afv-library/skills/agentforce-architecture-analyze/assets/mermaid/invocation_sequence.mmd

21 lines
1004 B
Plaintext

%% invocation_sequence.mmd - section 3 invocation sequence (sequenceDiagram)
%%
%% Placeholders (substituted by render_architecture.load_mermaid):
%% PARTICIPANTS placeholder: one `participant <Name>` line per lane,
%% newline-joined; deactivate with
%% `<Dst>-->>-<Src>: <response>`.
%% MESSAGES placeholder: one `<Src>->>+<Dst>: <label>` per message line.
%%
%% placeholder names above are written as bare tokens (no double-
%% curly) so load_mermaid's single-pass str.replace cannot substitute them
%% here. That keeps %% headers intact in the rendered output - Mermaid
%% ignores %% lines as comments at render time, and debuggability wins.
%%
%% Renderer (render_architecture.py::_render_invocation_sequence) is the sole
%% consumer. Generation-aware: classic ReAct uses Planner<->TopicClassifier
%% <->ActionExecutor; NGA orchestrations add one lane per sub-agent.
sequenceDiagram
autonumber
{{PARTICIPANTS}}
{{MESSAGES}}