ADR 0001 - Ecosystem-wide agent config via symlink to rstsr-agents

All repos in the rstsr ecosystem (rstsr, rstsr-book, rstsr-ffi, device crates) share one agent configuration by symlinking .agents/, .claude/, and AGENTS.md to the single rstsr-agents repo. No repo keeps local agent config, skills, or rules.

Chosen over copying (drifts out of sync) or a workspace-include (agent runtimes do not resolve Cargo-style paths): a relative ../ symlink is the one mechanism every agent runtime follows, and skills are auto-discovered through .claude/skills/.

1. Considered

• Copy-on-bootstrap - rejected (drift).
• Per-repo config - rejected (no single source of truth).

2. Consequences

A broken upstream rstsr-agents breaks every repo at once. Accepted for simplicity.

3. Amendment (2026-09-06)

• Links point directly at the sibling repository (.claude -> ../rstsr-agents, CLAUDE.md -> ../rstsr-agents/AGENTS.md, and so on), not through intermediate pack-root links; the agent-setup skill (skills/agent-setup/scripts/link.sh) in rstsr-agents bootstraps and repairs them. Uniform four links in every repo and at the pack root.
• The per-repo override layer now exists, via the gitignored *.local* catch-all: AGENTS.local.md (per-developer instructions, read in addition to AGENTS.md/CLAUDE.md) and settings.local.json (personal Claude Code settings placed inside rstsr-agents).
• See the AI Agent Workflow page for setup.