Skills: name parity between skill tool references and PipefyClient methods

[mocha06]

Problem

The skills name MCP tools. Some of these tools are MCP-layer conveniences, and no PipefyClient method has the same name. An agent that reaches Pipefy through the SDK gets "unknown tool" when it follows the skill.

MCP tool used by a skill	Implemented at (packages/mcp/src/pipefy_mcp/tools/)	Closest PipefyClient method
get_ai_automation	ai_automation_tools.py:131	get_automation
get_ai_automations	ai_automation_tools.py:180	get_automations, then ai_preflight.filter_ai_automation_summaries
delete_ai_automation	ai_automation_tools.py:244	delete_automation
get_labels	pipe_tools.py:837	get_pipe (then .labels)
get_pipe_report	report_tools.py:110	get_pipe_reports(pipe_uuid, first=1, report_id=)
remove_member_from_pipe	member_tools.py:180	remove_members_from_pipe
fill_card_phase_fields	pipe_tools.py:1179	get_phase_fields + update_card (the elicitation step is MCP-only)

Measured against origin/main 3442f0c8. The client has 179 public async methods, and the MCP registry has 187 names.

Proposal

For each row, pick one of these:

• (a) add a PipefyClient method with the MCP name, or
• (b) change the skills to name the client method, and keep the MCP tool as an alias.

We prefer (a) where the MCP tool adds real logic (get_ai_automations, remove_member_from_pipe with its removal check). We prefer (b) where it is only a projection (get_labels, get_pipe_report).

Non-breaking

Both options only add methods or change skill text. No MCP tool is removed.

Related: the skills-lint issue in this milestone.