fix(skills): use real --output-format flag in messaging-agents

[amelia-letta]

Summary

• The messaging-agents skill told agents to run letta -p ... --output json, but the CLI flag catalog only defines --output-format; strict parsing rejects --output with Unknown option
• The full-transcript pointer referenced letta messages list, but messages transcript is the command that renders tool calls; messages list returns raw JSON messages
• Added src/skills/messaging-agents.test.ts locking the documented headless flags to the CLI flag catalog so a stale flag reference fails CI

Builtin-skill-watch: messaging-agents@aa3e294d006b-a29c6f40c6ffdc12

Selected skill

src/skills/builtin/messaging-agents

Stale claim and current owning source

• --output json (SKILL.md line 105) vs src/cli/args.ts CLI_FLAG_CATALOG (output-format, no output alias) and src/index.ts strict parseCliArgs(processedArgs, true)
• letta messages list --agent <id> for "full transcript (including tool calls)" vs src/cli/subcommands/messages.ts (transcript action renders tool calls; list returns raw JSON messages)

Focused validation

• Parser probe: parseCliArgs(["-p","--from-agent","agent-x","--agent","agent-y","--output","json","hi"], strict) fails with Unknown option '--output'; --output-format json parses
• bun test src/skills/messaging-agents.test.ts (2 pass), bun test src/skills/dispatching-coding-agents.test.ts src/agent/skills-discovery.test.ts (18 pass), bun run check (12/12 pass)

Test plan

• bun run check passes
• New skill test locks --output-format and rejects --output json
• Reviewer confirms the JSON example matches actual letta -p --from-agent ... --output-format json output

👾 Generated with Letta Code

Co-Authored-By: Letta Code noreply@letta.com