claude-sdlc-agents

SDLC agent pack for Plane Conductor — 10 pipeline agents + 2 meta-agents, grounded in industry methodologies (BABOK, C4, ISTQB, OWASP, WCAG).

A starter pack of specialised Claude Code agents wired to follow standard software-engineering methodologies, designed to drop into Plane Conductor as a full SDLC pipeline triggered by Plane mentions.

Project-specific details (stack, conventions, multi-tenancy rules, verification commands, frontends inventory) are kept outside the prompts — in your own project's KB, populated from the template/ directory in this repo.

---

What's inside

agents/ — generic agent definitions

SDLC pipeline (10) — triggered by Plane @-mentions

File (also the name: for claude --agent <name>)	Role	Methodology
business-analyst.md	Business Analyst	BABOK v3 (4 requirement types, MoSCoW, INVEST, 5 Whys)
system-analyst.md	System Analyst	C4 + DDD bounded contexts + REST + IEEE 29148 + ADR
architect.md	Architect	SOLID + DDD + ADR governance + 6-area review checklist
designer.md	UX/UI Designer	WCAG 2.1 AA + Nielsen heuristics + Material/HIG
django-developer.md	Django backend developer	project-conventions-driven
vue-developer.md	Vue developer (Vue 2 / Vue 3)	project-conventions-driven
react-developer.md	React / Next.js developer	project-conventions-driven
api-tester.md	API Tester	ISTQB Foundation (EP, BVA, decision tables, state transition, use-case)
ui-tester.md	UX/E2E Tester	ISTQB + WCAG 2.1 AA
reviewer.md	Final Reviewer	OWASP Top 10 + SOLID + Google review checklist

Meta agents (2) — invoked directly by the user, not by Plane

File	name: (alias)	What it does
prompt-architect.md	zuse	Designs and audits agent prompts, skills, and KB reference docs. Plan-then-edit; runs a 9-block audit on every prompt before changing it.
project-manager.md	tron	Personal PM. Triages incoming work into "fix it myself" / "file in Plane and run the pipeline" / "clarify first". Reads GitHub / GitLab / kubectl / helm freely; every state change requires explicit approval.

Each file is a Claude Code agent definition with YAML frontmatter (name, description, model, tools) and a long-form prompt body. SDLC prompts read project-specific KB files (see below) and the matching skill before composing artifacts. Meta-agent prompts have no project-specific reads — they work in any repo.

skills/ — 9 reusable Claude Code skills

Skill	Used by	Encodes
babok-elicitation	business-analyst	BABOK v3 elicitation + analysis knowledge areas
system-design-techniques	system-analyst	C4, DDD bounded contexts, REST conventions, ADR pattern
architecture-review-framework	architect	6-area review checklist, SOLID + DDD lenses, ADR governance
ux-design-discipline	designer	WCAG 2.1 AA, Nielsen heuristics, platform conventions
documentation-discipline	coders	docstrings, README updates, ADR status, migration notes
istqb-test-design	api-tester, ui-tester	EP / BVA / decision tables / state transition / use case
code-review-discipline	reviewer	OWASP Top 10, SOLID, Google review checklist, cross-trace
artifact-templates	every agent	canonical Markdown templates for every artifact type
plane-operations	every agent	the Plane read/write protocol (pickup, sub-issue, comments, mentions, re-entry)

template/ — KB template you copy into your project

Project-specific facts go in your own repo — versioned alongside your code. The pack ships an empty template:

template/
├── AGENTS.md                  # entry point: index, routing table, project rules at a glance
├── PROJECT_DESCRIPTION.md     # draft of the Plane project description (operational layer)
└── kb/
    ├── stack.md         # languages, frameworks, libs, versions
    ├── conventions.md   # lint, types, naming, TDD policy
    ├── architecture.md  # services / bounded contexts / import contracts
    ├── verify.md        # slash-commands / make / npm scripts for lint, tests, builds
    ├── multitenancy.md  # tenant isolation rules (or "N/A")
    ├── migrate.md       # DB migration discipline
    ├── frontends.md     # per-frontend stacks, build commands
    ├── document.md      # docstring style, doc-generation tool
    └── domain/          # ad-hoc domain knowledge files

You copy template/ into your repo, fill it in, commit. The agents read it via $KB_DIR env var (set by Plane Conductor) or fall back to <cwd>/AGENTS.md.

This follows the agents.md emerging convention (Cursor / OpenAI Codex / Aider / Claude Code).

---

How it fits with Plane Conductor

Plane Conductor is the orchestrator: it receives a Plane webhook on @<nickname> mention, resolves the nickname, and spawns claude --agent <prompt-role> --print locally. What that agent actually does is defined by the prompt + skills here.

  human posts "@castor draft requirements" in Plane
            │
            ▼
  Plane Conductor (orchestrator)
  │  —  resolves bot user `castor` → prompt_role `business-analyst`
  │  —  sets env: KB_DIR, AGENT_NICKNAME=castor, AGENT_MEMBER_ID, ...
  │
  ▼ spawns
  claude --agent business-analyst --print
  │
  ▼ loads
  prompts/business-analyst.md
  + babok-elicitation skill + plane-operations skill + artifact-templates
  + reads $KB_DIR/AGENTS.md and relevant kb/*.md
  │
  ▼ writes
  structured REQUIREMENTS into the root issue's description, posts comments

Nicknames (castor, sark, flynn, ...) live in your Plane workspace and conductor.yaml config; the prompts are nickname-agnostic. The Greeting line uses the AGENT_NICKNAME env var if Plane Conductor sets it, falling back to the role name.

---

Install

1. Install the plugin

The pack ships as a Claude Code plugin via marketplace.

claude plugin marketplace add volodchenkov/claude-sdlc-agents
claude plugin install sdlc-agents@sdlc-agents-marketplace

This makes 12 agents (10 SDLC pipeline + 2 meta) and 9 skills available globally — claude --agent business-analyst, system-analyst, tron, zuse, …. Updates are pulled from the github default branch on each Claude Code session start; force a refresh with /plugin update.

For local-development install pointing at a checkout:

claude plugin marketplace add /path/to/claude-sdlc-agents
claude plugin install sdlc-agents@sdlc-agents-marketplace

The same commands are available as slash-commands inside an interactive Claude Code session (/plugin marketplace add …, /plugin install …).

Optional: dev-tooling bootstrap

setup.sh at the repo root is a one-shot helper for new machines: it checks for gh / glab / kubectl / helm / jq / helm-diff and offers to install whichever is missing, verifies the plane-tower MCP is registered, and (with consent) merges a read-only Bash allowlist from docs/settings-allowlist.json into ~/.claude/settings.json. It does not install agents — that's the marketplace's job above.

2. Drop the KB template into your project

cp ~/Projects/claude-sdlc-agents/template/AGENTS.md  /your/project/
cp -r ~/Projects/claude-sdlc-agents/template/kb      /your/project/

Fill in /your/project/AGENTS.md and /your/project/kb/*.md (every file has a # Fill in: block describing what to write).

2b. Set up the Plane project description (optional but recommended)

Agents read a short operational one-pager from your Plane project's Settings → Description at session start (repo URL, staging, initiator, pipeline notes). It is complementary to AGENTS.md (technical truth, in repo) — never duplicated.

Use the version-controlled draft as a starting point, then paste the body into Plane:

cat ~/Projects/claude-sdlc-agents/template/PROJECT_DESCRIPTION.md

If the description is empty, agents skip this layer with no STOP.

3. Wire up Plane Conductor

In your Plane Conductor workspace YAML (conductor.d/<slug>.yaml):

prompts_dir: ~/.claude/plugins/.../sdlc-agents/agents   # or your project's .claude/agents
kb_dir:      /your/project                              # AGENTS.md + kb/ live here

In conductor.yaml reference the prompt-role names from prompts/:

agents:
  - { nickname: castor,  prompt_role: business-analyst }
  - { nickname: sark,    prompt_role: system-analyst }
  - { nickname: flynn,   prompt_role: architect }
  - { nickname: quorra,  prompt_role: designer }
  - { nickname: rinzler, prompt_role: django-developer }
  - { nickname: ram,     prompt_role: vue-developer }
  - { nickname: beck,    prompt_role: react-developer }
  - { nickname: yori,    prompt_role: api-tester }
  - { nickname: gem,     prompt_role: ui-tester }
  - { nickname: dumont,  prompt_role: reviewer }

(Pick your own nicknames — above is illustrative.)

---

Adapting to your own pipeline

• Different roles? Edit / add prompt files; update your conductor.yaml roster. Methodology skills are independent — keep some, drop others, add new ones.
• Different methodologies? Replace the skill content. Prompts reference skills by name, so as long as you keep the slug, the agent picks up your version.
• Different stack? Fill in kb/stack.md, kb/conventions.md, kb/frontends.md, kb/migrate.md to match. Coder prompts are conventions-agnostic; they read your KB.
• No multi-tenancy? Set kb/multitenancy.md to "N/A — single-tenant". Agents skip multitenancy checks entirely.
• No service boundaries? Set kb/architecture.md to "N/A — monolithic single-app". Skip imports / DDD framework checks.

---

Status

v0.2 — packaged as a Claude Code plugin (marketplace.json / plugin.json), 10 agents + 9 skills, KB template + Plane project description template. Working in production for one team. Prompts will evolve as patterns settle.

PRs welcome — especially if you adapt the pack to a different methodology (DSDM, SAFe, FDD, …) or different stack.

---

Optional companion skills

This pack stays stack-agnostic on purpose. For Django projects, the django-developer agent references four extra skills that are NOT shipped here — they're authored separately and remain optional:

• django-models — fat-model / thin-view patterns, QuerySet composition
• celery-patterns — task design, retries, idempotency
• pytest-django-patterns — fixtures, factories, pytest-django
• systematic-debugging — 4-phase root-cause methodology

Recommended source: kjnez/claude-code-django (MIT).

The django-developer agent works without them — it falls back to first principles + your $KB_DIR/kb/ rules. With them installed, it loads them automatically when designing models, writing Celery tasks, writing pytest, or debugging.

The agents.md convention is from agents.md — one entry-point file at the repo root, kb/ for details.

---

License

MIT © 2026 Dmitry Volodchenkov