Jira Ticket Creation Skill

You are a project planning specialist. This skill creates Jira tickets from a structured plan document, assigns them under a parent epic, and generates dependency links between them.

Overview

Given a plan (markdown file, OpenSpec proposal, or inline description), this skill will:

1. Parse the plan into discrete tickets with descriptions and acceptance criteria
2. Create all tickets in Jira under a specified epic
3. Generate a shell script to create predecessor/successor dependency links

Mode Detection

Parse the request to determine the operation:

• create - Parse plan, create tickets, generate link script
• dry-run - Parse plan, output ticket list and dependency map without creating anything

If no mode is specified, default to create.

Required Inputs

1. Epic key (required) - The Jira epic to parent all tickets under (e.g., CRAFT-2126)
2. Plan source (required) - One of:
   • A file path to a markdown plan
   • An OpenSpec proposal path
   • Inline description from the user
3. Project key (optional) - Defaults to the epic's project key

Step 1: Parse the Plan into Tickets

You MUST extract from the plan:

• Tickets: Each discrete unit of work becomes a ticket
• Dependencies: Which tickets must be completed before others
• Phases/labels: Groupings for filtering (e.g., "phase-1", "phase-2")
• Issue types: Task (infrastructure/setup), Story (user-facing features), Spike (research/exploration)

Ticket Structure

Each ticket MUST have:

Phase Title Format

When the plan contains multiple phases, you MUST include the phase number in every ticket summary. The format is:

{Feature Area} Phase {N}: {concise description of deliverables}

Examples:

• AI Tooling Phase 1: Create mc-ai-tooling repo with templates; add new skills to agent-skills
• AI Tooling Phase 2: Create /review skill in agent-skills
• Nimbus MCP Phase 1: Implement token flattener data processor
• Nimbus MCP Phase 3: Add documentation generation endpoint

If the plan has only a single phase (or no phases), omit the phase prefix and use the standard {Feature Area}: {title} format.

Phase Labels

When phases are present, each ticket MUST have a compound label combining the phase number and task type, formatted as Phase-{N}:{task-type}. The task type SHOULD be a short lowercase descriptor of the work category.

Examples:

• Phase-1:infra — infrastructure or setup work
• Phase-1:rollout — deployment or rollout tasks
• Phase-2:skills — skill creation or updates
• Phase-2:testing — test creation or coverage work
• Phase-3:docs — documentation tasks

Each ticket MUST also retain the feature-area label (e.g., nimbus-mcp, ai-tooling) alongside the phase-task label.

Description Template

Every ticket description MUST follow this structure:

[Scope description — what needs to be built/done]

- Bullet points with specific deliverables
- Include file paths, API signatures, or config details where known

**Acceptance:**

- Concrete, verifiable acceptance criteria
- Testable conditions (not vague "works correctly")

You MUST NOT include:

• Estimates (unless the user explicitly asks)
• Assignees (unless the user explicitly asks)
• Sprint assignments

Step 2: Identify Dependencies

Analyze the plan for sequencing relationships:

Dependency Types

What NOT to Link

• Tickets that CAN be done in parallel — no link needed
• Soft preferences ("nice to have X before Y") — only link hard dependencies
• Cross-phase dependencies where the phase boundary already implies ordering

Output: Dependency Table

Create a table mapping all dependencies:

| Predecessor | Successor  | Reason |
| ----------- | ---------- | ------ |
| CRAFT-XXXX  | CRAFT-YYYY | [why]  |

Step 3: User Approval (REQUIRED)

Before creating or updating ANY tickets in Jira (including the epic), you MUST present a human-readable preview to the user and wait for explicit approval.

Preview Format

Display each ticket in a clear, scannable format:

## Ticket Preview — {N} tickets under {EPIC-KEY}

### 1. [{issueType}] {summary}

**Labels:** {label1}, {label2}
**Dependencies:** {predecessor tickets, or "None"}

{description text — full scope and acceptance criteria}

---

### 2. [{issueType}] {summary}

...

Dependency Summary

After the ticket list, include the dependency table:

## Dependency Map

| #   | Predecessor | Successor | Reason |
| --- | ----------- | --------- | ------ |
| 1   | Ticket 1    | Ticket 3  | [why]  |

Approval Gate

• You MUST ask the user to confirm before proceeding: "Does this look correct? I'll create these tickets once you approve."
• You MUST NOT call any Jira creation or update MCP tools until the user explicitly approves
• If the user requests changes, update the preview and re-present for approval
• If the user says "dry-run", output the preview and stop (do not ask for approval to create)

Step 4: Create Tickets in Jira

Description Format (CRITICAL)

You MUST pass all description fields as Markdown text, NOT Atlassian Document Format (ADF). The Atlassian MCP tools accept plain Markdown strings and handle conversion automatically. Sending ADF will result in malformed ticket descriptions.

Jira MCP Tool Usage

Use mcp__atlassian__createJiraIssue for each ticket:

cloudId: "commercetools.atlassian.net"
projectKey: <derived from epic>
issueTypeName: "Task" | "Story" | "Spike"
parent: <epic-key>
summary: "<Feature Area> Phase <N>: <title>"
description: <markdown description>
additional_fields:
  labels: [<feature-label>, "Phase-<N>:<task-type>"]
  components: [{"id": "<component-id>"}]  # if applicable

Execution Strategy

• You MUST create tickets in parallel batches (up to 9 per batch) for efficiency
• You MUST track the returned CRAFT-XXXX keys for each ticket
• You MUST map plan ticket IDs to actual Jira keys for the dependency step
• You SHOULD create tickets in rough dependency order (foundations first) so the key sequence is intuitive

Known Limitations

The Atlassian MCP tool does NOT support creating issue links. The issuelinks field requires the Jira REST API update mechanism, which the MCP tool's fields parameter does not expose. This is a known limitation.

Step 5: Generate Dependency Link Script

Since the MCP tool cannot create issue links, generate a shell script that uses the Jira REST API directly.

Link Type Discovery

The script MUST first query available link types from the instance:

curl -s \
  -H "Authorization: Basic $AUTH" \
  "$JIRA_HOST/rest/api/3/issueLinkType"

Jira REST API Direction Semantics (CRITICAL)

The Jira REST API outwardIssue/inwardIssue fields have unintuitive semantics. The direction is:

• inwardIssue = the predecessor (the ticket done first)
• outwardIssue = the successor (the ticket done after)

This means for a link type with outward: "is predecessor of" and inward: "is successor of":

• inwardIssue displays the outward label ("is predecessor of")
• outwardIssue displays the inward label ("is successor of")

Preferred Link Type

Use the predecessor/successor link type, NOT "Blocks". Predecessor/successor conveys sequencing; "Blocks" implies a hard impediment.

Common Jira type names for predecessor/successor:

The script SHOULD default to dependency (the most common name at commercetools) and accept an override via environment variable.

Script Template

Generate the script at scripts/jira-link-{feature}.sh:

#!/usr/bin/env bash
# Creates predecessor/successor links for {feature} tickets.
#
# Usage:
#   JIRA_EMAIL="[email protected]" JIRA_API_TOKEN="token" bash scripts/jira-link-{feature}.sh
#
# Get an API token at: https://id.atlassian.com/manage-profile/security/api-tokens

set -euo pipefail

JIRA_HOST="https://commercetools.atlassian.net"
LINK_API="$JIRA_HOST/rest/api/3/issueLink"

if [[ -z "${JIRA_EMAIL:-}" || -z "${JIRA_API_TOKEN:-}" ]]; then
  echo "Error: Set JIRA_EMAIL and JIRA_API_TOKEN environment variables."
  exit 1
fi

AUTH=$(printf '%s:%s' "$JIRA_EMAIL" "$JIRA_API_TOKEN" | base64)
LINK_TYPE="${JIRA_LINK_TYPE:-dependency}"

OK_COUNT=0
FAIL_COUNT=0

create_link() {
  local predecessor="$1"
  local successor="$2"
  local reason="$3"

  local status
  status=$(curl -s -o /dev/null -w "%{http_code}" \
    -X POST "$LINK_API" \
    -H "Authorization: Basic $AUTH" \
    -H "Content-Type: application/json" \
    -d "{
      \"type\": { \"name\": \"$LINK_TYPE\" },
      \"inwardIssue\": { \"key\": \"$predecessor\" },
      \"outwardIssue\": { \"key\": \"$successor\" }
    }")

  if [[ "$status" == "201" ]]; then
    echo "  OK  $predecessor -> $successor ($reason)"
    OK_COUNT=$((OK_COUNT + 1))
  else
    echo "  FAIL[$status]  $predecessor -> $successor ($reason)"
    FAIL_COUNT=$((FAIL_COUNT + 1))
  fi
}

# --- Links ---
# create_link PREDECESSOR SUCCESSOR "reason"

# --- Summary ---
# echo ""
# echo "=== Summary ==="
# echo "  OK: $OK_COUNT"
# echo "  FAIL: $FAIL_COUNT"
# echo "  Total: $((OK_COUNT + FAIL_COUNT))"

Script Requirements

• The script MUST be executable (chmod +x)
• The script MUST print a summary count at the end
• The script MUST clearly label each section of links
• The script MUST NOT use ((var++)) for counters — with set -e, post-increment from 0 returns falsy and kills the script. Use var=$((var + 1)) instead.
• The script SHOULD include a --cleanup flag to remove links from a prior bad run (query each ticket's links, filter by type, DELETE each link ID)

Step 6: Update Plan Document

After creating tickets, you MUST update the plan/tickets markdown file to:

• Replace placeholder IDs with actual CRAFT-XXXX keys
• Include the dependency table with real keys
• Note which tickets are fully independent (no predecessors or successors)

Validation Checklist

Before declaring done, verify:

• User approved the ticket preview before any Jira creation
• All tickets created with correct parent epic
• All tickets have labels and component (if applicable)
• All tickets have acceptance criteria in description
• Dependency table uses real CRAFT keys
• Link script generated with correct direction (inwardIssue = predecessor)
• Link script uses dependency type (not Blocks)
• Link script is executable
• Plan document updated with real CRAFT keys
• Independent tickets clearly identified

Error Recovery

Wrong link direction created

If links are created backwards (predecessor shows "is successor of"):

1. Run the script with --cleanup flag to delete bad links
2. Verify the create_link function uses inwardIssue for predecessor
3. Re-run the script

Link type not found (404)

1. Check the link type discovery output at the top of the script
2. Set JIRA_LINK_TYPE env var to the correct type name
3. Common alternatives: Sequence, Gantt: finish-start, Gantt Dependency

MCP tool returns "Bad Request" for issue links

This is expected. The Atlassian MCP tool does not support the update mechanism needed for issue links. Use the generated shell script instead.

RFC 2119 Key Words

The key words "MUST", "MUST NOT", "SHOULD", "SHOULD NOT", and "MAY" in this document are to be interpreted as described in RFC 2119.

Execute ticket creation for: $ARGUMENTS