sdlc-brief

sdlc-brief

Lightweight orchestrator for tasks a human has already sized as XS/S/M. Picking this skill is the complexity classification — there is no classifier inside it. Everything runs inline, in the current conversation, on the current feature branch.

Cost tradeoff: the full pipeline (sdlc-engine, invoked via sdlc-guided / sdlc-auto) buys audit depth with subagents-per-plan-task, an evidence trail, and worktree isolation. sdlc-brief spends none of that budget. If a task turns out to need it mid-flow, stop and hand it back rather than bolting the heavy machinery on partway through.

Entry points into the parent plugin

When the project owns implementation

Before any implementation, read .agentic/agentic-sdlc/config.json if it is there. Where it declares project orchestrators, this skill owns intake, research, spec, plan, review and QA — and nothing that writes code. Route the code-writing to the selected project orchestrator and consume only its summary and gate results. Do not invoke the project's fleet workers yourself, and do not open a competing orchestration state file; the project's own is authoritative for the work it is running.

Non-goals

• No effort-sizing step of its own — sizing-analyst only returns a routing verdict, never a gate.
• No meta.json, no per-phase status slots.
• No run ledger beyond the optional decisions.jsonl / events.jsonl that gate-arbiter writes on its own.
• No git worktrees under any host.
• No load of the heavy-pipeline memory directory (.agents/memory/sdlc/) — never read by this skill, in any stage, in any mode.
• No autonomous or stand-in approval at any judgment gate — HITL only, always.
• No per-task evidence/*.json — the conversation transcript is the record.
• No feature verification unless the ui input is explicitly true.