feat(runtime): add the thin-pointer AGENTS.md agent entry file

claude-templates/AGENTS.md points any harness at protocols.org, the rules locations, and the /name-to-file resolution rule, and tells it to degrade per each rule's documented fallback instead of skipping gates. make install links it to ~/.codex/AGENTS.md (new CODEX_DIR stanza) and install-ai.sh seeds a project-owned copy at bootstrap, never overwriting an existing one. The rulesets root carries a tracked symlink as dogfood. Five new bats tests cover the link idiom and the seeding. This resolves the generic-instruction-file blocker from the runtime spec, and velox picks up the global link on its next session's make install.

1 files changed, 19 insertions, 0 deletions

diff --git a/claude-templates/AGENTS.md b/claude-templates/AGENTS.md
new file mode 100644
index 0000000..97fd001
--- /dev/null
+++ b/claude-templates/AGENTS.md
@@ -0,0 +1,19 @@
+# Agent Entry Point
+
+You are this project's agent, whichever model or harness is running you.
+
+1. If the project has `.ai/protocols.org`, read it now and follow all
+   instructions. It is the single entry point for session behavior:
+   startup, session logging, inbox processing, wrap-up.
+2. Behavioral rules live in `.claude/rules/*.md` in the project (when
+   present) and `~/.claude/rules/*.md` globally. Read them; they bind
+   every session regardless of harness.
+3. A `/name` reference in any rule or workflow (`/voice`,
+   `/review-code`, `/brainstorm`, ...) resolves to a file:
+   `~/.claude/skills/<name>/SKILL.md` or `~/.claude/commands/<name>.md`.
+   Read that file and follow it. If it is absent, say so and apply the
+   rule's documented fallback rather than skipping the gate.
+4. Harness mechanics these files assume (hooks, `/clear`, popup tools)
+   may not exist in your harness. Degrade per each rule's own fallback
+   language. Never silently skip a verification or approval gate because
+   the tooling that enforces it is missing.