You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The current speckit workflow requires manual invocation of each command in sequence: specify → clarify → plan → tasks → analyze → implement. While the handoffs mechanism in the frontmatter provides UI buttons for the next step, the user still needs to manually click/invoke each transition.
This creates unnecessary friction — especially for users who have already internalized the workflow and just want to run the full pipeline from a feature description to implementation, stopping only when actual user input is needed (e.g., clarification questions, critical issues in analysis).
Proposal
Add a new pipeline.md command (/speckit.pipeline) that orchestrates the full SDD workflow automatically:
/speckit.specify — Create feature specification
/speckit.clarify — Resolve ambiguities (pauses for user input)
/speckit.plan — Generate technical plan
/speckit.tasks — Break plan into actionable tasks
/speckit.analyze — Cross-artifact consistency analysis (pauses on CRITICAL issues)
/speckit.implement — Execute all tasks
Key behaviors:
Auto-progresses between steps without returning control to the user
Only pauses when a step genuinely requires user input (clarification questions, critical analysis findings, incomplete checklists)
If any step fails, stops and reports the error instead of blindly continuing
Minimal progress reporting (individual commands already produce their own output)
Why not use existing solutions?
Handoffs: Require manual user interaction at each transition
Community extensions (Conduct, Fleet, AIDE): Solve a similar problem but with significantly more complexity (multi-agent orchestration, coordination files, subagent delegation). The pipeline command is a lightweight, zero-config alternative
Issue Fully Self-Automate Agent #1192 (Fully Self-Automate Agent): Asks for this capability but no simple solution has been proposed yet
Design
The command is a single markdown file (templates/commands/pipeline.md) that instructs the AI agent to sequentially invoke each speckit command. It leverages the agent's native ability to call slash commands (e.g., via Claude Code's Skill tool, Copilot's chat commands, etc.).
No changes to existing commands, scripts, or templates are required.
AI Disclosure
This issue and the associated PR were authored with assistance from Claude Code (Claude Opus 4.6). The concept, workflow design, and rationale are human-directed. The command template was written by Claude and reviewed by the contributor.
Problem
The current speckit workflow requires manual invocation of each command in sequence:
specify→clarify→plan→tasks→analyze→implement. While thehandoffsmechanism in the frontmatter provides UI buttons for the next step, the user still needs to manually click/invoke each transition.This creates unnecessary friction — especially for users who have already internalized the workflow and just want to run the full pipeline from a feature description to implementation, stopping only when actual user input is needed (e.g., clarification questions, critical issues in analysis).
Proposal
Add a new
pipeline.mdcommand (/speckit.pipeline) that orchestrates the full SDD workflow automatically:/speckit.specify— Create feature specification/speckit.clarify— Resolve ambiguities (pauses for user input)/speckit.plan— Generate technical plan/speckit.tasks— Break plan into actionable tasks/speckit.analyze— Cross-artifact consistency analysis (pauses on CRITICAL issues)/speckit.implement— Execute all tasksKey behaviors:
Why not use existing solutions?
Design
The command is a single markdown file (
templates/commands/pipeline.md) that instructs the AI agent to sequentially invoke each speckit command. It leverages the agent's native ability to call slash commands (e.g., via Claude Code's Skill tool, Copilot's chat commands, etc.).No changes to existing commands, scripts, or templates are required.
AI Disclosure
This issue and the associated PR were authored with assistance from Claude Code (Claude Opus 4.6). The concept, workflow design, and rationale are human-directed. The command template was written by Claude and reviewed by the contributor.
Related: #1192