원클릭으로
speckit-ralph-implement
Orchestrate task-by-task implementation (Ralph loop) until all tasks are complete.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Orchestrate task-by-task implementation (Ralph loop) until all tasks are complete.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Orchestrate iterative spec clarification and remediation (Homer loop) on spec.md until all findings are resolved.
Orchestrate iterative cross-artifact analysis and remediation (Lisa loop) on spec.md, plan.md, and tasks.md until all findings are resolved.
Orchestrate iterative code review and remediation (Marge loop) over the feature branch diff until all findings are resolved.
Orchestrate the full SpecKit pipeline (reconcile, specify, homer, phase, plan, tasks, lisa, split, ralph, marge) from feature description to reviewed implementation.
Post a PR review with inline comments for findings requiring human judgment — one-way doors, concurrency risks, architectural decisions, and project-specific patterns.
Analyze a feature branch diff against baseline and project-specific review packs; optionally remediate the auto-fixable findings in severity order.
| name | speckit-ralph-implement |
| description | Orchestrate task-by-task implementation (Ralph loop) until all tasks are complete. |
$ARGUMENTS
You MUST consider the user input before proceeding (if not empty).
Ralph uses two gates: an optional fast scoped gate (.specify/quality-gates-fast.sh) per iteration and the full gate (.specify/quality-gates.sh) once after the loop terminates.
Before configuring the loop, validate that quality gates are set up. Run the following checks using the Bash tool:
test -f .specify/quality-gates.sh && echo "EXISTS" || echo "MISSING"grep -v '^\s*#' .specify/quality-gates.sh | grep -v '^\s*$' | head -1test -f .specify/quality-gates-fast.sh && echo "EXISTS" || echo "MISSING"grep -v '^\s*#' .specify/quality-gates-fast.sh | grep -v '^\s*$' | head -1If the full gate file is MISSING or its grep produces no output, display this error and STOP:
ERROR: Quality gates file is missing or empty.
Expected: .specify/quality-gates.sh with executable commands
Found: File missing or contains only comments/whitespace
The full quality gates file is required for Ralph to validate task implementations.
Create or update .specify/quality-gates.sh with your project's quality gate
commands (e.g., npm test && npm run lint). The file must exit 0 for gates to pass.
Optionally, also create .specify/quality-gates-fast.sh with scoped commands
that check only changed files for faster per-iteration feedback.
If the fast gate file is missing, note this for the LOOP_CONFIG below — the full gate will be used per-iteration instead.
After resolving FEATURE_DIR (the orchestrator handles this), count tasks in FEATURE_DIR/tasks.md:
grep -c '^\s*- \[ \]' "<FEATURE_DIR>/tasks.md"grep -c '^\s*- \[x\]' "<FEATURE_DIR>/tasks.md"incomplete_tasks + 10Use the calculated value as MAX_ITERATIONS in the loop configuration below.
Set the following LOOP_CONFIG values for this execution:
Read and follow the instructions in .claude/agents/loop-orchestrator.md, using the LOOP_CONFIG values above. Pass $ARGUMENTS through for argument parsing.
Run only when the loop exited via the success path (<promise>ALL_TASKS_COMPLETE</promise> observed). Skip this step on max-iterations, stuck, or failure exits.
Run via Bash tool:
bash .specify/quality-gates.sh
If it exits non-zero, treat the loop as incomplete: the fast gate missed a regression in files outside the per-iteration scope. Set the completion status to failure with reason "end-of-loop full quality gates failed", surface the failing output in the report, and suggest rerunning ralph or fixing the issue manually before re-running this command.
If it exits zero, proceed to tasks verification.
After the loop orchestrator reports completion via the promise tag, also verify tasks.md directly — if no - [ ] lines remain and at least one - [x] exists, confirm completion. If incomplete tasks remain despite the promise tag, report the discrepancy.
/speckit-ralph-implement — Auto-detect spec dir from current branch, use default max iterations (incomplete_tasks + 10)/speckit-ralph-implement specs/003-fix-pipeline-delegation — Run for specific spec dir/speckit-ralph-implement 5 — Auto-detect spec dir, limit to 5 iterations/speckit-ralph-implement specs/003-fix-pipeline-delegation 5 — Specific spec dir with 5 max iterations