Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
Agenture avatar

Agn

  • 2 repo stars
  • Updated June 9, 2026
  • AgentureHQ/agenture-loop

Agentic SDLC for Claude Code. Automates spec, design, planning, implementation, and QA via /agn:* skills with built-in review gates and document consistency.

About

agn is a Claude Code skill in the Design & UI/UX category. Agentic SDLC for Claude Code. Automates spec, design, planning, implementation, and QA via /agn:* skills with built-in review gates and document consistency.

  • agn
  • Design & UI/UX
  • AI-coding skill

Agn by the numbers

  • Data as of Jul 21, 2026 (Skillselion catalog sync)
/plugin marketplace add AgentureHQ/agenture-loop
/plugin install agn@agenture

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
repo stars2
Last updatedJune 9, 2026
RepositoryAgentureHQ/agenture-loop

What it does

Agentic SDLC for Claude Code. Automates spec, design, planning, implementation, and QA via /agn:* skills with built-in review gates and document consistency.

README.md

agn — Agentic SDLC

Agentic software-delivery loop for Claude Code. Claude facilitates writing specs and designing architecture, then autonomously builds implementation plans, implements, and tests — all through structured /agn:* skills with built-in review gates and document consistency.

Install

From the agenture marketplace:

/plugin marketplace add AgentureHQ/agenture-loop
/plugin install agn@agenture

Work-unit hierarchy

The plugin supports a four-tier hierarchy. Each tier is optional one level up — a task may have no feature, and a feature may have no epic.

product → epic → feature → task
Tier Purpose File location
Product Vision, specification, requirements docs/vision.md, docs/spec.md, docs/requirements.md
Epic Functional block larger than a feature; decomposes into features tasks/epics/YYYYMMDD_<slug>.md
Feature Named product initiative; produces a plan and child tasks tasks/features/YYYYMMDD_<slug>.md
Task Unit of implementation work tasks/<state>/YYYYMMDD_<slug>.md (where state is backlog, active, or done)

Bugs are tasks with kind: bug. Same lifecycle. Can be attached to a feature or ad-hoc.

See rules/task-composition.md for the body and frontmatter model; see ./scripts/taskman.sh help for the persistence model (storage, naming, lifecycle).

Skills

The plugin ships 8 skills. Lifecycle skills follow the verb-noun pattern /agn:<verb> <level>; tool skills retain their action-named form.

SDLC workflow (lifecycle skills)

Verb Skill What it does
Define /agn:define <product|epic|feature|task> Define a work unit at the named tier — vision/spec/requirements (product), epic + linked features (epic), feature + linked tasks (feature), task/bug ticket (task). Delegates composition to the Planner sub-agent for epic/feature/task levels.
Design /agn:design <product|epic|feature> Focused revision of an existing unit's design; product produces docs/architecture.md. Epic and feature delegate to the Planner sub-agent for in-place body refinement.
Plan /agn:plan <epic|feature> Focused revision of an existing unit's decomposition (epic into features, feature into tasks). Delegates to the Planner sub-agent in refine + plan-only mode.
Implement /agn:implement <task|feature|epic> Execute implementation; task = detailed design → code → tests; feature/epic = iterate children with review gates
Validate /agn:validate <task|feature|epic|product> Quality gates. Task runs the task's own ## Quality gates in the main session. Feature, epic, and product delegate to the QA sub-agent for fresh-context validation against spec.

Code and maintenance

Skill What it does
/agn:code-review Read-only codebase audit; produces backlog tasks
/agn:code-comment Add explanatory comments to source code
/agn:code-commit Stage files and write a well-formed git commit message
/agn:docs-sync Process the doc-sync queue: review upstream docs for drift against recently closed units, propose diffs, apply after user approval, clear processed entries

Workflows supported

  • New product/agn:define product/agn:design product/agn:define epic (or /agn:define feature if no epic tier needed) → /agn:implement epic / /agn:implement feature/agn:validate product
  • Bug fix/agn:define task bug/agn:implement task/agn:validate feature/agn:validate product
  • Ad-hoc maintenance/agn:define task/agn:implement task/agn:validate feature
  • Incremental feature/agn:define feature (against existing product docs) → /agn:implement feature/agn:validate feature
  • Codebase optimization/agn:code-review/agn:define feature (or /agn:define epic for larger scope) → /agn:implement feature/agn:validate product

taskman.sh — task and feature CLI

All create / move / close / list operations on epics, features, and tasks go through scripts/taskman.sh. Skills compose content in dialog with the user, then hand off to taskman as the save step. Do not write task files directly.

./scripts/taskman.sh help                                # full surface
./scripts/taskman.sh new epic    --slug <s> --title T < body
./scripts/taskman.sh new feature --slug <s> --title T [--epic <s>] < body
./scripts/taskman.sh new task    --title T [--feature <s>] [--kind task|bug] < body
./scripts/taskman.sh finalize <path>                     # clear draft: true
./scripts/taskman.sh discard  <path>                     # delete a draft
./scripts/taskman.sh move <task-path> <backlog|active|done>
./scripts/taskman.sh list epics    [--status backlog|active|done]
./scripts/taskman.sh list features [--epic <s>] [--status backlog|active|done]
./scripts/taskman.sh list tasks    [--feature <s>] [--status backlog|active|done] [--kind task|bug]
./scripts/taskman.sh epic show    <slug>
./scripts/taskman.sh epic close   <slug>                 # fails unless all member features in done
./scripts/taskman.sh feature show  <slug>
./scripts/taskman.sh feature close <slug>                # fails unless all member tasks in done
./scripts/taskman.sh validate

epic close and feature close enforce the lifecycle rule: an epic can only close when every feature with matching epic: <slug> is in done, and a feature can only close when every task with matching feature: <slug> is in done.

Project layout created by workflows

docs/vision.md
docs/spec.md
docs/requirements.md
docs/architecture.md
docs/<area>/.../-spec.md         # feature-scoped specs
tasks/
  epics/                          # YYYYMMDD_<slug>.md
  features/                       # YYYYMMDD_<slug>.md
  backlog/                        # YYYYMMDD[_NN]_<slug>.md
  active/
  done/
  gaps/                           # YYYYMMDD-HHMMSS_<task-slug>.md — design-gap escalation log
  docs-sync-queue.txt             # PostClose hook queue; processed by /agn:docs-sync

Created incrementally as you run each stage — no upfront scaffolding.

Loading rules into your project

Plugins do not auto-load rules/. To activate the behavioral rules in your project, append this block to your project's CLAUDE.md:

@~/.claude/plugins/cache/agenture/agn/0.1.0/rules/first-principles.md
@~/.claude/plugins/cache/agenture/agn/0.1.0/rules/task-composition.md
@~/.claude/plugins/cache/agenture/agn/0.1.0/rules/writing-guideline.md
@~/.claude/plugins/cache/agenture/agn/0.1.0/rules/qa.md
@~/.claude/plugins/cache/agenture/agn/0.1.0/rules/doc-maintenance.md

Update the version (0.1.0) when you upgrade the plugin.

The five files:

  • first-principles.md — design discipline (YAGNI, KISS, DRY), surgical changes, goal-driven execution. Always-on.
  • task-composition.md — epic/feature/task frontmatter shapes, body section requirements, completion-summary template. Loaded by any skill that composes work units.
  • writing-guideline.md — crisp, no-fluff prose style. Loaded by document-writing skills.
  • qa.md — QA mindset, role separation, validation principles. Loaded by the QA sub-agent and /agn:validate skills.
  • doc-maintenance.md — what to check after a work unit closes. Loaded by the docs-sync skill / PostClose hook.

Persistence rules (storage layout, naming, lifecycle preconditions, CLI surface, validation behavior) live in ./scripts/taskman.sh help — the script is the single writer and the authoritative reference.

PostClose hook and docs-sync

Every successful close action — taskman.sh move <path> done, feature close <slug>, epic close <slug> — appends an entry to tasks/docs-sync-queue.txt and prints a "Doc-sync pending" hint. Run /agn:docs-sync to process the queue: it walks the dependency chain (vision → spec → requirements → architecture → linked spec) per rules/doc-maintenance.md, proposes diffs, applies after user approval, and clears each processed entry.

The queue is durable on disk. taskman.sh list prints a WARN hint whenever the queue is non-empty, so you see pending entries during normal task-state checks.

Optional: surface the queue at session start

For automatic surfacing of pending entries when a new Claude Code session opens in this project, add a SessionStart hook to your project's .claude/settings.json (or your user-level ~/.claude/settings.json):

{
  "hooks": {
    "SessionStart": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "test -s tasks/docs-sync-queue.txt && printf 'Doc-sync pending: %s entries in tasks/docs-sync-queue.txt. Run /agn:docs-sync.\\n' \"$(wc -l < tasks/docs-sync-queue.txt | tr -d ' ')\" || true"
          }
        ]
      }
    ]
  }
}

The hook is opt-in — the plugin does not install it for you. Without it, the queue still persists; you just see the hint via taskman.sh list rather than at session start.

Plugin layout

plugins/agn/
├── .claude-plugin/plugin.json    # plugin manifest
├── skills/                       # 9 /agn:* skills
├── agents/                       # planner (Design + Plan); qa (fresh-context validator)
├── rules/                        # first-principles, task-composition, writing-guideline, qa, doc-maintenance
├── scripts/taskman.sh            # task lifecycle CLI; PostClose hook emits to docs-sync-queue.txt
└── README.md

Related skills

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.