Speckit.Pipeline

---
description: Orchestrate the full SpecKit pipeline (specify, homer, plan, tasks, lisa, ralph, marge) from feature description to reviewed implementation.
---

## User Input

```text
$ARGUMENTS

Pre-Flight Check

Before doing anything else, verify that the required utility scripts are installed:

1. Check if .specify/scripts/bash/check-prerequisites.sh exists (use the Bash tool: test -f .specify/scripts/bash/check-prerequisites.sh && echo "EXISTS" || echo "MISSING")
2. If MISSING, display this error and STOP — do not proceed with any pipeline execution:

ERROR: Required utility script not found.

Missing: .specify/scripts/bash/check-prerequisites.sh

This script is required for feature directory resolution and prerequisite validation.
To install it, run the SpecKit setup command:

  /speckit.setup

3. If EXISTS, proceed to the agent file check below.

Agent File Check

Verify that all required agent files exist before starting the pipeline. Check each of these files using the Bash tool:

for f in specify homer plan tasks lisa ralph marge; do
  test -f ".claude/agents/${f}.md" && echo "${f}.md: EXISTS" || echo "${f}.md: MISSING"
done

If any agent file is MISSING, display this error and STOP — do not proceed with pipeline execution:

ERROR: Required agent file(s) not found.

Missing: .claude/agents/<name>.md

Agent files are required for pipeline sub-agents to execute. These files define
the behavior of each pipeline phase. Ensure all agent files are present:
  .claude/agents/specify.md
  .claude/agents/homer.md
  .claude/agents/plan.md
  .claude/agents/tasks.md
  .claude/agents/lisa.md
  .claude/agents/ralph.md
  .claude/agents/marge.md

If all agent files exist, proceed to the Overview section below.

Overview

Orchestrate the full SpecKit pipeline directly within this session. Each step spawns fresh sub agents (via the Agent tool) with isolated context windows. The pipeline can start from a feature description (using the specify step) or from an existing spec.md.

AUTONOMOUS EXECUTION: This pipeline runs unattended. Do NOT ask the user for confirmation between iterations or steps. Do NOT pause for permission requests. Execute all steps and iterations back-to-back until the pipeline completes or a failure condition is met.

STRICT SEQUENTIAL EXECUTION: Each sub agent MUST complete and return its result before the next sub agent is spawned. Never run multiple sub agents in parallel. Within each loop step, wait for one iteration to finish before starting the next. Between pipeline steps, wait for the entire step to complete before advancing. The pipeline order is non-negotiable:

The pipeline runs these 7 steps in sequence:

0. specify — Create feature spec from description (optional, auto-detected)
1. homer — Iterative spec clarification & remediation
2. plan — Generate technical implementation plan
3. tasks — Generate dependency-ordered task list
4. lisa — Cross-artifact consistency analysis
5. ralph — Task-by-task implementation with quality gates
6. marge — Iterative code review of the implementation

Between ralph and marge, two optional polish phases run automatically if and only if the corresponding skill is installed in the environment:

• simplify — invokes the /simplify skill for reuse/quality/efficiency fixes
• security-review — invokes the /security-review skill for a security audit

If either skill is absent, that phase is silently skipped. These phases are not part of the --from / --stop-after step mapping; they always run between ralph and marge when present, and are skipped when --stop-after ralph halts the pipeline.

Instructions

Step 1: Parse Arguments and Determine the spec directory

Parse $ARGUMENTS for the following (all are optional, can appear in any order):

• --from <step>: Starting step override. Valid values: specify, homer, plan, tasks, lisa, ralph, marge. If provided, the pipeline starts from this step instead of auto-detecting.
• --stop-after <step>: Stop-after step. Valid values: specify, homer, plan, tasks, lisa, ralph, marge. If provided, the pipeline halts after the specified step completes, skipping all subsequent steps. Store the value in STOP_AFTER_STEP. If --stop-after is NOT provided, STOP_AFTER_STEP MUST remain empty/unset so that all stop checks are no-ops and the pipeline runs all steps from the starting step through marge — identical to the behavior before --stop-after was added (FR-007). If --stop-after is present but no step name follows (e.g., it is the last argument or the next token is another flag), display an error: "Error: --stop-after requires a step name. Valid steps: specify, homer, plan, tasks, lisa, ralph, marge." and STOP.
• --description <text>: Feature description for the specify step. Capture the full text after --description (may be quoted).
• spec-dir: A directory path (e.g., specs/003-fix-pipeline-delegation). If provided, use it as FEATURE_DIR.

If no spec-dir is provided in $ARGUMENTS, resolve FEATURE_DIR automatically:

1. Determine if on a feature branch: Run git rev-parse --abbrev-ref HEAD via Bash tool. If the branch matches the pattern ^[0-9]{3}- (e.g., 004-fix-prereq-bootstrap), it is a feature branch.

2. Feature branch resolution: Run bash .specify/scripts/bash/check-prerequisites.sh --json --paths-only from repo root via Bash tool. This resolves paths without validating artifact existence. Parse the JSON output for FEATURE_DIR.

3. Non-feature branch (e.g., main, HEAD) with bootstrapping: If on a non-feature branch and --from specify is set or --description is provided, skip check-prerequisites.sh entirely (it would reject the non-feature branch). Proceed with an empty/unresolved FEATURE_DIR — the specify step will create the feature branch and directory.