Software Project Management Plugin (Wave 2)backlog_item

softwarefeatureagentspluginextension:software-kbwave-2
2 min read · Edit on Pyrite

Problem

Dev teams using AI agents (Claude Code, Codex, OpenClaw) lack structured project knowledge. Architecture decisions live in scattered docs, component relationships are undocumented, and agents can't natively collaborate on a project's knowledge graph. Existing project management tools (GitHub Issues, Jira) track tasks but not knowledge — they don't capture the "why" behind architecture, the relationships between components, or the editorial standards that govern the codebase.

Solution

Evolve the existing `extensions/software-kb/` into a full software project management plugin. The current extension already provides ADR, DesignDoc, Standard, Component, BacklogItem, and Runbook entry types with validators and CLI commands (`pyrite sw components`, `pyrite sw adrs`, `pyrite sw backlog`, `pyrite sw standards`, `pyrite sw new-adr`). The wave 2 work adds workflows, agent collaboration entity types, and task integration so agents can natively collaborate on a project.

What Exists (software-kb extension)

  • Entry types: ADR (7-status lifecycle), DesignDoc (5-status), Standard (category + enforced flag), Component (kind/path/owner/dependencies), BacklogItem (5-status workflow, priority, effort), Runbook (kind + audience)
  • Validators: ADR number uniqueness, status enum enforcement, component dependency resolution
  • CLI: `pyrite sw` subcommands for browsing and creating project artifacts
  • Workflows: Basic status transitions on ADRs and backlog items
  • Preset: `software` template for `pyrite init --template software`
  • What's New for Wave 2

    Enhanced workflows:

  • BacklogItem workflow integration with the task plugin — backlog items become assignable tasks with agent claim/checkpoint semantics
  • ADR approval workflow: draft → proposed → review → accepted with reviewer tracking
  • Sprint/iteration planning: group backlog items into time-boxed iterations
  • New entity types:

  • `sprint` — time-boxed iteration with goals, capacity, velocity tracking
  • `dependency_map` — cross-component dependency visualization data
  • `release` — release notes aggregated from completed backlog items and ADRs
  • Agent collaboration:

  • Agents can claim backlog items as tasks (via task plugin integration)
  • Agents can propose ADRs and submit for review
  • Agents can update component documentation when they modify code
  • Agents can generate release notes from completed items
  • MCP tools (beyond existing CLI):

  • `sw_claim_item` — agent claims a backlog item (atomic, no double-claims)
  • `sw_propose_adr` — create ADR in draft, auto-link to relevant components
  • `sw_update_component` — update component docs with dependency changes
  • `sw_sprint_status` — current sprint overview for agent context
  • Relationship to Existing Extension

    This is not a new plugin — it's the evolution of `extensions/software-kb/`. The existing entry types, validators, CLI commands, and preset remain. New features are additive. The `software` template continues to work. Backlog items should flow naturally into the backlog being tracked by this very project.

    Prerequisites

  • Task plugin phase 2 (atomic task_claim, task_decompose) for agent collaboration
  • Wave 1 platform shipped (0.8 alpha)
  • Success Criteria

  • Agent can claim a backlog item, implement it, update component docs, and mark it done — all via CLI or MCP
  • ADR lifecycle works end-to-end: agent proposes → human reviews → accepted/rejected
  • Sprint planning: create sprint, assign items, track velocity
  • `pyrite init --template software` produces a working project KB with all new types
  • Demo: "watch an agent collaborate on a software project" screencast
  • Launch Context

    This is the wave 2 plugin. Launches 1-2 weeks after the 0.8 platform alpha. Audience: dev teams, Claude Code/Codex users. Message: "Agents that understand your architecture." The existing software-kb extension gives us a massive head start — most of the schema and infrastructure already exists.