implement
SKILL.md
Implement: execute a task from a Breakdown
When the user types /implement and references a task file (e.g., @projects/<domain-name>/<project-name>/tasks/01-add-feature-flag.md), do the following:
1) Understand the task and its context (required)
- Read the task file end-to-end.
- Identify the goal, acceptance criteria, key touch points, and risks.
- Read the parent Plan (
plan.md) and Breakdown (tasks/*.mdoverview file) for broader context. - Check Dependencies — if a blocking task is incomplete, warn the user before proceeding.
- Search the codebase for the key touch points to understand existing patterns and conventions.
2) Shape Up principles for implementation
Follow the Shape Up "hill climbing" model:
Uphill (figuring it out)
- Explore unknowns first — if there are risks or unclear patterns, investigate before writing code.
- Find existing examples in the codebase that match the task's patterns.
- Identify the smallest working vertical slice that proves the approach.
Downhill (making it happen)
- Implement incrementally — small, reviewable commits.
- Follow TDD when the task specifies it: write failing tests first, then implement to pass.
- Don't gold-plate — implement exactly what's in the acceptance criteria, nothing more.
3) Implementation workflow
Step 1: Pre-flight checks
- Verify dependencies are complete (warn if not)
- Identify existing patterns to follow
- List the files that will be created or modified
Step 2: Write tests first (if TDD task)
For tasks marked "(TDD)":
- Create the spec file(s) listed in Key touch points
- Write test cases for each acceptance criterion
- Run tests — they should fail (red)
- Show the user the failing tests before implementing
Step 3: Implement the feature
- Create/modify files listed in Key touch points
- Follow existing codebase conventions (Sorbet types, module structure, naming)
- Reference the Implementation notes and Risks sections for guidance
- Run tests after each significant change
Step 4: Verify acceptance criteria
Go through each acceptance criterion checkbox:
- Criterion 1 — explain how it's satisfied
- Criterion 2 — explain how it's satisfied
- ...
Step 5: Final checks
- All tests pass
- Code follows existing patterns
- No TODOs or placeholders left behind
Step 6: Post-completion tooling (run as needed)
Run the following tools based on which files were changed:
Ruby files (.rb)
-
bundle exec rubocop -a <changed files>— ensure linting passes -
bin/tapioca dsl— regenerate Sorbet RBI files if models/classes were created/modified
TypeScript/React files (.tsx, .ts)
-
npm run prettier -- --write <changed files>— format files -
npm run tsc— ensure TypeScript compilation passes
GraphQL changes
-
rake graphql:schema:dump— regenerate schema if GraphQL types/resolvers changed -
npm run codegen— regenerate client types if.graphqlfiles or schema changed
Packwerk changes
-
bin/packwerk update-todo— update packwerk todo if package boundaries changed
4) Output format
Structure your response as:
## Task: [Task title]
### Pre-flight
- Dependencies: [status]
- Patterns identified: [list]
- Files to touch: [list]
### Implementation
[Show code changes with explanations]
### Acceptance criteria verification
- [x] Criterion 1 — [how satisfied]
- [x] Criterion 2 — [how satisfied]
...
### Next steps
- [Any follow-up or blocked tasks now unblocked]
5) Important rules
- Don't skip tests — if the task is marked TDD, write tests first.
- Don't over-engineer — implement only what's in the acceptance criteria.
- Don't break existing functionality — run existing tests if touching shared code.
- Ask before major decisions — if you encounter something not covered by the task, ask the user.
- Update the task file — after completion, offer to mark acceptance criteria as checked.
6) When blocked
If you cannot complete a task because:
- A dependency task is incomplete → list what's missing and which task provides it
- The codebase doesn't match expectations → describe the discrepancy and ask for guidance
- A decision is needed → present options with tradeoffs
Do not guess or make assumptions on architectural decisions not specified in the task.
Weekly Installs
4
Repository
dailydm/skillsFirst Seen
Feb 27, 2026
Security Audits
Installed on
opencode4
claude-code4
github-copilot4
codex4
amp4
cline4