| name | reyn_cheat_sheet |
| description | Reyn-specific usage cheat sheet -- which mechanism to reach for (skill/pipeline/mcp/hook/present), composition idioms, op essentials, and pointers to the full specs. Read this before authoring a new part or composing several. |
Reyn cheat sheet
This is the gap-filler between "reyn has these parts" and "you use them
correctly". concept/reference docs describe each mechanism in isolation;
this skill is the composition know-how in the cracks between them (0060
Addendum D1). Read on demand when deciding which mechanism to use, or before
authoring a new skill/pipeline/hook/present-view.
This L2 router stays deliberately small: it is enough on its own to decide
whether you need to go deeper and which references/ file answers your
question -- you should not need to open a reference just to find out if you
need it.
Decision tree (which mechanism)
- Need input (new data, or a reactive trigger) ->
hook | mcp |
the builtin build_and_query_rag_corpus skill (document search over the
operator's own files). Deeper detail (hook action schemes, a
CI-verified worked hook example, MCP tool discovery):
input-hooks-and-mcp.md
-- read when wiring a reactive trigger or calling an external tool/resource.
- Need workflow (multi-step orchestration) ->
skill | pipeline | an
mcp tool call mid-flow. Deeper detail (pipeline step grammar, the
flagship CI-verified worked pipeline, SKILL.md authoring shape):
workflow-pipelines-and-skills.md
-- read when writing pipeline DSL steps or authoring a new skill.
- Need output (show a result, or write externally) ->
present |
render_template | an mcp write. Deeper detail (present's blueprint
catalog and the input/output caveat, the self-review gate before
promotion):
output-present-and-self-review.md
-- read when deciding whether/how to show a result via present, or
self-reviewing an authored asset before it becomes a reused part.
Reuse before authoring: check the existing catalog (list_actions) for a
part that already covers the need. Author only when nothing fits, and
self-review anything you author or promote (an agent step + schema, see
output-present-and-self-review.md)
before it becomes a reused asset -- an ungated authored part is a liability,
not a shortcut.
