Configure instructions, three projects and a saved-prompt file for your real work, then measure the time saved over a week.
Write custom instructions, create three projects for your three most recurring areas of work with files and instructions, and keep a file of five saved prompts with variables. Use the setup for a week and log time saved per task and edits made.
Prepare a real repository so Claude Code can work in it safely: conventions, tests, checks and a release script, proven by one test-first feature.
Take a repository you own. Write CLAUDE.md, ensure the test command runs green, add a release-check script that prints an evidence table, then implement one small feature test-first with the tool and ship it with the table in the pull request.
From first session to a repeatable workflow: orientation, test-first changes, review and release.
Use Claude Code to make verified changes to a real codebase with tests, reviews and evidence.
Skills: Writing a CLAUDE.md that steers the tool, Running test-first changes, Reviewing diffs the tool proposes, Keeping the tool inside safe boundaries
Set up a Claude project that reviews contracts against your checklist and produces a referenced summary, tested on three real contracts.
Build a project whose instructions define a review checklist (payment, termination, liability, IP, renewal), an output format with section references and quotes, and a rule that ambiguous clauses are quoted and listed for a lawyer. Test on three contracts; verify every reference.
Build a read-only agent that answers questions with cited findings, within a fixed budget, and passes an evaluation set.
Build an agent with two tools — search and fetch — that answers a question with findings (each with URL and date), inferences (each with reasoning), and gaps. Enforce a budget of fifteen fetches and forty steps in code. Validate the report shape in code and reject uncited findings. Evaluate on twenty questions.
From a single tool call to a budgeted, evaluated, approval-gated agent that can be left running.
Design, build, evaluate and operate a tool-using agent with side effects gated by a person.
Skills: Defining tools with schemas and failure returns, Writing a system prompt as a contract, Enforcing budgets in code, Building an evaluation set, Gating side effects
Build five reusable prompts for one role's recurring tasks, each with variables, a reference run and a ten-case test set.
Choose one role (support, sales, marketing, ops). Identify its five most repeated writing tasks. For each, produce a prompt template with variables, a filled reference run, and a ten-case test set with rubrics and a recorded pass rate. Deliver as a folder someone in that role could use on Monday.
Write a two-page guide for your team on what your chosen assistant does well, where it fails, and the rules for using it, with every claim tested by you.
Produce a two-page internal guide for a real team: what the assistant is, three tasks it does well with evidence from your own tests, three ways it fails with examples you produced, and five rules for use. The guide must not contain a claim you did not test yourself.
Milestone 1: Test the strengths
Milestone 2: Reproduce the failures
Any assistant
Instant delivery after paymentLicence stated on every productRe-download anytime from My LibrarySecure checkout — Polar or FungiesNew products every week
Instant delivery after paymentLicence stated on every productRe-download anytime from My LibrarySecure checkout — Polar or FungiesNew products every week
