fbdl-mcp: Skill for Codex

Using the FBDL MCP server

This skill is the operations manual for fbdl-mcp.

When to use it

Use it whenever the user wants to:

• write or fix an FBDL script
• submit a script as a real FBDL run
• inspect or archive previous runs
• understand why create_fbdl_run was rejected

Tools

Read-only:

• list_entities
• list_actions
• validate_fbdl
• explain_fbdl
• resource fbdl://reference

API-backed:

• create_fbdl_run({fbdl_code, note})
• list_fbdl_runs({?limit, ?after})
• get_fbdl_run({id})
• archive_fbdl_run({id})

Standard workflow

1. Understand the scenario.
2. Use list_entities / list_actions if names or params are unclear.
3. Write the script with block grammar.
4. Call validate_fbdl before presenting or submitting it.
5. Only when the user explicitly wants a real run, call create_fbdl_run.
6. Poll with get_fbdl_run if results are needed.
7. Archive old runs when the user is done with them.

Run-submission rules

create_fbdl_run is rate-limited by the server itself:

1. Only one create_fbdl_run can be in flight at a time.
2. After success there is a 30 second cooldown.
3. After failure there is a 60 second cooldown.
4. Validation failures do not trigger cooldown because they never hit the API.
5. If the account is at the max active-runs limit, the server may return a hint telling the caller to list and archive old runs first.

Do not silently retry or parallelize submissions.