licanfu/multica-best-practices/multica-platform-confluence

Claude Code and agent skill by licanfu/multica-best-practices

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Platform-layer skill (placeholder shell): read/write capability for Confluence / Wiki-class systems — pull existing pages for Agents to consume, land PRD / design artifacts and return a stable page link. Decoupled from the Issue system; called by artifact-orchestration skills. Concrete platform URLs / spaces / parent pages live in config.yaml, never in role prompts.

Install

npx vskill@latest install licanfu/multica-best-practices/multica-platform-confluence

SKILL.md

Platform · Confluence (placeholder shell)

This is a platform-layer placeholder shell. The public multica-best-practices binds to no specific company's internal URLs or credentials.
To onboard your own Wiki, only edit this skill's config.yaml and scripts/; all upstream roles and orchestration skills stay untouched.

Purpose

Provide read + write capability for "knowledge-base / Wiki" class platforms:

  • Read: fetch an existing page by pageId / title as Markdown for downstream Agents.
  • Write: land PRD / technical design artifacts to the team Wiki and return a stable page link.

Decoupled from multica-platform-jira (Issue system); this skill only handles Wiki, not Issue field writes. Role prompts never contain Wiki URLs / pageIds; swap platforms by swapping only this skill.

Default landing (team-configured)

Artifact typeWiki parent page (config.yaml)Local draft (Agent writes first)
PRDconfig.yaml → wiki.default_parent_page_idstructured by multica-pm-requirement-spec, landed by req-sync
Technical designconfig.yaml → wiki.design_parent_page_iddocs/design/<ISSUE-KEY>/design.md

Files (suggested structure)

text
multica-platform-confluence/
├── SKILL.md
├── config.yaml
├── .env.example
└── scripts/
├── credentials.sh
├── wiki.sh # Read/Write CLI
├── fetch_page.py # Wiki → local Markdown
├── publish_design.py # Markdown design upsert
└── lib/

Read (downstream / Leader pulls upstream artifact)

bash
bash scripts/wiki.sh fetch-page <page_id> [output_dir] [issue_key]
bash scripts/wiki.sh get-page <page_id>
bash scripts/wiki.sh find-page "<title>" [space_key]

Write (artifact landing)

PRD page
bash
bash scripts/wiki.sh create-page "<title>" "<parent_page_id>" "<html_or_md>" "<space_key>"

Called by multica-pm-artifact-publish.

Technical design (Markdown → Wiki)
  1. @Architect writes local via multica-technical-design: docs/design/<ISSUE-KEY>/design.md.
  2. Publish under the design parent page:
bash
python scripts/publish_design.py <ISSUE-KEY> docs/design/<ISSUE-KEY>/design.md \
[--space SPACE] [--parent PAGE_ID] [--title "Title"] [--json]
  1. Return url / page_id from the JSON; multica-artifact-architect then calls the Issue-platform skill to write the link back to the Issue.

Upsert rule: same space + same title → update version; title may get an [AI] suffix (team-configurable).

Agent Compatibility

  • Credentials injected via runtime env vars (.env.example gives the variable-name template); never print passwords, never write into role prompts.
  • Confirm before external writes: space, parent pageId, title.
  • Prefer scripts/; do not call REST directly.

Adapting To A New Team

  1. Edit config.yaml: wiki.url, default_space, *_parent_page_id.
  2. Change the design-doc parent page ID to the team Wiki directory page.
  3. Maintain space / directory lists in spaces.json (if applicable).

Why it works

Wiki auth, spaces, and parent pages differ per team; isolating them in a platform skill means Issue-system / notification / Git-change scripts don't affect Wiki scripts, and PRD + design share one read/write capability. A swappable platform layer is the core of "copy-paste-run": role prompts only ever say multica-artifact-*, and the real internal details stay in this layer.

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/licanfu/multica-best-practices/multica-platform-confluence/badge)](https://verified-skill.com/skills/licanfu/multica-best-practices/multica-platform-confluence)

Something wrong with this skill or its page? Tell us.