#1125 Honor invocation tokens in rulebook bodies

[williamthorsen]

What

A rulebook can now route an agent to another skill, subagent, or rulebook as a direct invocation rather than a name mentioned in prose; each delivery renders that invocation in the form the receiving harness understands. A rulebook that routes to a missing target, or to one that deploys nothing an agent can invoke, now fails the run. Naming a routing target is now enough to have it deployed alongside the rulebook that names it.

Why

A hub-and-spoke guidance set exists to send an agent to the right reference, and a rulebook had no way to say where. Prose routing left the reader a name to type, with nothing guaranteeing the named artifact had been deployed and nothing failing when it did not exist -- a gap that widens as more packages ship a cheatsheet plus a set of reference rulebooks.

Details

🎉 Features

• A {rulebook:<slug>} token in a rulebook body renders the harness skill sigil plus the skill name its target deploys under, honoring a skill-name override on that target.
• {skill:} and {subagent:} tokens render in a rulebook body as they already do in skills and subagents, in both ambient and skill delivery.
• A token naming a rulebook that deploys no skill to invoke fails the run before anything is written, reporting every offending token at once and pointing at dependencies: for the relationship it cannot express.
• The error for an unusable token names the file carrying it, so an author goes straight to the source.
• Linking to a sibling rulebook still fails, and the rejection now names the {rulebook:<slug>} token that addresses it instead.
• The content specification documents the new kind, why only a rulebook body renders one, why an ambient-only target is rejected, and which token kinds are checked against the deployed set.

🐛 Bug fixes

• A declared subagent whose body fails to render aborts the sync before the ambient host and skill files are written, and fails --dry-run rather than passing it. This covers an unmapped {tool:NAME} placeholder alongside the newly rejected token.

🏗️ Internal features

• The invocation-token grammar recognizes a rulebook kind alongside skill and subagent, resolving a target to the skill name it deploys under or reporting why it cannot.
• A token in a rulebook body is a closure edge, so a named target deploys without a duplicate dependencies: entry. A rulebook naming itself keeps the reference without becoming a dependency cycle, and a {rulebook:} token counts as an edge only from a rulebook body.

♻️ Refactoring

• The subagent render is extracted from its deploy so the pre-write gate and the write share one path.
• The rulebook slug derived in the sibling-rulebook rejection no longer shadows the enclosing link-target parameter.

🧪 Tests

• The shipped-content suite builds a rulebook catalog from every shipped rulebook, turning it into a build-time gate on tokens as well as links.
• A subagent named by a token in a rulebook body is asserted to enter the deployed closure, alongside the skill and rulebook edges.

Closes #1125

[github-actions]

Dependency audit

Production dependency audit passed.