agentic-sdlc

Orchestrates multi-agent software delivery waves using Seeds queues, Git worktrees, and provider-native delegation.

Updated Jun 30, 2026
One-click install
npx skills add https://github.com/Codeseys-Labs/agentic-sdlc --skill agentic-sdlc-codeseys-labs
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agentic-sdlc
Source: https://github.com/Codeseys-Labs/agentic-sdlc/tree/main/plugin/skills/agentic-sdlc
Command: npx skills add https://github.com/Codeseys-Labs/agentic-sdlc --skill agentic-sdlc-codeseys-labs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Coordinating multi-agent software development across a whole project is usually improvised ad hoc: agents edit the same checkout, worker summaries get trusted as acceptance evidence, and nobody tracks what was actually verified. This Skill provides a repeatable Frame-to-Ship loop with a Seeds queue of record, isolated Git worktrees per workstream, and evidence-backed acceptance rules. ## Core Features & Use Cases - Phased SDLC loop: Frame, Discover, Research, Plan, Act, Review, Reconcile, and Ship phases with bounded backflow and hard stops for unsafe operations. - Seeds-backed worktree waves: A pinned, receipt-verified Seeds CLI acts as the queue of record while each write-capable worker runs in its own .worktrees/<seed-id>-<slug>/ worktree. - Provider-native delegation: Routes work to Claude Code subagents/workflows, Codex role agents, or direct execution, with cmux as an optional view/event layer only. - Use Case: You open a repository with a backlog of open seeds and ask the agent to drive it to zero; the Skill frames the mission, fans out implementers into worktrees, requires independent review and a re-run gate on the merged snapshot, and records wave evidence before any PR is proposed. ## Quick Start Ask the agent to run the agentic SDLC loop on this repository, starting with activation via /sdlc-init if no Seeds queue exists, then framing and executing the next ready seed in a worktree wave.

Frequently Asked Questions about agentic-sdlc

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-agent implementation wave on a repository?

Activate the repo first with /sdlc-init to establish the Git baseline, Seeds queue, and pinned gates. Then frame the work, dispatch one implementer per workstream into its own .worktrees/<seed-id>-<slug>/ worktree, and reconcile findings through the conductor-only Seeds record seam.

What is the Seeds queue in an agentic SDLC workflow?

Seeds is the queue of record for all work items, accessed through a pinned, receipt-verified launcher running exact Node 22.23.2 and Bun 1.4.0. Only the conductor mutates the queue, via a compare-and-swap record verb that refuses unrequested changes.

Does this workflow require tmux or cmux to run?

No. The baseline uses only the host's native agent and subagent capabilities. cmux is an optional view/event layer loaded only when already active or explicitly requested, and tmux is never part of the baseline probe.

Can a worker's summary be used as acceptance evidence?

No. Worker messages and status lines are advisory only. Acceptance requires reading commits and diffs from Git, a gate receipt re-run on the merged snapshot, independent review by a different node, and an operator approval dated before the fan-in.

What happens when no certified delegation route is available?

The run fails closed rather than improvising. A Frame may authorize exactly one bounded, non-delegated conductor execution for a small seed with an observable done condition, but any need for retries, parallel work, or additional workers stops the run.