From bb
Begin a coding session. Read the reference, orient to the code, repeat back intent, wait for confirmation, then build. Use when the user invokes /bb:dev-start.
How this skill is triggered — by the user, by Claude, or both
Slash command
/bb:dev-startThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Start a coding session cleanly. The goal is to make sure you and the user agree on what's being built before any code is written — even if there's a written plan, your mental model and theirs may not match yet.
Start a coding session cleanly. The goal is to make sure you and the user agree on what's being built before any code is written — even if there's a written plan, your mental model and theirs may not match yet.
If the line below has content after the colon, treat it as the user's pointer to the change (a plan file path, an issue link, a free-text description). Otherwise ask.
Pointer: $ARGUMENTS
If the user provided a plan file, issue link, or other document: read it in full. Don't skim. If the document references other docs (design notes, related plans, decision records), read those too if they're load-bearing.
If they gave a free-text description, treat the conversation itself as the spec.
If there's no pointer at all, ask the user to describe what they're working on before continuing.
Check the current state of the code the change will touch. Names, shape, tests, callers. If a plan was written days or weeks ago, code drift is common — note any gap between what the plan assumes and what's there now.
If the project has a CLAUDE.md, read it for conventions (test commands, lint, anything project-specific you'd otherwise have to guess).
If anything is ambiguous, under-specified, or seems wrong, ask before building. Batch questions into one message where possible.
If everything is clear, skip this step rather than inventing questions.
In your own words, state:
Keep it tight. A few bullets, not a paragraph.
Do not start building until the user confirms. If they correct your understanding, revise and confirm again.
Once confirmed, implement. Follow the project's normal workflow — write tests as you go, commits at meaningful checkpoints, align to the project's CLAUDE.md conventions.
When the implementation work is done, hand back to the user. This skill ends at "I'm done coding." Wrapping up — tests, decisions worth recording, the commit — is /bb:dev-finish, a separate invocation.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
npx claudepluginhub govtech-bb/team-skills --plugin bb