talos

evo-talos Devkit Plugin

End-to-end SDLC orchestration kit for Claude Code and Codex. The plugin packages role agents, slash commands, on-demand skills, operating rules, safer settings, runtime hooks, and Codex-compatible skill metadata so a project can move from requirements to architecture, design, implementation, deployment, and QA with explicit artifacts and gates.

For a visual walkthrough, open the static guide:

• Usage guide - workflow diagram, entry tracks, design flows, FE/BE framework tracks, QA/deployment tracks, and artifact ownership.

Token Usage Notice

This kit can consume a large amount of tokens. That is intentional: evo-talos is designed for SDLC governance, artifact traceability, role separation, validation gates, QA evidence, and keeping a project on track rather than for minimal one-shot code edits.

Use it when the cost of coordination mistakes, missing requirements, weak QA, or uncontrolled implementation drift is higher than the cost of extra model usage. For small throwaway changes, a direct Claude Code prompt may be more efficient.

What This Plugin Does

evo-talos turns Claude Code from a single coding assistant into a structured SDLC coordinator. It gives Claude a project-local operating model with:

• Role-based agents for BA, SA, TL, UI/UX Designer, BE Dev, FE Dev, DevOps, QA-Author, QA-Exec, and support roles.
• Slash commands for project setup and autonomous SDLC progression.
• Skills that load detailed instructions only when needed, reducing runtime token overhead.
• Hooks that guard privacy, plan consistency, SRS sign-off, source-code writes, Docker scope, QA evidence, and dependency safety.
• Artifact ownership rules so each role writes only the documents it owns.
• Worktree isolation and plan-update.json protocols for parallel execution.

When To Use It

Use this kit when you want Claude Code to help run a disciplined software delivery flow rather than just edit files. It supports:

• Greenfield projects starting from a PRD or raw requirement document.
• Projects with an existing SRS and per-story/per-FR documentation.
• Requirements stored across a folder of source documents.
• External-source ingestion from systems such as Confluence, Notion, Jira, or SharePoint when connectors are available.
• Brownfield onboarding where code exists but trustworthy SRS/architecture docs do not.
• Figma-first product work where design content needs to feed the SRS.
• Iterative scope changes after sign-off.

Main Workflow

1. Install or enable the plugin in Claude Code.
2. Run /sdlc-init inside the target project.
3. Provide the starting source: PRD, SRS, requirements folder, external source, Figma URL, or existing codebase.
4. Run /sdlc-loop.
5. Answer blocking questions only when the kit halts for required human context or approval.
6. Review generated artifacts, deployment reports, and QA reports before accepting completion.

The high-level flow is:

/sdlc-init
  -> BA requirements ingestion and SRS sign-off
  -> UI/UX design extraction, mapping, creation, import, revision, or incorporation
  -> SA architecture, ADRs, integration adequacy, and instrumentation contracts
  -> TL task planning
  -> BE/FE implementation
  -> DevOps local deployment
  -> QA-Exec runtime verification
  -> iteration or done

Commands

/sdlc-init

Initializes the current project for the kit.

/sdlc-init

Optional flags:

/sdlc-init --target codex
/sdlc-init --target claude
/sdlc-init --target both
/sdlc-init --dry-run
/sdlc-init --force-hooks
/sdlc-init --skip-agents
/sdlc-init --skip-settings
/sdlc-init --skip-hooks

What it does:

• Injects the plugin guidance into the project's AGENTS.md inside a managed block by default.
• Warns about overlapping or conflicting headings.
• Skips .claude/settings.json and .claude/hooks for the default Codex target.
• With --target claude, writes CLAUDE.md, merges original Claude settings into the project, and mirrors hook scripts into .claude/hooks.
• With --target both, writes both AGENTS.md and CLAUDE.md, then applies Claude settings/hooks.

Run /sdlc-init again after plugin updates when you want the target project to receive updated kit guidance.

/sdlc-loop

Runs the SDLC orchestrator loop. It reads current artifacts, dispatches the next eligible role, enforces gates, and advances tasks through the plan when evidence exists.

/sdlc-loop

The loop is intentionally gate-driven. It should halt for unresolved SRS questions, missing design confirmation, external integration adequacy gaps, dependency approval, failed QA, malformed deploy reports, or unsafe operations.

Supported Tracks

Requirements Tracks