Getting started with DexDev

Turn how I want my AI to behave into a DexDev soul using the DexDev MCP tools.

1. Gather what you know about the persona I want: in a chat app, my custom instructions, personalization and saved memories about tone and style; in a local agent, the persona, tone and "how to work with me" parts of CLAUDE.md, AGENTS.md, GEMINI.md or Cursor rules. Show me a short summary of what you found.
2. Call search_souls first. If a soul already covers this, propose changes to it instead of creating a new one.
3. Draft SOUL.md with YAML frontmatter (name, a one-sentence description of when to use it, tags) and the sections "## Identity", "## Voice and tone", "## Values", "## Boundaries" and "## Working with me". Keep it concrete and short enough to load at the start of every session. Ask me anything you cannot infer.
4. After I approve, save it with create_soul (or update_soul with the expected_revision from get_soul). Never include secrets.
5. Tell me the slug, and that connected clients will load it at the start of a session.

Help me turn my reusable instructions into DexDev skills using the DexDev MCP tools.

1. Look at what you can see: project or custom GPT instructions, skills enabled in this chat, and tasks I have asked you to do the same way more than once. List the candidates (name, one line on when to use it) and let me pick.
2. For each one I pick, call search_skills first; if DexDev already has it, improve it with update_skill instead of creating a duplicate.
3. Write a SKILL.md with YAML frontmatter (name, a one-sentence description of when to use it, tags) and a body with "## When to use", "## Steps" (numbered, concrete) and "## Notes". Save it with create_skill. Never include secrets.
4. Finish with a table: skill, slug, created or updated.

Copy what you remember about me into DexDev using the DexDev MCP tools.

1. List everything you know about me that should carry over to other AI tools: saved memories, custom instructions, preferences, facts about my work and decisions we made. Show me the list before saving anything.
2. Split it into single ideas: one preference, fact, decision or note per memory, each with a short title, the right type and a few tags.
3. For each one, call search_memories with its key words first; if DexDev already has it, update that memory with update_memory instead of creating a duplicate. Otherwise call create_memory with source_kind "agent_explicit" and a client_request_id.
4. Never save secrets: skip passwords, tokens, keys and account numbers.
5. Finish with a table: title, type, created or updated, skipped (with reason).

Turn how I want my AI to behave into a DexDev soul using the DexDev MCP tools.

1. Gather what you know about the persona I want: in a chat app, my custom instructions, personalization and saved memories about tone and style; in a local agent, the persona, tone and "how to work with me" parts of CLAUDE.md, AGENTS.md, GEMINI.md or Cursor rules. Show me a short summary of what you found.
2. Call search_souls first. If a soul already covers this, propose changes to it instead of creating a new one.
3. Draft SOUL.md with YAML frontmatter (name, a one-sentence description of when to use it, tags) and the sections "## Identity", "## Voice and tone", "## Values", "## Boundaries" and "## Working with me". Keep it concrete and short enough to load at the start of every session. Ask me anything you cannot infer.
4. After I approve, save it with create_soul (or update_soul with the expected_revision from get_soul). Never include secrets.
5. Tell me the slug, and that connected clients will load it at the start of a session.

Upload all of my agent skills to DexDev using the DexDev MCP tools.

1. Find every skill you can access: skill folders that contain a SKILL.md (for example ~/.claude/skills, .claude/skills in this repository, ~/.codex/skills, plugin skill folders) and any other reusable instruction files you load as skills. List what you found (name, path, file count) before uploading anything.
2. For each skill, call search_skills with its name first. If DexDev already has the same skill, skip it and tell me.
3. Upload each new skill with create_skill: pass the complete SKILL.md (keep or add YAML frontmatter with name and a one-sentence description of when to use it) and every supporting file in its folder (scripts, references, assets) as files with folder-relative paths. Use base64 for binary files.
4. Never upload secrets: skip .env files, keys, tokens and credentials, and replace secrets inside files with placeholders such as $API_KEY.
5. When you are done, give me a table: skill, slug, files uploaded, skipped (with reason).

Upload the memories my local agents keep to DexDev using the DexDev MCP tools.

1. Find every memory store you can read: Claude Code auto-memory (~/.claude/projects/*/memory/*.md and MEMORY.md), "remember" or memory sections in CLAUDE.md / AGENTS.md / GEMINI.md, Codex and Cursor notes, and any notes file you load at start. List what you found (path, number of entries) before uploading anything.
2. Split them into single ideas: one preference, fact, decision or note per memory, with a short title, the right type and a few tags. Drop anything stale or machine-specific (absolute paths, temp files).
3. For each one, call search_memories with its key words first; if DexDev already has it, update that memory with update_memory instead of creating a duplicate. Otherwise call create_memory with source_kind "agent_explicit" and a client_request_id. Memories that belong to one project get project_ids with that project's slug.
4. Never upload secrets: skip tokens, keys, passwords and account numbers, and replace them inside content with placeholders such as $API_KEY.
5. Finish with a table: title, type, created or updated, skipped (with reason).

Help me turn my DexDev skills into one clean, de-duplicated library using the DexDev MCP tools.

1. Call search_skills with all_projects=true and no query, following next_cursor until you have every skill. Read each one with get_skill.
2. Group skills that cover the same job (duplicates, near-duplicates, older and newer variants, the same skill uploaded from different agents).
3. Show me a plan before changing anything: for each group, which skill to keep, what to merge into it, the final slug and name, and which skills to archive. Also flag skills with vague descriptions or missing steps.
4. After I approve, apply the plan: update the kept skill with update_skill (merged SKILL.md, combined files, clear description), then archive the merged-away skills with archive_skill. Use the expected_revision from get_skill and re-read on conflicts.
5. Finish with a summary of kept, merged, renamed and archived skills. Never delete anything; archived skills can be restored on the DexDev website.

Bring this repository's agent context into the DexDev project "<project-slug>" using the DexDev MCP tools.

1. Call list_projects and confirm a project with slug "<project-slug>" exists. If it does not, stop and tell me to create it at https://dex-dev.com/projects/new first.
2. Collect what this repository teaches an agent: CLAUDE.md, AGENTS.md, GEMINI.md, .cursor/rules, docs that record conventions or decisions, and the skills in .claude/skills or .codex/skills. List what you found before uploading.
3. Upload each project skill with create_skill and project_ids ["<project-slug>"] (call search_skills with project_id "<project-slug>" first to avoid duplicates), including supporting files with folder-relative paths.
4. Turn conventions and decisions into memories: one idea each, create_memory with project_ids ["<project-slug>"], source_kind "agent_explicit" and a client_request_id. Call search_memories with project_id "<project-slug>" first and use update_memory when one already covers it.
5. Never upload secrets (.env files, keys, tokens); replace any inside files with placeholders such as $API_KEY.
6. Finish with a table: item, kind (skill or memory), slug or title, status.