From dl
Implements tasks from a plan and design, applying rules and saving implementation notes. Use when coding a planned feature or working through tasks one at a time.
How this skill is triggered — by the user, by Claude, or both
Slash command
/dl:implementThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Execute each section in order. Copy the checklist and check off items as you complete them. Do not proceed past a **Gate** until verified.
Execute each section in order. Copy the checklist and check off items as you complete them. Do not proceed past a Gate until verified.
Artifact: Write an implementation note to .work/implementations/. Verify with ls before wrapping up — even if the task is incomplete.
Constraint: You MUST implement only the selected task. Note out-of-scope discoveries in the implementation note rather than acting on them.
Check .work/active/ for marker files. If exactly one exists and no $ARGUMENTS provided, auto-select that feature's slug. If multiple markers exist, list them and ask. Arguments always override.
Parse $ARGUMENTS as <slug> or <slug> <task-N>. Find the plan in .work/plans/ and matching design in .work/designs/. Missing design → print "No design found. Run /design first." and stop. No argument → list plan/design pairs; none: stop; one: confirm; many: ask to pick.
Task Progress:
- [ ] Find plan and design by slug
- [ ] Read plan, design, and .mmd diagrams
- [ ] Sync plan tasks to Claude Code task list (match by subject, don't duplicate)
- [ ] Pick task (from args or ask) and mark in_progress
- [ ] Run /resolve-rules for applicable rules
- [ ] Read relevant files and run baseline tests
- [ ] Implement against task spec
- [ ] Re-run tests; mark completed
- [ ] Write implementation note (see implementation-note.md)
- [ ] Gate: verify note with ls; update active marker
- [ ] Wrap up: summarize, note deviations, suggest commit
Read plan and design in full. Read .mmd diagrams from .work/designs/diagrams/ to understand proposed architecture. Sync plan tasks to Claude Code task list: call TaskList, then TaskCreate for any missing tasks — match by subject before creating. Print task list with completion status.
If $ARGUMENTS includes a task number, use it. Otherwise ask. Warn if dependencies are incomplete. Mark in_progress with TaskUpdate.
If the task spec has a **Rules:** line, run /resolve-rules mode:explicit <titles> as a subtask. Otherwise, run /resolve-rules mode:keyword <task terms>. If unavailable, continue without rules. Follow matched rules when modifying files under their scope. List applied rules in your response.
Read all relevant files before editing. Run relevant tests to establish a baseline. Implement against the task spec (Goal, Interfaces, Acceptance criteria from design). Re-run tests — note failures or unexpected results. Mark completed with TaskUpdate when fully done.
Write the note per implementation-note.md. Save to .work/implementations/YYYY-MM-DD-<slug>-task-N.md. In the Deviations section, list what design specified vs. what was done and why. This is read by /review to distinguish intentional deviations from violations.
Verify implementation note with ls; write if missing. Update .work/active/<slug>.md with stage: implement and today's date.
Summarize changes (files created/modified). Note deviations — suggest /design refine if significant. If out-of-scope work discovered, suggest /plan refine. Suggest a git commit scoped to this task.
npx claudepluginhub minusblindfold/devloop --plugin dlExecutes implementation plans from tasks.md files with plan review, batch task processing, verifications, and code review checkpoints.
Implements approved plans from .claude/scratchpad files by reading code fully, executing phases sequentially, verifying work, and updating checkboxes.
Executes approved implementation plans with checkpoint validation, progress tracking, and stakes-based enforcement. Locates plans, verifies approval, and runs verification criteria before proceeding.