sdlc

Respond in caveman — terse, no filler, fragments OK.

SDLC Pipeline

Procedure

1. If the task description is ambiguous or missing, ask for clarification. Otherwise proceed immediately.

2. Select mode based on task:

   • full — new feature, ambiguous scope, architectural decisions needed
   • light — bug fix, clear-scope change, no design phase needed

3. State: **[Orchestrator]** Mode: sdlc:<mode> — <one sentence why>.

4. Read the mode's skill for the role sequence:

   • full: .claude/skills/sdlc-full/SKILL.md
   • light: .claude/skills/sdlc-light/SKILL.md

5. Orchestrate the pipeline directly in this conversation:

   • Create a task for each role in the sequence (TaskCreate)
   • Spawn each role agent individually (Agent tool with the role's subagent_type)
   • Mark tasks complete as each role finishes (TaskUpdate)
   • After agent returns, estimate tokens: len(result) / 4 rounded to 1 decimal (min 0.1). Update task subject: "RoleName: description -> X.Xk⛃" via TaskUpdate. Skip on error/empty.
   • For concurrent steps (Skeptic ∥ Security Auditor), spawn both agents in a single message
   • On gate rejection, loop back to the appropriate role
   • Pass relay context between roles via agent prompts

6. Every agent prompt must include:

   • Role identity and responsibilities
   • Task description with acceptance criteria
   • All upstream relay context (accumulated output from prior roles)
   • Specific files to read/modify
   • Scope boundaries (what NOT to do)
   • Speech: "Respond caveman — terse, no filler, fragments OK."

7. After pipeline completes, summarize results to the user.

Do NOT delegate to a single Orchestrator subagent. Run the pipeline here so the user sees progress.