Claude · Where each workflow layer lives
This Step 2 mapping tells you where to place each workflow layer inside Claude: project-level instructions, project knowledge (uploaded reference material), reusable Skills, runtime chat input, and verification gates. Start from the layer (Step 1), then place it in the correct Claude surface below.
Choose your Claude setup (fast decision)
- Project: best for ongoing workflows with a dedicated knowledge base + project chat history + project instructions.
- Skill: best for repeatable workflows packaged as a reusable unit (instructions + scripts + resources), loaded dynamically.
- Regular chat: best for a one-off request outside a project (no project knowledge base).
1) Claude surfaces you can configure (official)
- Projects (workspace)
- Projects are self-contained workspaces with their own chat histories and knowledge bases. Inside a project, you can upload documents/text/code/files to the project knowledge base and define project instructions.
- Project Knowledge (uploaded reference material)
- The project’s knowledge base: upload documents/text/code/files Claude should consult as context for chats in that project. When project knowledge approaches context limits, Claude can use Retrieval-Augmented Generation (RAG) to retrieve relevant info via a project knowledge search tool and expand capacity by up to 10x. Note: Claude Help Center pages currently differ on whether “enhanced project knowledge with RAG” is paid-only vs “RAG for projects” available for all plans—treat availability as plan/account-dependent and rely on the in-product RAG indicator.
- Project Instructions (project-scoped rules)
- A project-level instruction surface: use it for workflow rules that should apply across all chats in that project.
- Skills (reusable workflows)
- Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. Skills require code execution and file creation to be enabled, and are managed under Customize → Skills (with org-level controls on Team/Enterprise). Claude can automatically use built-in Skills when relevant once enabled.
- Chat search (paid plans) + Memory (cross-chat)
- On paid plans, Claude can search past chats using RAG and show tool calls; search boundaries differ for project vs non-project chats. Memory from chat history is available for all users; Claude can generate a memory summary for standalone chats and a separate memory space + dedicated project summary per project. You can toggle search and memory in Settings → Capabilities. Incognito chats (ghost icon) are not saved to chat history and are excluded from future memory summaries and searches.
- Web search (when enabled)
- Web search can be toggled on in chat (UI toggle in the input interface). When enabled, Claude includes citations in responses. On Team/Enterprise, web search must be enabled by an Owner at the workspace level before members can use it.
- Artifacts (output surface)
- Artifacts are substantial standalone outputs in a dedicated window (apps/tools/content) that you can modify and reuse. Treat artifacts as an output container; they do not replace instructions or knowledge surfaces.
- Sharing chats (governance)
- Chats are private by default; you can create shareable snapshots via the Share button. Shared snapshots include messages and artifacts up to the share point, but attached files are not included, and MCP tool-call raw data is hidden.
2) Layer → where to put it in Claude (Step 2 mapping)
The goal is correct placement: rules go into instruction surfaces, reference material goes into project knowledge, repeatable procedures become Skills (or a Project workflow), current requests stay in the current chat, and verification stays as a gate/checklist before you use output.
- Instruction layer
-
Put stable workflow rules here (rules that must persist across messages).
- Project Instructions (Project) — project-scoped rules for workflows you run inside a Project.
- Skill instructions (Skill) — packaged workflow rules that Claude loads for repeatable tasks.
- Memory (not an instruction surface) — use Memory only for personal/work preferences; do not store workflow rules as memory if they must be deterministic.
- Reference / source material layer
-
Put grounding material here (policies, docs, examples, evidence the model should consult).
- Project Knowledge — upload the documents/text/code/files Claude should use as context for project chats.
- RAG for projects (automatic when needed) — Claude uses a project knowledge search tool to retrieve relevant information from uploaded documents and can expand capacity up to 10x; you’ll see an indicator when RAG is active.
- Reusable workflow layer
-
Put the repeatable procedure here (so you don’t rewrite it every time).
- Skill — best for a packaged repeatable workflow (instructions + scripts + resources).
- Project — best when the workflow depends on shared project knowledge + project instructions + project chat history.
Enabling Skills: Settings → Capabilities (enable code execution + file creation if needed), then Customize → Skills, then toggle Skills on/off. On Enterprise, org Owners enable Skills under Organization settings → Skills.
- Runtime prompt layer
-
Put the one-off task input for the current run here.
- Current chat message inside the relevant Project (or a standalone chat if you are not using Projects).
- Use chat search only when needed (paid plans): ask Claude to find prior context; you can toggle “Search and reference chats” in Settings → Capabilities.
- Incognito chats: use the ghost icon to start a temporary chat that won’t be saved to history and won’t be used for memory summaries or past-chat search.
- Verification layer
-
Verification is a required step after Claude produces an answer.
In Claude, you run verification as a deliberate second pass before you use the output.
- Default (works everywhere): send a second message: “Verification pass: check the answer against the guide checklist and project knowledge. Return PASS/FAIL + issues + required fixes.”
- Enforced (recommended): put the verification requirement in Project Instructions or Skill instructions so every run ends with a verification summary.
- Web-grounded workflows: enable web search in chat and require citations; verify by opening cited sources.
Plan note: Projects exist on free accounts (with a free-tier limit), chat search is paid-plan only, Skills require code execution + file creation, and collaboration/sharing permissions vary by Team/Enterprise. Use in-product availability as the final gate.