Skip to content

Repository files navigation

Agentic SDLC

A public catalog of Claude Code skills, agents, commands, workflow templates, and helper automation for building an end-to-end agentic software delivery lifecycle. The repo is intentionally template-first: private project material has been removed, placeholders are explicit, and the shipped defaults bias toward reviewable, opt-in automation.

Quick Stats

Category Count Description
Skills 24 system + 6 project (in 2 starter packs) Trigger-matched capabilities (deploy, review, triage, design, orchestration)
Agents 4 Specialized agent definitions (iOS, GCP, dead code)
Commands 24 User-invoked slash commands (agent-os, ops, ios, design-os)
Workflows 16 GitHub Actions (CI/CD, testing, ops, AI automation)
Workflow Helpers 10 Companion .github/actions, .github/scripts, and drift checks for the workflow templates
Patterns 5 Documented reusable patterns
Scripts 1 Standalone developer tooling (tmux orchestrator)
Plugins 25+ Claude Code marketplace plugins
MCP Servers 2 Tool integrations (Docker recommended, Xcode)

Architecture Overview

flowchart LR
    subgraph INPUT["Input"]
        IDEA["Idea"]
        BUG["Bug"]
    end

    subgraph PLAN["Plan"]
        TR["/triage"]
        PRD["/prd-taskmaster"]
        SPEC["/shape-spec"]
    end

    subgraph BUILD["Build"]
        FEAT["/feature-dev"]
        AGENTS["Agents"]
        SIMP["/simplify"]
    end

    subgraph REVIEW["Review"]
        CODEX["4-Agent\nAdversarial\nReview"]
    end

    subgraph SHIP["Ship"]
        SELF["Self-Healing\nDeploy\n(7 phases)"]
        PROD["Production"]
    end

    IDEA --> TR --> PRD --> SPEC
    BUG --> TR
    SPEC --> FEAT --> AGENTS --> SIMP --> CODEX --> SELF --> PROD

    style INPUT fill:#e1f5fe
    style PLAN fill:#f3e5f5
    style BUILD fill:#e8f5e9
    style REVIEW fill:#fff3e0
    style SHIP fill:#fce4ec
Loading

Catalog

Skills (skills/)

System-level skills available across all projects:

Skill Category What It Does
self-healing-deploy Deploy 7-phase autonomous merge-to-production pipeline
prd-to-pr Pipeline End-to-end PRD to merged PR orchestration
bug-to-pr Pipeline End-to-end bug report to fix PR pipeline
pr-handoff-to-codex Review 4-agent adversarial Codex review
prd-taskmaster Planning PRD generation and GitHub issue publishing
triage Planning Batch work item classification and routing
senior-architect Architecture Multi-platform system architecture design
gh-fix-ci CI/CD GitHub Actions failure debugging
code-simplifier Quality Code clarity and consistency review
dead-code-cleanup Quality Swift dead code removal in isolated worktree
ios-design iOS SwiftUI design system compliance
figma Design Figma-to-code translation via MCP
find-skills Discovery Skill discovery and installation
workflow-orchestrator Orchestration Coordinates multiple pipelines in parallel when the task warrants it

Full skills catalog

Agents (agents/)

Agent Domain Description
ios-architect iOS Senior iOS architect (Swift 6.0+, SwiftUI, MVVM)
ios-developer iOS Day-to-day Swift/SwiftUI implementation
dead-code-agent iOS Dead code detection and removal
gcp-platform-engineer GCP Cloud Run, Cloud SQL, Terraform infrastructure

Full agents catalog

Commands (commands/)

Suite Commands Purpose
agent-os 5 commands Standards-driven development framework
ops 7 commands Cost, performance, and security management
ios 2 commands iOS build and design
design-os 10 commands Product design lifecycle (vision to export)

Full commands catalog

Design-OS is included because I think the approach is promising, but it is the least battle-tested part of this repo in my own workflow so far.

Workflows (workflows/)

16 GitHub Actions workflows organized by function. The repo also includes the companion .github/actions/, .github/scripts/, and Infrastructure/scripts/ helpers referenced by those templates so adopters are not left with broken local-action paths.

Category Workflows Purpose
ci-cd 6 Build, deploy, promote (dev -> staging -> prod)
testing 4 Backend, iOS, UI, performance tests
ops 4 DB migrations, artifact retention, ops review
ai 2 Claude bot, AI config hotfix

Full CI/CD documentation

Key Patterns

4 Codex agents independently review each PR (architecture, security, quality), then an orchestrator synthesizes and autonomously fixes issues.

7-phase pipeline that merges, monitors CI, self-heals failures, runs migrations, applies terraform, and promotes through environments.

8-step orchestration from PRD to production: generate -> spec -> counsel -> implement -> simplify -> review -> fix -> deploy.

5-command framework that discovers, indexes, and injects project standards into agent context before implementation.

10-command product design pipeline from vision to export, enforcing sequential design methodology.

This is still an exploratory part of the framework for me. I kept it in the repo because it feels promising, but I have not used it nearly as heavily as the engineering and delivery paths yet.

Scripts (scripts/)

Standalone developer tooling that complements the skills and commands.

Script What It Does
orchestrate Tmux-based orchestrator that runs full Claude Code skill pipelines (/bug-to-pr, /prd-to-pr, /triage) in parallel, each in its own git worktree with complete skill access. Solves the sub-agent Skill tool limitation by spawning full interactive CLI sessions. 3-column tmux layout with health monitoring, cost guardrails, queue management, and session recovery.

Documentation

Public Hardening

This public version has been scrubbed and hardened for reuse:

  • Embedded secrets and private redemption tooling were removed.
  • Nested VCS metadata and absolute-path symlinks were removed and ignored.
  • The sample Claude Code settings no longer auto-run shell hooks.
  • Third-party GitHub Actions are pinned to immutable SHAs.
  • Cloud SQL Proxy downloads are checksum-verified.
  • The Claude workflow is limited to trusted collaborators instead of arbitrary public commenters.

Security Note

The configs/settings/monorepo-fullstack.json example is a safer baseline for repository analysis and controlled edits. It keeps core Claude file permissions enabled, removes auto-hooks, and only allows a narrow set of local shell reads by default:

  • No auto-enabled shell hooks - hooks is empty by default
  • No network download commands - curl, wget, ssh, scp, rsync, and nc are explicitly denied
  • No wildcard shell execution - the example only permits local inspection commands plus git

Review and tighten this file to match your own security posture before adoption.

Workflow files use <YOUR_*> placeholders and ${{ secrets.YOUR_* }} names. Replace them with your own project identifiers, secret names, and service accounts before enabling any workflow in production.

Getting Started

Adopting Individual Pieces

Each directory is self-contained. To adopt a skill, agent, or command:

  1. Copy the directory to your project's .claude/ folder.
  2. Adjust any project-specific references and placeholders.
  3. Restart Claude Code to pick up the new configuration.

Adopting Workflow Templates

These workflows are shipped as a catalog under workflows/, not as live .github/workflows/ files. To use them in your own repo:

  1. Copy the workflows you want from workflows/** into .github/workflows/.
  2. Copy the helper directories from this repo:
    • .github/actions/
    • .github/scripts/
    • Infrastructure/scripts/
  3. Replace every <YOUR_*> placeholder with repo-specific values.
  4. Configure the required GitHub Environment secrets and approvals.
  5. Dry-run the workflows in a non-production repo before trusting them with real deploy credentials.

Adopting the Full Pipeline

To replicate the complete agentic SDLC:

  1. Install Superpowers by Jesse Vincent from the obra/superpowers-marketplace if it fits your workflow.
  2. Copy skills/system/ to ~/.agentConfig/skills/.
  3. Copy relevant skills/project/ directories into your repo-scoped .claude/skills/.
  4. Copy relevant agents/ to your project's .claude/agents/.
  5. Copy relevant commands/ to your project's .claude/commands/.
  6. Start from configs/settings/monorepo-fullstack.json, then tighten it further.
  7. Treat hooks/session-init.sh as an opt-in template and only enable it after review.
  8. Adapt the workflow templates and helper actions/scripts for your CI/CD environment.
  9. Configure MCP servers per docs/package-inventory.md.

Origin

This catalog was extracted from real production usage across a mobile + cloud monorepo and then generalized into reusable templates. The goal is to preserve realistic patterns without shipping private infrastructure details, private repo references, embedded secrets, or non-portable local paths.

Upstream Attribution

  • Agent OS naming and standards-driven workflow attribution belongs to Brian Casel / Builder Methods. This repo includes an adapted public implementation of that style of workflow, not an official distribution. See docs/upstream-attribution.md.
  • Design OS naming and design-process attribution belongs to Brian Casel / Builder Methods. This repo keeps an adapted version because the approach looks promising, but it is still one of the least battle-tested parts of my own setup.
  • Superpowers is by Jesse Vincent and is distributed through obra/superpowers and the obra/superpowers-marketplace.
  • The bundled prd-taskmaster skill is sourced from anombyte93/prd-taskmaster and references Task Master / task-master-ai for downstream task orchestration.

Contributing

See CONTRIBUTING.md for guidelines on adding new skills, agents, commands, or patterns.

License

MIT

About

Public agentic SDLC framework for Claude Code skills, commands, workflows, and helper automation

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages