Noticed while auditing the plugin: all three agents — fable-advisor, fable-implementer, codex-implementer — have strong, specific description: frontmatter with good trigger phrases, but none include <example> blocks.
Claude Code's own agent guidance treats 2+ <example> blocks (with Context: / user: / assistant: scenarios) as the reliable-triggering baseline — they help the router pick the right agent when a request is ambiguous. The prose triggers you have are good; adding a couple of concrete dialogue examples per agent would make routing more robust, e.g.:
<example>
Context: A routine CRUD endpoint the spec fully determines
user: add a DELETE /todos/:id endpoint following the existing pattern
assistant: I'll route this to codex-implementer — it's spec-determined, no judgment call.
<commentary>Straightforward feature → default codex lane.</commentary>
</example>
Filing as a suggestion rather than a PR since the examples are best written in your own voice / against the scenarios you actually care about — but happy to send a PR with drafted blocks for all three agents if that's easier. (Separately opened #9 for a couple of mechanical fixes.)
Noticed while auditing the plugin: all three agents —
fable-advisor,fable-implementer,codex-implementer— have strong, specificdescription:frontmatter with good trigger phrases, but none include<example>blocks.Claude Code's own agent guidance treats 2+
<example>blocks (withContext:/user:/assistant:scenarios) as the reliable-triggering baseline — they help the router pick the right agent when a request is ambiguous. The prose triggers you have are good; adding a couple of concrete dialogue examples per agent would make routing more robust, e.g.:Filing as a suggestion rather than a PR since the examples are best written in your own voice / against the scenarios you actually care about — but happy to send a PR with drafted blocks for all three agents if that's easier. (Separately opened #9 for a couple of mechanical fixes.)