{
  "markdown": "# Agenture\n\n**Skills and a methodology for agentic software development - from vibe coding towards an autonomous agentic development.**\n\nThis repo is set of coding skills and plugins for Claude Code and Codex, installable and useable with `agn` command prefix.\nAgenture's north star is a software factory you run by setting direction and approving gates, not by writing every line. Today Agenture supports a spec-driven, review-gated lifecycle that takes work from requirements to validated code. \n\n \n## Quick start\n\nAdd the marketplace to Claude Code once, then install the plugin:\n\n```\n/plugin marketplace add AgentureHQ/agenture-loop\n/plugin install agn@agenture\n```\n\nRestart Claude Code (or run `/reload-plugins`) and the `/agn:*` skills become available.\n\nThen walk the lifecycle for whatever you're building. The smallest useful loop — a single change or bug:\n\n```\n/agn:define task              # write the requirements (WHAT/WHY)\n/agn:implement task <path>    # detailed design → code → tests\n/agn:validate task            # run the task's quality gates\n```\n\n## The workflow\n\n`agn` automates the **software development lifecycle**, not a ticket tracker. Work moves through the phases a disciplined engineering team uses, with a review gate between each:\n\n```\ndefine → design → plan → implement → validate   (+ maintenance, anytime)\n```\n\nThe contract is the same at every phase: **you define WHAT and WHY, the agent derives HOW, and a review gate stands between phases.** Specs and plans hold requirements and acceptance criteria — never implementation code. The agent generates implementation from the spec, not from a ticket title or chat history.\n\nEach phase has a skill. Most take a `<level>` so you run the same phase at any granularity.\n\n### 1. Define — *what & why*\n`/agn:define <level>` where `<level>` ∈ `product | epic | feature | task`\n- **product** → vision / spec / requirements in `docs/`\n- **epic** → an epic file + its linked feature files\n- **feature** → a feature file + its linked task files\n- **task** → a single task or bug ticket in `tasks/backlog/`\n\nProduces requirements (WHAT/WHY) — never implementation detail.\n\n### 2. Design — *architecture*\n`/agn:design <level>` where `<level>` ∈ `product | epic | feature`\n- **product** drafts `docs/architecture.md` in-session\n- **epic / feature** refine that unit's design in place\n\n### 3. Plan — *decomposition*\n`/agn:plan <level>` — revises how a unit breaks down: an epic into features, or a feature into tasks. Plan-only; writes no code.\n\n### 4. Implement — *code & tests*\n`/agn:implement <level> <id>`\n- **task** (takes a file path) → detailed design → code → tests; halts on an upstream design gap\n- **feature** (takes a slug) → runs every open task in order\n- **epic** (takes a slug) → runs every linked feature in order\n\n### 5. Validate — *quality gates*\n`/agn:validate <level>` where `<level>` ∈ `task | feature | epic | product` — runs the gates for that tier before it can close. Task-level gates run in the main session; feature / epic / product QA runs in a fresh context via the QA sub-agent.\n\n### Maintenance — *anytime*\n- `/agn:code-review` — read-only audit, emits backlog tasks\n- `/agn:code-commit` — staged, well-formed commit\n- `/agn:code-comment` — add explanatory comments\n- `/agn:docs-sync` — reconcile upstream docs after a unit closes\n\n### State is managed by `taskman.sh`\nSkills compose content in dialog with you, then hand off to `./scripts/taskman.sh` as the save step. Files move `backlog → active → done`; a feature can't close until all its tasks are `done`, an epic until all its features are `done`.\n\n## Common paths\n\nMatch the entry point to where you are. Pick the smallest one that fits the scope.\n\n**Greenfield product** — start at Define and walk the full cycle.\n```\n/agn:define product           # vision + spec + requirements → docs/\n/agn:design product           # architecture → docs/architecture.md\n/agn:define epic              # design + plan for an epic (Planner sub-agent)\n/agn:implement epic <slug>    # iterate features, stop per task for review\n/agn:validate product         # full-system QA via the QA sub-agent\n```\n\n**Incremental feature** — docs already exist; enter at Define.\n```\n/agn:define feature           # design + plan for the new feature\n/agn:plan feature             # revise the task breakdown (optional)\n/agn:implement feature <slug> # each task in order\n/agn:validate feature         # integration tests via the QA sub-agent\n/agn:docs-sync                # reconcile docs after close\n```\n\n**Single change or bug** — smallest work item.\n```\n/agn:define task              # task or bug — requirements only\n/agn:implement task <path>    # detailed design → code → tests; halts on upstream design gap\n/agn:validate task            # task-level quality gates (main session)\n```\n\n**Optimization** — surface debt, then schedule it.\n```\n/agn:code-review              # read-only audit → backlog tasks\n/agn:define feature           # group the findings worth doing\n/agn:implement feature <slug>\n/agn:validate feature\n```\n\nSee [plugins/agn/README.md](plugins/agn/README.md) for the full skill reference and the `taskman.sh` lifecycle CLI, and [docs/agn-specification.md](docs/specification.md) for the normative methodology and plugin specification.\n\n## Work units: product, epic, feature, task\n\nThe lifecycle above runs at four granularities. Pick the smallest one that fits the scope — the phases are the structure; sizing is a convenience.\n\n| Size | Use when | File |\n|------|----------|------|\n| **Product** | The whole system — vision, spec, architecture | `docs/` |\n| **Epic** | A functional block spanning multiple features | `tasks/epics/` |\n| **Feature** | One coherent unit of work, usually one branch | `tasks/features/` |\n| **Task** | A single implementation step or bug fix | `tasks/backlog/` → `active/` → `done/` |\n\nA product holds many epics, an epic many features, a feature many tasks — but the hierarchy is open, not fixed: every level is optional one up. A feature can have no parent epic; a task can be ad-hoc with no parent feature or epic.\n\n```mermaid\ngraph TD\n    P[Product]\n\n    P --> E1[Epic A]\n    P --> E2[Epic B]\n    P --> F3[Feature C]\n    P --> T7[Task]\n\n    E1 --> F1[Feature A1]\n    E1 --> F2[Feature A2]\n    E2 --> F4[Feature B1]\n\n    F1 --> T1[Task]\n    F1 --> T2[Task]\n    F2 --> T3[Task]\n    F4 --> T4[Task]\n    F4 --> T5[Task]\n    F3 --> T6[Task]\n```\n\n**Feature C** has no parent epic, and the top-level **Task** is ad-hoc.\n\n## Available plugins\n\n| Plugin | What it does | Docs |\n|--------|--------------|------|\n| `agn` | Agentic SDLC loop — define, design, plan, implement, and validate through structured `/agn:*` skills with built-in review gates | [plugins/agn/README.md](plugins/agn/README.md) |\n\n## Repository layout\n\n```\nagenture-loop/\n├── .claude-plugin/\n│   └── marketplace.json          # marketplace manifest\n├── plugins/\n│   └── agn/                      # the agentic-SDLC plugin\n│       ├── .claude-plugin/plugin.json\n│       ├── skills/\n│       ├── rules/\n│       ├── scripts/\n│       └── README.md\n├── docs/                         # product docs for the marketplace\n├── tasks/                        # this repo's own SDLC tracking (dogfoods agn)\n├── LICENSE\n├── PRIVACY.md\n└── README.md\n```\n\n## Adding a new plugin to the marketplace\n\n1. Create `plugins/<your-plugin>/` with its own `.claude-plugin/plugin.json` and any of `skills/`, `agents/`, `commands/`, `hooks/`, `.mcp.json`. All paths must be self-contained inside the plugin directory.\n2. Add a new entry to the `plugins` array in `.claude-plugin/marketplace.json`.\n3. Update the **Available plugins** table above.\n\nSee [Claude Code's plugin marketplace docs](https://code.claude.com/docs/en/plugin-marketplaces.md) for the full schema.\n\n## Developing and testing plugins locally\n\nThis section is for contributors who develop the marketplace itself. End users should follow [Quick start](#quick-start) above.\n\n### Test an unpublished change from another project\n\nLocal changes are not on GitHub yet, so install the marketplace from your local checkout instead of the GitHub shorthand. Use the **absolute path** — a relative `./` only resolves when Claude Code runs inside this repo.\n\n```\n/plugin marketplace add /absolute/path/to/agenture-loop\n/plugin install agn@agenture\n```\n\nThen run `/reload-plugins` (or restart Claude Code) so the `/agn:*` skills load.\n\n### Refresh after editing the plugin\n\nA locally added marketplace is cached. After changing `marketplace.json`, a `plugin.json`, or any skill, refresh the cache:\n\n```\n/plugin marketplace update agenture     # re-read this marketplace\n/reload-plugins                          # reload skills into the session\n```\n\nIf an entry is broken or stale (for example, a failed earlier `add`), remove and re-add it:\n\n```\n/plugin marketplace remove agenture\n/plugin marketplace add /absolute/path/to/agenture-loop\n```\n\n### Dogfooding inside this repo\n\nWhen you run Claude Code **inside this repo**, the `agn` skills load automatically without `/plugin install` — `.claude/` symlinks point at the plugin sources:\n\n```\n.claude/skills -> plugins/agn/skills\n.claude/rules  -> plugins/agn/rules\n```\n\n### Publish\n\nA local install pins the marketplace to your machine's path and resolves only for you. For the [Quick start](#quick-start) command (`AgentureHQ/agenture-loop`) to work for everyone, push your commits to `origin/main` so GitHub serves the updated `.claude-plugin/marketplace.json`.\n\n## License\n\n[Apache 2.0](LICENSE). See [PRIVACY.md](PRIVACY.md) for the privacy policy.\n",
  "bytes": 9569,
  "sha": "d944df7848e74508fd9cc783973c49c10a6c5f2528bc696ec034053ff71011d5",
  "repo_slug": "vkroz/spec-to-code",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_vkroz_spec_to_code_spec_to_code_ee4eb822/readme"
}