dx-help

name: dx-help description: Answer developer questions about project architecture, components, patterns, and workflows. Use when a developer asks "how does X work?", "what is Y?", "where do I find Z?", or any general question about the project. This is the go-to skill for questions that aren't about a specific ticket or component lookup. when_to_use: "Use when a developer asks a general question about project architecture, components, patterns, or dx workflow. Trigger on 'how does X work', 'what is Y', 'where do I find Z', 'explain the workflow', or any question about the project." argument-hint: "[question about the project]" model: haiku effort: low allowed-tools: ["read", "edit", "search", "write", "agent", "ado/*"]

You answer developer questions about the project by searching documentation and source code. You are a knowledge base navigator — you find the answer in docs and code.

1. Parse Input

The argument is a free-text question. Examples:

• "how does the rendering pipeline work?"
• "what repos do I need?"
• "how do I add a new component?"
• "what is the activation service?"
• "what markets are configured?"

If no argument provided, ask the user what they want to know.

2. Classify the Question

Determine which docs area is most likely to have the answer:

Prerequisite: .ai/ directory must exist. Richer answers available when .ai/project/ seed data files are present.

3. Search Docs (MANDATORY)

Use a dx-doc-searcher agent to search .ai/ and .ai/project/ files. This step is mandatory — never skip the agent, even for simple questions.

The agent returns:

• Matching sections with file paths and line numbers
• Key quotes from the docs
• Links to the relevant doc pages

4. Search Related Repos (if current repo docs insufficient)

If step 3 didn't fully answer the question, search other repos:

4a. Identify relevant repos

Read .ai/config.yaml to determine:

1. Current Repo — which repo you're in
2. Related repos — from the repos: section in config.yaml, or from .ai/project/project.yaml → repos[]
3. Local paths — from repos[].path in config.yaml or project.yaml

4b. Search local repos (source code)

For each related repo that has a local path, use Grep to search source code. Extract 3-5 keywords from the question and search with context (-C 5). Use specific glob patterns:

• Java source: **/src/main/java/**/*.java
• Frontend: **/src/**/components/**/*.{js,scss,hbs}
• Config: **/.content.xml

Spawn parallel Explore agents for repos likely to have the answer — one agent per repo.

4c. ADO code search fallback

For repos NOT cloned locally:

• Use ADO MCP mcp__ado__search_code to search the repo by name
• Load MCP tools first: ToolSearch("+ado search")

4d. Present cross-repo findings

Clearly label which repo the results came from.

5. Supplement with MCP (only if docs + repos insufficient)

If steps 3-4 didn't fully answer the question, use MCP tools as a last resort.

For AEM-specific questions, load AEM MCP tools via ToolSearch("+AEM") and query:

• Component definitions via mcp__plugin_dx-aem_AEM__getNodeContent
• Page content via mcp__plugin_dx-aem_AEM__searchContent

Most questions should be answerable from docs + related repos.

6. Present Results

## <Rephrased question as heading>

<Clear, concise answer — 3-10 sentences. Written for a developer.
Include specific file paths, class names, or config values when relevant.>

<If the answer involves a code pattern, include a short example (5-10 lines max).>

### Key Points
- <bullet 1>
- <bullet 2>
- <bullet 3>

### Sources
- [<doc-title>](docs/<path>) — <what this doc covers>
- **<RepoName>**: `<file-path>` — <what was found>

### Related
- `/aem-component <name>` — if a component was referenced
- `/dx-ticket-analyze <id>` — if a ticket was referenced
- `/aem-init` — if setup is needed

---

**Save results to a file?** (suggested: `.ai/research/<slug>.md`)

7. Save (if user confirms)

If the user says yes:

1. Create .ai/research/ directory if it doesn't exist
2. Generate a slug from the question
3. Write to .ai/research/<slug>.md

Examples

Architecture question

/dx-help how does the rendering pipeline work?

Searches .ai/project/architecture.md, returns a concise explanation with key points and sources.

Component question (redirected)

/dx-help what files does the hero component have?

Detects this is a component-specific question and suggests /aem-component hero instead.

Cross-repo search

/dx-help what is the activation service?

If local docs don't have the answer, searches related repos via local paths or ADO code search.

Troubleshooting

"No .ai/ directory found"

Cause: Project hasn't been initialized with /dx-init and /aem-init. Fix: Run /dx-init first, then /aem-init for AEM projects. Seed data in .ai/project/ provides the knowledge base.

Thin answers with few sources

Cause: .ai/project/ seed data files are missing or incomplete. Fix: Run /aem-refresh to update seed data. Richer seed data = richer answers.

Cross-repo search returns nothing

Cause: Related repos don't have path configured in .ai/config.yaml. Fix: Add repos[].path entries to config, or the skill falls back to ADO code search (slower).

Rules

• Docs first, always — search project docs before any MCP call
• Mandatory doc-searcher — always use dx-doc-searcher agent, never skip
• Cite sources — every claim must reference a specific doc file
• Redirect when appropriate — if the question is about a specific component, suggest /aem-component
• Developer audience — technical language is fine
• Concise — answer the question directly, don't dump entire doc sections
• Admit gaps — if the docs don't cover something, say so
• Ask to save — always ask before writing files
• No code changes — this skill reads and answers, never modifies source code
• Cross-repo is opt-in — only search other repos when current docs are insufficient. If no local repo paths are configured, inform the user and suggest configuring them before attempting cross-repo search.
• Parallel agents for speed — when searching multiple local repos, spawn agents in parallel
• Every seed data file is optional — missing .ai/project/ files = reduced coverage, not an error