atlassian-rest（阿特拉斯休息）

Atlassian REST API Skill

Portable Jira & Confluence integration via REST APIs. Works in any agent environment with Node.js 18+ — zero dependencies, no MCP server required.

First-Use Setup

<skill-path> throughout this document refers to the directory where this skill is installed. Resolve it from the skill invocation context (e.g., the path shown by claude plugin list or the skill's directory in .claude/plugins/).

Before any operation, verify the user has credentials configured. Run:

node <skill-path>/scripts/setup.mjs

If it fails, guide the user through the setup — the script prints step-by-step instructions.

Required Environment Variables

How to Use This Skill

When the user asks you to do something with Jira or Confluence, follow these principles:

1. Resolve ambiguity first. If the user says "create a ticket" but hasn't specified a project, run node <skill-path>/scripts/jira.mjs projects to list available projects, then ask which one. Same for issue types — run node <skill-path>/scripts/jira.mjs issue-types <projectKey> if unsure.
2. Confirm before mutating. Before creating issues, transitioning tickets, or publishing Confluence pages, show the user what you're about to do and get confirmation. Read operations (search, get, list) don't need confirmation.
3. Never delete. This skill does not support delete operations (issues, pages, attachments, boards, projects, accounts, etc.). If the user asks to delete something, direct them to the Atlassian web UI. This restriction is intentional and must not be bypassed.
4. Compose operations naturally. Many user requests require multiple script calls. For example, "assign PROJ-123 to Sarah" requires: (a) lookup-user "Sarah" to get the account ID, then (b) edit PROJ-123 --assignee <accountId>.
5. Prefer sync.mjs for document-based operations. When creating Jira issues from a local markdown file (story docs, specs, epics), use sync.mjs instead of raw jira.mjs create — it handles field mapping, link tracking, and sync state automatically. See the Document Sync Operations section below for details. Only use jira.mjs create for ad-hoc issues not backed by a local document.
6. Use workflows for complex tasks. If the user's request matches one of the workflows below, read the corresponding file and follow its step-by-step process.
7. Read reference docs when needed. Before writing JQL/CQL queries, consult references/query-languages.md. Before creating tickets, consult references/ticket-writing-guide.md. The reference docs exist to help you produce high-quality output — use them.

Passing Long Content to Scripts

For descriptions, comments, or page bodies longer than ~200 characters or containing special characters (backticks, quotes, $, newlines), write the content to a temp file and use the file-based flag:

Write plain markdown to the file — scripts handle conversion to ADF (Jira) or storage format (Confluence) automatically. Prefer file-based input to avoid shell escaping issues.

Jira Operations

Script: node <skill-path>/scripts/jira.mjs <command> [args]

Search Issues

jira.mjs search 'project = PROJ AND status = "In Progress"' --max 20

Get Issue Details

jira.mjs get PROJ-123
jira.mjs get PROJ-123 --fields summary,status,assignee

Create Issue

jira.mjs create --project PROJ --type Task --summary "Implement feature X" \
  --description "Details here" --priority High --assignee <accountId> \
  --labels "backend,urgent" --components "API,Auth"
jira.mjs create --project PROJ --type Story --summary "User login" --parent PROJ-100
# For long descriptions, use a file:
jira.mjs create --project PROJ --type Task --summary "Feature X" \
  --description-file /tmp/desc.md --priority High

When creating child stories under an Epic, include --priority Medium unless the user specifies a different priority.

Edit Issue

jira.mjs edit PROJ-123 --summary "Updated title" --priority Medium
jira.mjs edit PROJ-123 --labels "backend,v2" --components "API"
# For long descriptions, use a file:
jira.mjs edit PROJ-123 --description-file /tmp/desc.md

Comments

jira.mjs comment PROJ-123 "Fixed in PR #456"
# For long comments, use a file:
jira.mjs comment PROJ-123 --body-file /tmp/comment.md

Transitions (move ticket status)

jira.mjs transitions PROJ-123          # List available transitions first
jira.mjs transition PROJ-123 31        # Then transition by ID

Always list transitions first to get the correct ID — don't guess.

Projects & Issue Types

jira.mjs projects                      # List all visible projects
jira.mjs issue-types PROJ              # List issue types for a project

Issue Links

jira.mjs link-types                    # List available link types first
jira.mjs link PROJ-1 PROJ-2 --type "relates to"

User Lookup

jira.mjs lookup-user "john"            # Returns account ID needed for --assignee

Worklog

jira.mjs worklog PROJ-123 --time 2h --comment "Code review"

Document Sync Operations

Script: node <skill-path>/scripts/sync.mjs <command> [args]

When creating Jira/Confluence items from local markdown documents, prefer sync.mjs over raw jira.mjs/confluence.mjs — it auto-updates the source document with links and maintains sync state.

Setup Field Mapping (first time per doc type)

sync.mjs setup-mapping --type story --sample PROJ-200   # Auto-detect fields from existing ticket
sync.mjs setup-mapping --type epic --sample PROJ-100    # Creates memory/jira-epic-field-mapping.json

Field mappings are stored in <skill-path>/memory/ and define how markdown sections map to Jira fields. See references/sync-mapping-guide.md for the full schema.

Link & Create from Document

sync.mjs link <file> --type story --project PROJ --create    # Create Jira issue + update doc
sync.mjs link <file> --type epic --project PROJ --create     # Create epic + child stories
sync.mjs link <file> --type story --ticket PROJ-123          # Link to existing ticket

Push/Pull Changes

sync.mjs push <file>                    # Push local changes to Jira/Confluence
sync.mjs push <file> --delete-orphans   # Push + prompt to delete orphaned Sub-* subtasks
sync.mjs pull <file>                    # Pull remote changes to local
sync.mjs diff <file>                    # Show per-section diff
sync.mjs status <file>                  # Show sync status

When push reports orphaned subtasks (sections removed from local doc), ask the user if they want to delete them, then run with --delete-orphans. Only Sub-* issue types can be deleted — parent issues are skipped.

Custom Instructions in Mapping Config

The field mapping JSON (memory/jira-<docType>-field-mapping.json) supports an instructions field for additional agent guidance:

{
  "instructions": "Always set priority to High. Add label 'team-alpha'. Use Sub-Imp type for child items."
}

When present, instructions are printed to stdout during push and link operations so the calling agent can follow them.

Batch Operations

sync.mjs batch                # Scan all linked docs and report status

Confluence Operations

Script: node <skill-path>/scripts/confluence.mjs <command> [args]

Search Pages

confluence.mjs search 'type = page AND text ~ "architecture"' --max 10

Get Page

confluence.mjs get-page 12345
confluence.mjs get-page 12345 --format view

Create Page

confluence.mjs create-page --space TEAM --title "Sprint Report" --body "Report content"
confluence.mjs create-page --space TEAM --title "Sub Page" \
  --body "<h2>Heading</h2><p>Content</p>" --parent 12345
confluence.mjs create-page --space TEAM --title "Full Doc" --body-file /tmp/body.md

The --body flag accepts markdown (recommended), plain text, or raw HTML storage format (if it starts with <). The script automatically converts markdown to Confluence storage format — headings, lists, tables, and code blocks (converted to ac:structured-macro ac:name="code" with language detection) are all handled. Prefer writing markdown and letting the script handle conversion rather than manually constructing storage format XHTML. Use --body-file for long documents that would exceed shell argument limits.

Update Page

confluence.mjs update-page 12345 --title "Updated Title" --body "New content"
confluence.mjs update-page 12345 --title "Updated Title" --body-file /tmp/body.md

Version is auto-incremented — no need to track it manually. Use --body-file for large page updates.

Comments

confluence.mjs comment 12345 "Reviewed and approved"
# For long comments, use a file:
confluence.mjs comment 12345 --body-file /tmp/comment.md

Attachments

confluence.mjs attach 12345 ./screenshot.png --comment "Architecture diagram"
confluence.mjs list-attachments 12345 --max 10

Use attach to upload local files (images, PDFs, etc.) to a page. After uploading, embed images in the page body using <ac:image><ri:attachment ri:filename="screenshot.png" /></ac:image> — see references/confluence-formatting.md for sizing guidelines.

Sync Confluence Space to Local Markdown

# Download an entire page tree to local markdown with attachments
node <skill-path>/scripts/sync-confluence-space.mjs --root <pageId> --output ./docs

# Preview without writing files
node <skill-path>/scripts/sync-confluence-space.mjs --root <pageId> --output ./docs --dry-run

# Skip attachment downloads (faster, markdown only)
node <skill-path>/scripts/sync-confluence-space.mjs --root <pageId> --output ./docs --skip-attachments

Downloads an entire Confluence page tree to local markdown files with hierarchy, images, and linked titles. Rewrites both image references and file attachment links to point to local assets. See workflows/sync-confluence-space.md for full details and customization options.

Spaces & Navigation

confluence.mjs spaces --max 20
confluence.mjs descendants 12345       # Get child pages

Format Conversion Utilities

Script: <skill-path>/scripts/confluence-format.mjs

This module provides bidirectional markdown ↔ Confluence storage format conversion. It's used internally by confluence.mjs but can also be imported directly for custom sync scripts.

Exported Functions

import { markdownToStorage, storageToMarkdown, htmlInlineToMarkdown } from '<skill-path>/scripts/confluence-format.mjs';

What storageToMarkdown Handles

The converter handles real-world Confluence storage format patterns including:

• Structured macros: code blocks (with/without language), panels (info/tip/warning/note → GitHub alerts), expand → <details><summary> collapsible sections, jira references, view-file → attachment links, toc/children (stripped)
• Unknown macros: Any unrecognized ac:structured-macro types have their body content preserved (not silently dropped)
• Tables: <colgroup>, <tbody>, <thead> wrappers stripped; attributes on <table>, <tr>, <th>, <td> handled
• Task lists: <ac:task-list> with <ac:task-id> → markdown checkboxes
• Images: URL and attachment images → ![alt](url) with filename as fallback alt text; parentheses and spaces in filenames URL-encoded
• Code blocks: Protected from HTML tag stripping via placeholder system — generic types like Array<string> preserved inside fenced blocks
• Expand/collapsible bodies: Full content support inside expand macros — code blocks, images, tables, panels, task lists, and nested macros all convert correctly
• Entities: Full HTML entity decoding (’, “, →, `, numeric {`)
• Block separation: \n\n around headings, images, lists, <hr>, tables, code blocks
• Attribute tolerance: All HTML tag regexes accept optional attributes (handles Confluence's local-id, data-layout, breakoutWidth, etc.)
• Bold/italic cleanup: Trailing spaces inside markers fixed (**text ** → **text**)
• Stray angle brackets: Escaped outside code blocks to prevent markdown renderer confusion

Syncing a Confluence Space

For most use cases, use the built-in sync script directly:

node <skill-path>/scripts/sync-confluence-space.mjs --root <pageId> --output ./docs

For custom sync scripts, import storageToMarkdown and follow this pattern:

import { storageToMarkdown } from '<skill-path>/scripts/confluence-format.mjs';

// Fetch page via Confluence v2 API
const page = await apiGet(`/wiki/api/v2/pages/${pageId}`, { 'body-format': 'storage' });
const html = page.body.storage.value;

// Convert to markdown
let markdown = storageToMarkdown(html);

// Add linked title using page._links.base + page._links.webui
const pageUrl = `${page._links.base}${page._links.webui}`;
markdown = `# [${page.title}](${pageUrl})\n\n${markdown}`;

Key gotchas discovered in production use:

• The v2 API _links.webui is a relative path (e.g., /spaces/BTH/pages/123/Title). Combine with _links.base (e.g., https://company.atlassian.net/wiki) for the full URL — do NOT use env.domain + _links.webui directly (missing /wiki prefix).
• Attachment filenames with spaces and parentheses must be URL-encoded in markdown links: ![alt](Screenshot%202025-11-05%20at%2016.50.02.png).
• When writing a hierarchical sync, use getChildrenTree() (recursive children API) to build the tree, then decide per-node: pages with children → directory/index.md, leaf pages → flat PageName.md.
• Keep attachments in a shared assets/imgs/ and assets/pdfs/ directory. Use depth-aware relative paths (../assets/imgs/ at depth 1, ../../assets/imgs/ at depth 2, etc.).
• The sync script must rewrite both image references (![alt](filename)) and file attachment links ([filename](filename)) to local paths — Confluence view-file macros produce regular links, not image references.
• Confluence adds attributes (local-id, breakoutWidth, data-layout) to most HTML tags. Any regex matching tags must use [^>]* for optional attributes.
• Expand macro bodies can contain any content (code blocks, images, tables). The inner HTML converter must handle the same set of elements as the main converter.

Running Tests

node <skill-path>/scripts/test-format.mjs

100+ tests covering forward conversion, reverse conversion, round-trip preservation, block element spacing, code block protection, image URL encoding, entity decoding, expand macros with nested content, view-file macros, unknown macro catch-all, and attribute tolerance on HTML tags.

Workflows

For complex multi-step operations that require user interaction across several turns, read the corresponding workflow file and follow its step-by-step process. For simple one-shot commands, use the operations sections above directly.

Error Handling

Reference Documentation

Load these as needed — don't read them all upfront: