sdlc-task

sdlc-task Skill

Platform-agnostic definition for the autonomous (ungated) SDLC pipeline.

This skill runs the same 7 roles as the standard SDLC pipeline (Architect, Security Architect, Team Lead, Engineer, Code Reviewer, Quality Engineer, Security Auditor) but without human approval gates. All gates are advisory — agents produce their artifacts, surface findings, and self-advance.

When a later gate discovers something that changes architecture, scope, or security posture, the pipeline loops back to Gate 1 to re-evaluate with the new information, then continues forward. This makes the pipeline self-improving, not just advisory.

Behavior

• All 7 roles execute in sequence and produce their standard artifacts
• No gate blocks on human approval — agents self-advance
• Self-correction: discoveries that change architecture, scope, or security posture trigger a loop back to Gate 1 (max 2 re-evaluation cycles to prevent infinite loops)
• Finding mitigation is driven by the invoker's directions — findings within scope are mitigated, findings outside scope are flagged
• Audit trail is mandatory (REQ-2), entries marked as UNGATED
• All common files (CYNEFIN.md, PERSONALITY.md, LESSONS.md, REQUIREMENTS.md) are loaded

Pipeline

See .agents/skills/sdlc-task/PIPELINE.md for the full pipeline process definition.

Invocation Context

This skill may be invoked:

1. Directly by a human for lower-ceremony tasks
2. By any pipeline that needs autonomous SDLC-quality analysis without blocking