tags?}
E -- No --> F[Write (unmodified) HTML to output]
E -- Yes --> G[For each Mermaid
]
G --> H[Extract diagram code & caption]
H --> I[Sanitize caption → base filename]
I --> J[Write .mmd file]
J --> K{executable provided?}
K -- No --> L[Skip rendering (leave HTML unchanged)]
K -- Yes --> M[run_external_executable_with_ai_retry()]
M -- Success --> N[Replace
with 
]
M -- Failure --> O[Leave
intact]
N --> P[Count +1 success]
L & O --> P
P --> Q[After loop: write modified HTML to output]
Q --> R[End]
Detailed sub-flow of run_external_executable_with_ai_retry
flowchart TB
Start(["Start run_external_executable_with_ai_retry"])
Start --> ReadMMD["Read original .mmd into current_code"]
ReadMMD --> LoopAttempt{"attempt_num ≤ max_ai_attempts?"}
LoopAttempt -- No --> FailAll["Revert to original .mmd\nreturn False"]
LoopAttempt -- Yes --> WriteMMD["Write current_code to .mmd file"]
WriteMMD --> BuildCmd["Build command_parts & cwd"]
BuildCmd --> Exec["subprocess.run(...)"]
Exec -->|exit 0 + PNG exists| Success["return True"]
Exec -->|exit 0 but no PNG| CollectErr["mark failure + collect stderr/stdout"]
Exec -->|CalledProcessError| CollectErr
CollectErr --> CheckAI{"attempt_num < max_ai?"}
CheckAI -- No --> FailAll
CheckAI -- Yes --> AIcall["call_openai_to_fix_mermaid_iteratively"]
AIcall -->|new code| UpdateCode["current_code ← suggestion"]
UpdateCode --> LoopAttempt
AIcall -->|no suggestion| FailAll
Sequence diagram of the AI‐correction call
sequenceDiagram
participant Script
participant OpenAI
Script->>Script: prepare prompt (code + error + history)
Script->>OpenAI: send prompt via chat or completion API
OpenAI-->>Script: raw response (possibly with ```mermaid fences)
Script->>Script: strip fences & whitespace
alt cleaned code valid
Script-->Script: return cleaned Mermaid code
else no useful fix or “cannot fix”
Script-->Script: return None
end