> workflow / project-kickoff
Project Kickoff
Align repo, context package, hooks, and model/tool policy so the team has a Definition of Ready before architecture deepens.
What should I do next?
Complete the steps and evidence below, clear the human gates, then continue to Architecture.
When to use
- Brief is approved and a repo (or monorepo path) will host the work
- Team IDE/model policy is unset or inconsistent across contributors
- No CLAUDE.md / project rules / secrets boundaries exist yet
- You need hooks and context before the first architecture ADR
Step-by-step
- Confirm repo access, branch protection, and secret scanning are on.
- Install AIES kit baseline (rules, CLAUDE.md template, stack skill pointer) for the chosen IDE.
- Build a context package: system map, data classes, allowed MCPs, forbidden paths.
- Wire pre-planning.hook.sh; do not enable broad tool access yet.
- Document model providers allowed for this initiative and data that must never enter prompts.
- Assign AIDLC roles (PM, Architect, Engineer, Reviewer) with named humans.
- Pass Definition of Ready before scheduling architecture workshops.
Decisions to make
- Which IDEs and models are approved for this repo
- Which MCPs are allowlisted vs deferred
- Where the context package lives and who updates it
- Hook policy: which gates are blocking vs advisory
Evidence to collect
- Context package committed and linked from README
- Hook baseline installed (at least pre-planning)
- Named owners for context, secrets, and model policy
- Definition of Ready checklist marked complete
Role-specific next actions
- PM: Confirm DoR; keep backlog frozen until architecture owns boundaries.
- Architect: Draft the initial system map and data-class inventory in the context package.
- Engineer: Install IDE rules/hooks and verify the harness runs locally.
- DevOps: Enable secret scanning, branch protection, and CI skeleton.
Inputs
- Approved brief and risk tier
- Repo access and team roster
- IDE + model policy constraints
Outputs
- Kickoff checklist complete
- Committed context package
- Hook and secrets baseline
Required prompts
- Implementation plan
Optional early scaffold only — full plan comes after architecture.
- Documentation
Bootstrap README / CLAUDE.md / onboarding docs from the context package.
Required skills
- Context
Builds the minimal high-signal package agents must respect.
- Documentation
Keeps kickoff artifacts living in-repo, not in slides.
Required MCP servers
- Filesystem MCP
Controlled local file access once allowlisted — not a substitute for context packages.
- GitHub MCP
Repo metadata and PR scaffolding after branch protection exists.
- Jira MCP
Optional: sync epics only after DoR — never invent scope from tickets alone.
Tooling & handbook
- How to give context
Practical context packaging for agents.
- Hooks overview
Install and configure the hook set.
- pre-planning hook
Blocks coding agents from skipping plan gates.
- CLAUDE.md template
Repo harness baseline for Claude Code / compatible setups.
- Definition of Ready
Exit criteria for kickoff.
- MCP when and how
Allowlist discipline before enabling tools.
Recommended agents & patterns
- Context pattern
Primary pattern for kickoff context packages.
- Context engineer agent
Assists packaging; human owns accuracy.
- Planner agent
May draft kickoff task lists — not product decisions.
Human approval gates
- Tech lead approves context package and tool/MCP allowlist
- Security (or delegate) confirms secrets and PII boundaries
Automation opportunities
- Generate CLAUDE.md / project rules from templates
- Wire pre-planning.hook.sh in CI and local agent runners
- Seed issue labels for risk tier and AIDLC role
Enterprise best practices
- Inventory models and data classes touched before first PR
- Assign accountable owner per AIDLC role with backup
- Record MCP allowlist in the same place auditors can find ADRs
Common mistakes
- Starting coding without Definition of Ready
- Omitting secrets and PII boundaries from the context package
- Enabling every MCP "just in case"