any lawyer edits as a feedback entry appended to the shared memory file
4. Transitions the Jira ticket to Done

This is the only command that sends emails. Every other command stops at "draft created."

Constants

• Atlassian Cloud ID: fb47470f-f5c2-44bc-8182-f2a22f059adb
• n8n workflow ID: VAKq9Bra0RA0SdCO
• n8n production webhook: https://myposai.app.n8n.cloud/webhook/legal-copilot-filing

Input

$ARGUMENTS:

• <TICKET-KEY> (required) -- e.g., LEGAL-4321
• --dry-run (optional) -- show what would happen without sending or transitioning

---

Step 1: Find the AI draft in Outlook

Search the user's Outlook AI Drafts folder for a draft tagged with the ticket key. The /triage command tags drafts with a custom property x-mypos-legal-ticket: <TICKET-KEY>.

mcp__microsoft-365__outlook_email_search filtered to:

• folder: AI Drafts
• custom property match: x-mypos-legal-ticket eq '<TICKET-KEY>'

If no draft is found, ask the user:

No AI draft found for {ticket_key} in your AI Drafts folder.

Possible reasons:
  (a) The ticket hasn't been triaged yet -- run /triage {key} first.
  (b) You moved or renamed the draft -- paste the message ID or paste the draft content here.
  (c) Another lawyer triaged it on their machine -- the draft is in their AI Drafts, not yours.

If multiple drafts found (e.g., re-triaged multiple times), pick the most recent and warn the user.

---

Step 2: Read the current draft

Get the draft body, subject, and recipient list. Compare against what /triage originally produced (parse the AI Triage Jira comment for the original draft text).

Detect lawyer edits:

Comparison	Disposition
Identical	No edits captured. Continue.
Whitespace/formatting only	No edits captured. Continue.
Material changes	Capture as feedback in Step 6 below.

---

Step 3: Confirm before sending

Show the lawyer:

About to send this email and close {ticket_key}.

To: {recipient list}
Subject: {final subject}
Body preview:
  {first 300 chars of body}

Attachments: {count} files
Material edits since AI draft: {yes/no}

Confirm? (y / n / show-full-body)

If --dry-run, print the above and stop without sending. Otherwise wait for explicit y.

---

Step 4: Send the email

mcp__microsoft-365__outlook_email_send -- sends the draft, removes it from AI Drafts, places the sent copy in the user's Sent Items.

Capture the sent message ID for filing.

---

Step 5: File the final email + new documents to SharePoint

Invoke the sharepoint-filer skill (n8n-backed -- see skills/sharepoint-filer/SKILL.md) with:

• ticket_key
• ticket_summary (for filenames)
• matter_type (from the AI Triage Jira comment)
• files:
  • The sent email rendered as .eml (download via M365 MCP, encode as base64 with mime_type: message/rfc822, role: final_email)
  • Any new attachments since /triage ran (skip ones already filed -- list the case folder via mcp__microsoft-365__sharepoint_list_items first)
• memory_notes -- a brief markdown block summarising:
  • That the ticket was closed
  • The recipient list
  • Whether material edits were captured (and the category if so)
• triage_metadata -- pass through the metadata from the AI Triage comment

The skill calls the n8n workflow which uploads the .eml to myPOS Legal/{matter_folder}/{ticket_key}/ and appends a ## Case: {ticket_key} close-out entry to legal_copilot_memory.md.

If the workflow returns success: false, halt before transitioning the ticket to Done. Surface the error and let the lawyer retry.

Capture the returned sharepoint_folder_url, documents_filed[].sharepoint_url, and memory_file_url for use in Step 7.

---

Step 6: Capture lawyer edits as feedback (memory file)

If material edits were detected in Step 2, the memory_notes block from Step 5 should already include a structured feedback section. Make sure it has:

### Lawyer feedback ({ticket_key})
**Skill used:** {legal-triage-* skill, from AI Triage comment}
**AI draft excerpt:** "{first material-changed sentence as written by AI}"
**Lawyer changed to:** "{first material-changed sentence as sent}"
**Category:** {best guess: factual correction | tone | scope | jurisdiction-specific | other}
**Captured by:** /reply-and-close

The n8n workflow appends this to myPOS Legal/Claude skills memory/Copilot/_knowledge/legal_copilot_memory.md under the case entry. The next /triage-board run will read this back from SharePoint and surface emerging patterns.

This gap (AI draft vs. what the lawyer actually sent) is the highest-signal training data we have.

---

Step 7: Update Jira

Post a final comment on the ticket:

**Reply sent and ticket closed**

Sent to: {recipient list}
Filed to SharePoint: [{matter_type}/{ticket_key}/]({sharepoint_folder_url})
- Final email: [{final.eml}]({sharepoint_url_of_final_eml})
- {N} new attachments (if any)

{if material edits: "Lawyer-edited draft -- {summary of edits} -- captured to memory file."}

Memory file: [legal_copilot_memory.md]({memory_file_url})

*Closed by Legal Copilot via n8n workflow VAKq9Bra0RA0SdCO.*

Then transition to Done:

1. mcp__atlassian__getTransitionsForJiraIssue to find the Done transition ID.
2. mcp__atlassian__transitionJiraIssue to apply it.

---

Step 8: Summarise

Replied and closed {ticket_key}:
  Sent: ✓ to {recipient list}
  Filed to SharePoint: {sharepoint_folder_url}  (final email + {N} new attachments)
  Memory file updated: {memory_file_url}
  Jira: transitioned to Done
  {if edits captured: "Lawyer edits captured ({category}) -- will improve future drafts after the next /triage-board run"}

---

Hard rules

• NEVER send without explicit y confirmation in Step 3 (unless --dry-run, which doesn't send).
• NEVER bypass the Devil's advocate review -- if the original /triage flagged human_review_required = true, refuse to send and ask the user to confirm they've reviewed the flag.
• NEVER close a ticket where the AI Triage comment says risk gates triggered unless the lawyer has explicitly added a comment acknowledging the risk and approving the send. Look for a comment with text like approved for send or risk reviewed from the lawyer in the last hour.
• NEVER auto-resolve risk-flagged tickets even with confirmation -- always require an explicit lawyer comment in Jira saying the risk has been reviewed.
• ALWAYS file the final email to SharePoint via the n8n workflow before transitioning to Done. If the workflow returns success: false, do not transition -- the audit trail must be complete on SharePoint before the ticket closes.
• NEVER bypass the n8n workflow with a direct M365 SharePoint write. The workflow is the only sanctioned write path.
• NEVER fall back to local-Desktop saving on workflow failure. Surface the error and let the lawyer retry.
• NEVER fabricate the recipient list -- it must come from the actual draft.