Project Leadership: Claude Code Context

This file gives Claude persistent context about who we are, how we work, and the standards that apply to every task in this project. Edit the [PLACEHOLDER] fields before your first session. Keep it under 200 lines; move reference material into a skill.

Firm & Practice Context

• Firm: [YOUR FIRM NAME]
• Practice: [e.g., Enterprise Advisory & Transformation]
• Engagement type: [e.g., ERP implementation / Data modernization / Process transformation]
• Client: [CLIENT NAME]: [one-liner: industry, company size, what they do]
• Engagement phase: [e.g., Discovery / Design / Build / UAT / Hypercare]
• Engagement lead: [Your name and title]
• Key client contacts: [Name, Title; Name, Title]
• Contract value / duration: [e.g., $850K / 12 months ending Dec 2026]

Communication Standards

All client-facing content must be:

• Professional and precise: no filler phrases, no hedging without substance behind it
• Audience-appropriate: executive summaries at the C-suite level; working docs for the delivery team
• Action-oriented: every recommendation names an owner and a deadline
• Formatted for reuse: headers, bullets, and tables so output drops into PowerPoint or Word with minimal editing

Internal working documents may be more direct and less formal.

When tone is not specified, default to professional consulting style: confident, concise, recommendation-forward.

Confidentiality Rules

• Reference client files by path; never paste raw client data into the prompt. Claude reads the file directly.
• Treat all PII, financial figures, and proprietary process details as confidential.
• When in doubt, describe the data structure rather than sharing the data itself.
• Real engagement files live in my-projects/, which is gitignored. Session-specific sensitive notes go in CLAUDE.local.md, also gitignored.
• Do not include client names or identifying details in commit messages.
• Never name one client in another client's deliverable.

What Claude Does Not Do Here

These lines hold on a busy day because they are written down, not because anyone remembers them.

• Never sends. Email, chat messages, anything outbound: drafts only. A person presses send.
• Never invents a number, an owner, or a date. Use [TBD] or [NEEDS INPUT]. A blank is fixable in ten seconds; an invented figure that reached a client is not.
• Never settles a RAG rating. Propose it with the reasoning. The delivery lead confirms or overrules.
• Never treats memory as a source. Anything recalled from an earlier session is a hint about where to look. Re-read the file before writing a fact from it.

See docs/keep-leadership-human.md for the two questions behind these rules.

Common Artifact Types

Default Output Formats

• Status: RAG indicators (🔴 Red / 🟡 Amber / 🟢 Green) for overall status and each workstream. Those three words only; no "yellow," "at risk," or "on track" as a status value.
• Action items: Always include: Action, Owner, Due Date, Priority (High/Med/Low)
• PowerPoint-ready content: Short bullets (12 words or fewer), no nesting deeper than 2 levels
• Word/email content: Full sentences, structured headers, professional closing
• Tables: Use Markdown tables; they paste cleanly into Word and most PM tools
• Provenance: Every client-facing draft ends with Drafted with Claude Code on YYYY-MM-DD; reviewed by [NAME]. The reviewer fills in their name before it goes out. A draft with the placeholder still in it has not been reviewed.

Skills

Skills live in .claude/skills/. Type / to browse them, or describe the task in plain language and Claude picks the matching skill.

Subagent: ask for "a second look" or "fresh eyes" on a file and Claude delegates to the second-reviewer agent in .claude/agents/, which reads the file cold with no memory of the drafting conversation.

Project Notes

Add running notes here as the engagement progresses. Commit updates so the whole team shares context. Claude also keeps its own auto memory per repository; this section is the shared, human-written record.

Key decisions:

• [Date] | [Decision] | [Rationale] | [Owner]

Open risks:

• [Risk description] | [Owner] | [Status]

Team / capacity notes:

• [Person] | [Role] | [Availability / travel / coverage notes]