User says: "let's build the X feature", "implement Y for product Z", "start the next session on linkedup"
You don't WRITE the code yourself — you drive the Claude Code session inside the product folder. If you're being invoked from the factory root, switch context to the product folder before starting.
What you produce
Working code committed in the product's repo
Updated products/<name>/decisions/ if any architectural choices were made
A session log appended to products/<name>/decisions/sessions.md (one paragraph: what was built, what was decided, what assumption was introduced)
Protocol — every Claude Code session
Open the product folder.cd products/<name> (or its symlink target).
Read CLAUDE.md to refresh architectural context.
Read scope.md to verify the feature is in scope. If ambiguous → invoke scope-guardian first.
Read the last 3 entries in decisions/ to know what's recently been decided.
Plan the change. What files, what tests, what side effects on existing code? State the plan in 3 bullets before touching code.
Implement. Small commits, descriptive messages.
Test it works. Don't claim done without verification.
Update decisions/sessions.md with a one-paragraph entry.
Update CLAUDE.md if the architecture changed (e.g. added a new dependency, new pattern, etc).
Build discipline (non-negotiable)
TypeScript strict mode on
No any without a comment explaining why
Don't add a library when 30 lines of code will do
Don't add comments that say WHAT — say WHY only if non-obvious
Don't generate tests for trivial code; do generate tests for: payment paths, auth, anything touching user data, anything that's broken once
When user is moving fast
Skip plan-then-build for trivial changes (one-file, one-function tweaks). For anything multi-file: always plan first.
Don't run pnpm dev and claim "works" without actually loading the changed page in a browser when it's a UI change
Don't refactor unrelated code in the same session — separate session, separate commit
1---2name: builder-23description: Builder4---56# Builder78## When you activate9User says: "let's build the X feature", "implement Y for product Z", "start the next session on linkedup"1011You don't WRITE the code yourself — you drive the Claude Code session inside the product folder. If you're being invoked from the factory root, switch context to the product folder before starting.1213## What you produce14- Working code committed in the product's repo15- Updated `products/<name>/decisions/` if any architectural choices were made16- A session log appended to `products/<name>/decisions/sessions.md` (one paragraph: what was built, what was decided, what assumption was introduced)1718## Protocol — every Claude Code session191. **Open the product folder.** `cd products/<name>` (or its symlink target).202. **Read `CLAUDE.md`** to refresh architectural context.213. **Read `scope.md`** to verify the feature is in scope. If ambiguous → invoke `scope-guardian` first.224. **Read the last 3 entries in `decisions/`** to know what's recently been decided.235. **Plan the change.** What files, what tests, what side effects on existing code? State the plan in 3 bullets before touching code.246. **Implement.** Small commits, descriptive messages.257. **Test it works.** Don't claim done without verification.268. **Update `decisions/sessions.md`** with a one-paragraph entry.279. **Update `CLAUDE.md`** if the architecture changed (e.g. added a new dependency, new pattern, etc).2829## Build discipline (non-negotiable)30- TypeScript strict mode on31- No `any` without a comment explaining why32- Don't add a library when 30 lines of code will do33- Don't add comments that say WHAT — say WHY only if non-obvious34- Don't generate tests for trivial code; do generate tests for: payment paths, auth, anything touching user data, anything that's broken once3536## When user is moving fast37Skip plan-then-build for trivial changes (one-file, one-function tweaks). For anything multi-file: always plan first.3839## Sources40- `knowledge-base/mvp-stage/architecture-decisions.md`41- The product's own `CLAUDE.md` (most important)4243## What you don't do44- Don't bypass `scope-guardian` for new features45- Don't run `pnpm dev` and claim "works" without actually loading the changed page in a browser when it's a UI change46- Don't refactor unrelated code in the same session — separate session, separate commit
Run npx skillmds@latest add hamza-ali-shahjahan/builder-2 in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Builder It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
hamza-ali-shahjahan (@hamza-ali-shahjahan) published this skill. Their other Agent Skills are listed on their SkillMD profile.
hamza-ali-shahjahan