Claude Project: Build a Persistent PM Assistant That Knows Your Project

What This Builds

A persistent Claude workspace that holds your project's context — charter, risks, stakeholder map, past status reports — so you never re-explain the project from scratch. You open the Project, ask "draft this week's status report" or "what risks should I flag at tomorrow's steering committee," and Claude already knows the engagement, the client, the politics you've written in, and your preferred output style. It's the difference between briefing a new contractor every week and working with a senior analyst who's been on the project since day one.

Prerequisites

• Claude Pro subscription ($20/month at claude.ai) — Projects require Pro
• Key project documents ready to upload: project charter or SOW, current risk register, stakeholder list, at least 2 past status reports
• 30–45 minutes to write your Project Instructions (the most important step)

The Concept

Think of a Claude Project like hiring a very capable research assistant and giving them a dedicated filing cabinet for your project. Every time you walk into the room, they've already read the files and are ready to help — you don't need to hand them the charter again. The "Project Instructions" you write are their briefing document: who the client is, what the project is trying to achieve, how you like to communicate, what they should never do (like invent budget figures). The uploaded files are the filing cabinet. Every conversation you start inside the Project picks up from that baseline.

Build It Step by Step

Part 1: Create the Project and Write Instructions

1. Log in to claude.ai. In the left sidebar, click Projects → New Project.
2. Name it clearly: [Client/Project Name] — PM Workspace. Example: Meridian Insurance — ERP Implementation PM.
3. Click Edit Project Instructions (the gear icon or "Instructions" tab within the project).
4. Write your instructions using the template below. Fill in every bracketed section — the more specific you are, the better Claude performs:

## Project Context
You are a senior project management analyst supporting me, [YOUR NAME], as PM on the following engagement:

**Project:** [Project name and one-sentence description]
**Client:** [Client name, industry, approximate size]
**My firm:** [Your firm name]
**Phase:** [Current phase — e.g., "Phase 2: Build and Test (of 4 phases)"]
**Timeline:** [Start date → End date, current status vs. plan]
**Budget:** [Total budget; note if tracking is sensitive]
**Key stakeholders:**
- [Name, Title, Organization] — [their priority/concern in one phrase]
- [Name, Title, Organization] — [their priority/concern]
- [Name, Title, Organization] — [their priority/concern]

## My Communication Style
- I write status reports in RAG (Red/Amber/Green) format for a non-technical executive audience
- I prefer bullet-heavy formats over dense prose
- I use "we" (not "I") in client-facing writing
- Tone: confident, direct, solutions-oriented — never defensive or vague

## What You Help Me With
- Drafting status reports, executive summaries, and steering committee materials
- Identifying risks I may be overlooking based on project context
- Drafting difficult stakeholder emails
- Reviewing my writing for clarity and tone
- Preparing for meetings — agenda design, anticipated questions, talking points

## Critical Rules
- Never invent specific numbers (budget figures, percentages, dates) unless I provide them
- If I ask for a status report and haven't given you this week's data, ask me for it before drafting
- Always flag if something I'm describing sounds like a scope creep scenario
- If I ask about a risk, check the risk register I've uploaded before giving generic advice
- Do not include client-identifying information in any response I could accidentally share externally — use "[Client]" as a placeholder in draft documents unless I tell you it's safe to use their name

1. Click Save Instructions.

Part 2: Upload Your Project Documents

1. Inside the project, click Add Content or the upload icon.
2. Upload the following documents (PDF, Word, or plain text all work):

   • Project charter or SOW (this is the most important file — it establishes scope)
   • Current risk register (XLSX exported as CSV or PDF works well)
   • Current stakeholder map or RACI matrix
   • Your last 2–3 weekly status reports (so Claude learns your format and voice)
   • Any key decisions log or steering committee minutes
3. After uploading, start a new conversation within the project and test: "Summarize the current state of this project in 3 bullets." Claude should reference your documents rather than making things up. If it hallucinates details, re-upload the charter as a cleaner file (avoid scanned PDFs — text-based PDFs work best).

Part 3: Run Your First Real Workflow

1. Start a conversation inside the project with this prompt:

I need to draft this week's status report. Here is my raw data from this week:

- Completed: [list what got done]
- In progress: [list active work]
- Blocked: [list blockers]
- Upcoming milestones: [list with dates]
- New risks or issues: [describe any]
- Budget: [current spend vs. plan]

Using the project context you have and my usual status report format, draft this week's report. Flag anything that looks like it should be Amber or Red based on what we've discussed before.

1. Review the output. On your first run, you'll likely want to adjust 2–3 things. Go back to Project Instructions and add a note: "Based on first run: [what to change]." This refinement takes one session and then the output quality improves significantly.

Real Example: Digital Transformation Engagement

Setup: Claude Project named "Hartwell Financial — Digital Claims Platform", instructions include 4 stakeholder entries, SOW uploaded, 4 past status reports uploaded.

Input (one Monday morning):

Draft this week's status report. Raw data:
- Completed: API integration spec signed off by IT, Sprint 6 demo went well — client happy
- In progress: Data mapping for legacy claims fields (30% done, originally scoped at 60% by now)
- Blocked: Client hasn't provided access to production data environment — 2-week-old blocker
- Upcoming: Sprint 7 planning Thursday, steering committee next Tuesday
- New risk: Their IT team flagged a potential conflict between our integration approach and their SOX audit requirements — TBD
- Budget: $847K spent of $1.2M (70.6% of budget, 58% through timeline)

Output (Claude, using project context):

Overall Status: Amber — Data mapping is running 30 points behind plan and the production environment access blocker is now in its third week. Budget pacing is ahead of schedule.

This Week's Accomplishments:
- API integration specification received client IT sign-off — removes a key dependency from Sprint 8
- Sprint 6 demo completed with positive client feedback; all acceptance criteria met
- Budget tracking at $847K (70.6%); timeline 58% complete — currently pacing ahead on spend

Upcoming Milestones:
- Sprint 7 planning: [Thursday date]
- Steering committee review: [Tuesday date]
- Data mapping Phase 1 completion (revised): targeting [date — needs client confirmation]

Risks and Issues:
- Production environment access (Issue — Escalation needed): Blocker is now 14 days old. Impact: data validation for Sprint 8 cannot begin until resolved. Recommended action: escalate to [Client IT Director name from stakeholder list] at steering committee.
- SOX audit conflict (Risk — Amber): Hartwell IT has flagged a potential conflict between our integration approach and upcoming SOX audit requirements. We are scheduling a technical call this week to assess. Impact TBD.
- Data mapping pace (Risk — Amber): Currently 30% complete vs. 60% target. Root cause: source system field documentation is incomplete. Mitigation: PM and data lead to agree on revised scope by Thursday planning.

Decisions Needed:
- Escalation of production access blocker — request client leadership intervention at Tuesday steering committee

PM review time: 6 minutes — added specific names, confirmed dates, sent.

What Claude used from the project files: The stakeholder name and title came from the uploaded stakeholder map. The tone and section structure matched the past status reports. The SOX risk was accurately flagged as "TBD" rather than invented.

What to Do When It Breaks

• Claude ignores the uploaded files and gives generic answers → The files may be too large or poorly formatted. Split a large SOW into sections and re-upload. Scanned PDFs (image-based) aren't readable — export as text-based PDF or copy-paste into a .txt file.
• Claude invents numbers or facts not in your documents → Strengthen the "Critical Rules" section in your instructions. Add: "If you are not certain a fact appears in the documents I uploaded, say so explicitly and ask me to confirm."
• Output tone doesn't match your voice → Upload 2 more past status reports. The more examples of your writing Claude sees, the better it matches your style. You can also add specific instructions: "My status reports never exceed one page. Always use the exact section headers I use in past reports."
• Project context gets stale mid-engagement → Update your documents every 2–4 weeks. Replace the old risk register with the current version. Add new steering committee minutes. The project context is only as good as the documents you maintain.
• Confidentiality concern → By default, your Claude Pro data is not used to train Anthropic's models (check your settings at claude.ai → Settings → Privacy). For highly sensitive client work, check your firm's data policy before uploading client documents. If in doubt, use anonymized versions: replace client name with "[Client]" and remove PII before uploading.

Variations

• Simpler version: Don't upload any documents. Just write detailed Project Instructions with the key facts manually entered (scope, stakeholders, timeline). No file upload needed — just a well-written briefing. You lose document-search capability but gain a PM assistant that knows the project basics for any conversation.
• Extended version: Create one Claude Project per active engagement. Name them clearly. Now you have dedicated AI workspaces per project — switching between them is like switching between project files in your mind, except Claude holds the context.
• Team version: Claude for Teams ($30/user/month) allows you to share a Project with your full PM team. A shared project means your co-PM, PMO analyst, and junior coordinator all work from the same AI context — consistent tone, shared project knowledge, no one re-explaining scope to a chatbot.

What to Do Next

• This week: Set up the Project and upload your files. Run one status report draft through it. Compare the quality to what you'd write in a blank ChatGPT window — the difference in relevance will be immediate.
• This month: Add meeting notes from your last 3 steering committee sessions. Ask Claude "based on the pattern in these meetings, what is [Stakeholder Name] likely to push back on next Tuesday?" Use the answer to prep talking points.
• Advanced: Combine this with the Zapier pipeline (see guide: "Zapier Automation: Auto-Draft and Deliver Your Weekly Status Report"). Let Zapier pull and format Jira data, then paste that data as your input in the Claude Project each week. The Project provides the context; Zapier provides the data feed.

Advanced guide for project manager professionals. These techniques use more sophisticated AI features that may require paid subscriptions.