Repository Structure

Repository Structure

specroute/
├── AGENTS.md                  canonical, vendor-neutral root context
├── CLAUDE.md / GEMINI.md      per-vendor delegation shims
├── README.md / ROADMAP.md     project framing
│
├── agentic-docs/              conceptual deep references (philosophy, decision frameworks)
├── prds/                      PRD templates, lifecycle dirs (active / deprecated / archive), examples
├── specs/                     spec triplet templates (requirements, design, tasks), ADR, feature spec
├── agents/                    agent template, 7 archetypes, examples, cross-vendor roster
├── skills/                    skill-template/SKILL.md + examples (folder-per-skill convention)
├── commands/                  per-vendor command templates (Claude markdown, Gemini JSON)
├── hooks/                     per-vendor hook templates covering all six vendors
├── prompts/                   shared/, claude/, codex/ prompt sets + master-prompt trio
├── workflows/                 5 end-to-end execution playbooks
├── rules/                     engineering / code-review / security / docs rules + per-vendor surfacing
├── runtimes/                  copy-pasteable per-vendor runtime layouts + MCP single source
├── tools/                     sync-skills.py and other utility scripts
├── examples/                  sample-project/ canonical worked example
├── assets/                    images and diagrams
│
├── .claude/                   THIS repo's Claude Code runtime — not the consumer template
    ├── agents/                12 implementation agents
    ├── skills/                6 contributor skills (scaffold-artifact, all-hands, doc-currency-check, …)
    ├── commands/              4 slash commands (/audit, /parity, /sanitize, /status)
    ├── hooks/                 SessionStart banner, PreToolUse sanitize gate, PostToolUse frontmatter
    └── agent-memory/          per-agent persistent context
│
└── .agents/                   THIS repo's vendor-neutral runtime root
    └── skills/                same 6 skills, agentskills.io frontmatter — what Codex reads

Two distinct concerns

Consumer-facing templates live under top-level directories (prds/, specs/, agents/, skills/, commands/, hooks/, prompts/, rules/, runtimes/, examples/).

The implementation team that builds SpecRoute itself lives under .claude/ (Claude Code) and .agents/ (everything else). Don't confuse those with runtimes/. The consumer drops runtimes/.claude/ into their own repo; they do not copy .claude/.

.agents/skills/ exists because runtimes/.<vendor>/ is a template, not a runtime — no CLI reads that path. Until it was added, running Codex inside this repository found no skills at all. See Implementation Team.

Where do new docs go?

What you're adding	Where
A new template or example	The corresponding top-level directory (e.g. prds/templates/, agents/examples/)
Conceptual or framework-level reference	agentic-docs/
Recurring engineering procedure	workflows/
Standing engineering standard	rules/
Directory-specific guidance	That directory's README.md
Per-vendor configuration / setup	runtimes/.<vendor>/

See Documentation Structure for the full decision tree.

Per-directory READMEs

Every directory ships a README.md. Treat those as the authoritative answer to "what belongs here and how do I add to it?" The wiki pages summarize and link out:

• PRDs · Specs · Agents · Skills · Commands · Hooks · Prompts · Rules
• Workflows · Vendor Matrix (for runtimes/)
• Worked Example (for examples/)

File extensions and conventions

• Templates and content: .md (Markdown).
• Frontmatter: YAML at the top, fenced by ---. Required fields vary by artifact — see Frontmatter Contracts.
• Scripts: bash for hooks; Python 3 for tooling. No external runtime dependencies.
• Vendor-specific files keep their native shape (Gemini's .gemini/commands/*.toml and config.toml stay TOML; Codex's config.toml stays TOML; Kiro's .kiro/hooks/<name>.json stays JSON).

What this repo is NOT

• Not a package manager — there's no package.json or build step.
• Not a code generator — no runtime emits files.
• Not vendor-specific tooling — patterns work across all six supported CLIs.
• Not a place for proprietary business logic — sanitization is non-negotiable. See Sanitization.

See also

• Quickstart · Worked Example
• Documentation Structure — the full decision tree for new docs
• Implementation Team — what .claude/ contains and how it's wired