Base Coat — Project Goals¶
Mission¶
Base Coat exists to make GitHub Copilot useful at enterprise scale by providing a curated, governed, and composable library of AI customization assets that teams adopt across repositories through a single sync command.
Primary Goals¶
1. Full SDLC Agent Coverage¶
Provide specialized AI agents for every phase of the software development lifecycle — not just code generation. Base Coat covers architecture, coding, testing, security, DevOps, process management, and meta-tooling (agent design, prompt engineering, MCP).
Current state: 50 agents across 6 disciplines (v2.1.0):
| Discipline | Agents | Examples |
|---|---|---|
| 🔨 Development | 4 | backend-dev, frontend-dev, middleware-dev, data-tier |
| 🏗️ Architecture | 5 | solution-architect, api-designer, ux-designer, app-inventory, legacy-modernization |
| 🔍 Quality | 10 | code-review, security-analyst, guardrail, performance-analyst, chaos-engineer |
| 🚀 DevOps | 4 | devops-engineer, agentops, release-manager, self-healing-ci |
| 📋 Process | 6 | sprint-planner, product-manager, issue-triage, retro-facilitator, sprint-retrospective |
| 🧰 Meta | 6+ | agent-designer, prompt-engineer, mcp-developer, tech-writer, memory-curator |
2. One Entry Point, Zero Memorization¶
The /basecoat router skill provides a single entry point that routes to any of
the 50 agents. Users say /basecoat backend build a REST API and the router resolves
the right agent, attaches paired skills, and ensures ambient instructions are active.
Design philosophy: Users should never need to memorize agent names. Discovery mode
(/basecoat) shows a categorized catalog; delegation mode (/basecoat [discipline] [prompt])
routes directly.
3. Composable Three-Primitive Architecture¶
Base Coat separates concerns into three primitives that compose cleanly:
- Agents define who does the work and how (workflow, persona, model)
- Skills provide what knowledge they use (templates, checklists, decision trees)
- Instructions enforce what rules everyone follows (ambient, cross-cutting)
This separation means a new security policy updates one instruction file and every agent inherits it — not 49 agent files edited individually.
4. Enterprise Governance by Default¶
Base Coat is infrastructure for governed AI assistance:
- Ambient instructions enforce security, naming, quality, and process standards in every Copilot conversation automatically
- Guardrail agents validate outputs before delivery
- Secret scanning hooks block credentials in commits
- CI validation ensures all assets have valid frontmatter, structure, and catalog entries
- Version-pinned distribution prevents drift across consuming repositories
5. Opinionated but Extensible Framework¶
Ship battle-tested defaults that work out of the box, but allow every asset to be customized. Teams adopt Base Coat as a baseline, then layer their own domain-specific agents, skills, and instructions on top.
Distribution methods: Git submodule, sync scripts (PowerShell + Bash), release artifact downloads, and template-based bootstrapping — all with SHA256 verification.
6. Agentic Workflow Enablement¶
Beyond individual agents, Base Coat supports multi-agent orchestration:
- Parallel dispatch patterns for fleet-mode sprints
- Merge coordination to prevent conflicts when multiple agents work simultaneously
- Structured handoff protocols between agents
- Sprint planning that decomposes goals into agent-assignable issues with wave dependencies
- Retrospective tooling that measures agent effectiveness
7. Cost-Aware Model Routing¶
Every agent carries a model field in YAML frontmatter for direct VS Code integration
plus a ## Model section with rationale and minimum viable model. Token economics
instructions guide budget-aware model selection so organizations can optimize cost
without sacrificing quality.
Model distribution (v2.1.0): claude-sonnet-4.6 (28), gpt-5.3-codex (16), claude-haiku-4.5 (3), claude-sonnet-4-5 (2), claude-sonnet-4 (1).
8. Adoption Measurement and Feedback Loops¶
Base Coat includes tooling to measure its own impact:
- Adoption scanner detects which repos have synced assets and tracks version drift
- Metrics collector correlates Base Coat coverage with PR velocity, CI success, and issue resolution
- Dashboard visualizes adoption trends and degradation alerts
- Copilot usage analytics track per-session cost and model distribution
Non-Goals¶
| What Base Coat is NOT | Why |
|---|---|
| A hosted service or SaaS product | It is infrastructure — files in your repo |
| A single-domain tool | It covers the full SDLC, not just one discipline |
| A replacement for human judgment | Agents stop and ask when scope is ambiguous |
| A code generation library | It governs how AI generates code, not what code |
| A runtime dependency | Consuming repos work fine if Base Coat is removed |
Success Criteria¶
- New repos start governed — bootstrap from a pinned release in under a minute
- Standards are ambient — instructions load automatically, no opt-in needed
- Updates are safe — version-pinned distribution with validation gates
- Agents are discoverable — one router, categorized catalog, keyword search
- Impact is measurable — adoption metrics, cost tracking, feedback loops
- The framework practices what it preaches — Base Coat is maintained using its own agents