Generate/Update PR Description

Generates or updates PR description with incremental information, auto-updates title, and links Linear tickets.

Prerequisites

# Check project setup (thoughts, CLAUDE.md snippet, config)
if [[ -f "${CLAUDE_PLUGIN_ROOT}/scripts/check-project-setup.sh" ]]; then
  "${CLAUDE_PLUGIN_ROOT}/scripts/check-project-setup.sh" || exit 1
fi

Process:

1. Read PR description template

# Check if template exists
if [ ! -f "thoughts/shared/pr_description.md" ]; then
    echo "❌ PR description template not found"
fi

If missing:

❌ PR description template missing

Your humanlayer thoughts setup is incomplete. Create a template at:
  thoughts/shared/pr_description.md

See the PR description template you created earlier for reference.

Read template fully to understand all sections.

2. Identify target PR

If argument provided:

• Use that PR number: /describe_pr 123

If no argument:

# Try current branch
gh pr view --json number,url,title,state,body,headRefName,baseRefName 2>/dev/null

If no PR on current branch OR on main/master:

# List recent PRs
gh pr list --limit 10 --json number,title,headRefName,state

Ask user: "Which PR would you like to describe? (enter number)"

3. Extract ticket reference

From multiple sources:

# 1. From branch name
branch=$(gh pr view $pr_number --json headRefName -q .headRefName)
if [[ "$branch" =~ ([A-Z]+)-([0-9]+) ]]; then
    ticket="${BASH_REMATCH[0]}"
fi

# 2. From PR title
title=$(gh pr view $pr_number --json title -q .title)
if [[ "$title" =~ ([A-Z]+)-([0-9]+) ]]; then
    ticket="${BASH_REMATCH[0]}"
fi

# 3. From existing PR body
body=$(gh pr view $pr_number --json body -q .body)
if [[ "$body" =~ Refs:\ ([A-Z]+-[0-9]+) ]]; then
    ticket="${BASH_REMATCH[1]}"
fi

4. Read existing descriptions

Read current PR body from GitHub:

current_body=$(gh pr view $pr_number --json body -q .body)

Read saved description (if exists):

saved_desc="thoughts/shared/prs/${pr_number}_description.md"
if [ -f "$saved_desc" ]; then
    # Read fully
    # Note what sections exist vs what's new
fi

Check for metadata header:

<!-- Auto-generated: 2025-10-06T10:30:00Z -->
<!-- Last updated: 2025-10-06T14:45:00Z -->
<!-- PR: #123 -->
<!-- Previous commits: abc123,def456 -->

5. Gather comprehensive PR information

# Full diff
gh pr diff $pr_number

# Commit history with messages
gh pr view $pr_number --json commits

# Changed files
gh pr view $pr_number --json files

# PR metadata
gh pr view $pr_number --json url,title,number,state,baseRefName,headRefName,author

# CI/CD status
gh pr checks $pr_number

6. Analyze changes incrementally

If this is an UPDATE (saved description exists):

# Extract previous commit list from metadata
prev_commits=$(grep "Previous commits:" $saved_desc | sed 's/.*: //')

# Get current commits
current_commits=$(gh pr view $pr_number --json commits -q '.commits[].oid' | tr '\n' ',' | sed 's/,$//')

# Compare
new_commits=$(comm -13 <(echo "$prev_commits" | tr ',' '\n' | sort) <(echo "$current_commits" | tr ',' '\n' | sort))

Analysis:

• Identify what's NEW since last description
• Deep analysis of:
  • Code changes and architectural impact
  • Breaking changes
  • User-facing vs internal changes
  • Migration requirements
  • Security implications

7. Merge descriptions intelligently

Auto-generated sections (always update):

• Summary - regenerate based on ALL changes
• Changes Made - append new changes, preserve old
• How to Verify It - update checklist, rerun checks
• Changelog Entry - update to reflect all changes

Preserve manual edits in:

• Reviewer Notes - keep existing unless explicitly empty
• Screenshots/Videos - never overwrite
• Manually checked boxes - preserve [x] marks for manual steps
• Post-Merge Tasks - append new, keep existing

Merging strategy:

## Changes Made

### Backend Changes

[Existing changes from previous description]

**New changes** (since last update):

- [New change 1]
- [New change 2]

### Frontend Changes

[Existing + new merged together]

Add change summary at top:

<!-- Auto-generated: 2025-10-06T15:00:00Z -->
<!-- Last updated: 2025-10-06T15:00:00Z -->
<!-- PR: #123 -->
<!-- Previous commits: abc123,def456,ghi789 -->

---

**Update History:**

- 2025-10-06 15:00: Added validation logic, updated tests (3 new commits)
- 2025-10-06 10:30: Initial implementation (5 commits)

---

8. Add Linear reference

If ticket found:

## Related Issues/PRs

- Fixes https://linear.app/{workspace}/issue/{ticket}
- Related to #NNN (reference sibling work by its **GitHub PR number**)

CTL-623 — sibling reference format (REQUIRED): When referencing related/sibling work in prose, reference it by its GitHub PR number (#NNN), never by a bare Linear token (TEAM-NNN) or a Linear issue URL. A bare sibling TEAM-NNN token in the body is auto-linked by Linear's GitHub integration and drags that sibling's workflow status (Done → Implement) on PR open/merge. Do not emit bare sibling Linear tokens in prose. The own ticket's Fixes https://linear.app/... line is correct and stays — that link/transition is intended. Sibling neutralization is handled mechanically by the guard block appended at write-back time (see below).

Get Linear ticket details via direct SQL against the replica (see the linearis skill's "Reading Linear" section). Extract title and description with jq. Use ticket title and description for context.

9. Generate updated title

Title generation rules:

# If ticket exists, read its title via direct SQL against the replica
# (linear_read_ticket — replica-first, loud linearis fallback; CTL-1397).
source "${CLAUDE_PLUGIN_ROOT}/scripts/lib/linear-read-replica.sh"
if [[ "$ticket" ]]; then
    ticket_title=$(linear_read_ticket "$ticket" 2>/dev/null | jq -r '.title // empty')
    if [[ -n "$ticket_title" ]]; then
        title="$ticket: ${ticket_title:0:60}"
    else
        # Fallback: generate title from branch name + commits
        title="$ticket: $(echo "$branch" | sed "s/^.*$ticket-//" | tr '-' ' ')"
    fi
else
    # No ticket: generate from primary change
    title="Brief summary of main change"
fi

Auto-update without prompt - title is auto-generated section.

10. Run verification checks

For each checklist item in "How to Verify It":

# Example: "- [ ] Build passes: `make build`"
# Extract command: make build

# Try to run
if command -v make >/dev/null 2>&1; then
    if make build 2>&1; then
        # Mark as checked
        checkbox="- [x] Build passes: \`make build\` ✅"
    else
        # Mark unchecked with error
        checkbox="- [ ] Build passes: \`make build\` ❌ (failed: $error)"
    fi
else
    # Can't run
    checkbox="- [ ] Build passes: \`make build\` (manual verification required)"
fi

Common checks to attempt:

• make test / npm test / pytest
• make lint / npm run lint
• npm run typecheck / tsc --noEmit
• make build / npm run build

Document results:

• ✅ if passed
• ❌ if failed (with error)
• Manual required if can't automate

11. Save and sync

IMPORTANT: Document Storage Rules

• ALWAYS write to thoughts/shared/prs/ for PR descriptions
• NEVER write to thoughts/searchable/ — this is a read-only search index

Save description:

# Add metadata header
cat > "thoughts/shared/prs/${pr_number}_description.md" <<EOF
<!-- Auto-generated: $(date -u +%Y-%m-%dT%H:%M:%SZ) -->
<!-- Last updated: $(date -u +%Y-%m-%dT%H:%M:%SZ) -->
<!-- PR: #$pr_number -->
<!-- Previous commits: $commit_list -->

[Full description content]
EOF

Sync thoughts:

humanlayer thoughts sync

12. Update PR on GitHub

CRITICAL: NO CLAUDE ATTRIBUTION

Befo