agentic_sdlc

Enterprise agentic SDLC. This repository is a Claude Code plugin marketplace that ships the agentic-sdlc-bootstrap skill.

What the skill does

agentic-sdlc-bootstrap sets up a complete, versioned and traceable agentic SDLC for any project, software or not (research programs, operations changes, launches, content or data pipelines, multi-workstream programs). It interviews you, proposes the artifact chain and its dependency graph, then scaffolds the document system with gates, agent roles and validation. Every process or flow is documented with an embedded Mermaid diagram. See skills/agentic-sdlc-bootstrap/SKILL.md.

Install

Claude Code (plugin marketplace, recommended)

One step (Claude Code 2.1.275 or later):

/plugin install agentic-sdlc --marketplace rafapc2/agentic_sdlc

Or in two steps:

/plugin marketplace add rafapc2/agentic_sdlc
/plugin install agentic-sdlc@agentic-sdlc-marketplace

From a shell, use claude plugin marketplace add ... and claude plugin install ... with the same arguments. Update later with /plugin marketplace update agentic-sdlc-marketplace.

Claude Code (manual copy)

git clone https://github.com/rafapc2/agentic_sdlc
cp -r agentic_sdlc/skills/agentic-sdlc-bootstrap ~/.claude/skills/        # all projects
# or: cp -r ... <project>/.claude/skills/                                  # one project

Claude.ai / Claude desktop

Zip the skill folder so that agentic-sdlc-bootstrap/SKILL.md is at the top level of the archive, then upload it under Settings, Capabilities, Skills. Organization admins can provision it for the whole team from the same place.

Use

Ask for what you need in your own words; the skill triggers on requests to set up an agentic SDLC or workflow. When installed as a plugin the explicit command is /agentic-sdlc:agentic-sdlc-bootstrap. Example first prompt:

I want to set up an agentic SDLC from scratch for moving our 120-truck fleet to electric: procurement, charging, driver training and budget. It is not software.

Day to day in a generated project:

python sdlc/_tools/new_doc.py sdlc spec "Title" --derived-from INT-0001@1.0.0   # new document
python sdlc/_tools/bump.py sdlc SPEC-0001 minor --note "Added R-4"              # new version + impact
python sdlc/_tools/validate_docs.py sdlc                                         # links, versions, diagrams, graph

Repository layout

.claude-plugin/
  plugin.json            plugin manifest (name: agentic-sdlc)
  marketplace.json       marketplace manifest (name: agentic-sdlc-marketplace)
skills/
  agentic-sdlc-bootstrap/
    SKILL.md             workflow and rules
    references/          profiles, artifact catalog, versioning, linking, governance, diagramming
    assets/templates/    document, CLAUDE.md, agent-role and hook templates
    scripts/             scaffold.py, validate_docs.py, bump.py, new_doc.py
    evals/               example prompts

Develop and test locally

claude plugin validate .                           # must print "Validation passed"
claude plugin marketplace add ./                   # use the working copy
claude plugin install agentic-sdlc@agentic-sdlc-marketplace

Release by bumping version in .claude-plugin/plugin.json; users get the new version on their next marketplace update.

Data and privacy

This plugin is a set of Markdown instructions, templates and local Python/Bash scripts. It does not run a server, does not call any network service, does not collect telemetry, and does not send any data anywhere. The scripts only read and write files inside the project folder you generate (including the optional .index/ reports), and the optional validation hook runs locally. It declares no MCP servers and no npx or uvx launchers, so there are no launcher versions to pin and no privacy-policy URL is required. Anything you type in the interview is handled by Claude under your own Claude plan's terms, not by this plugin.

License

MIT, see LICENSE.