Agentic Playbook

Agentic product lifecycle

An agentic SDLC that starts at design, not at code: agents do the making and people call the shots. It takes an idea to production from nothing, and works just as well inside a client's existing stack.

The board

Agents write the spec, lay down the design system, scaffold the project, build the screens, run the checks, ship, and watch what is live. A person says what to make, whether it is good and whether it goes out. Quality does not depend on anyone remembering: gates on every commit and pull request enforce it, and a build that fails accessibility or performance does not get to argue.

Most of the time the work is a client's codebase, not a blank one, so each skill owns a purpose rather than a tool and reads which tool fills it from the project's own stack manifest. The client keeps their framework, CMS, tracker, CI and the way their team already works. We keep our opinions: a spec written to be proved wrong, gates rather than ceremony, accessibility and performance as thresholds a build clears rather than numbers admired afterwards, and a person deciding what ships.

None of this works if a skill is just a page of instructions. What goes into one, and how it gets better every time it runs, is what makes a good agentic skill.

Every cell opens its own record, in place: what the skill is for, and which tools can fill that purpose. Install any of them with npx skills add madebymany/madebyagents --skill <name>, or read one on GitHub from its record — how it is built, and what holds it, are in the SKILL.md itself. Both need access to the repo.