id(world-line): PR #808 — join pump --command-mode kills the session on a parse Message (SSH shows it and continues), and /logs [seconds] is parsed then discarded by both consumers #819

Problem

PR #808 introduces the shared line grammar (pkgs/id/src/world_line.rs, branch id/phase-11) whose stated contract is that one line "works identically as an SSH command, a piped script line, and a CLI one-off" — and the PR body claims the id world join --command-mode pump and SSH command mode both speak it (#546). Two places break that parity:

1. A typo in the join pump kills the session; on SSH it just shows a message

• world_line.rs defines WorldLine::Message(String) as "A command that did not parse; the message is shown to the user" (unknown commands, usage: /input <hex>, usage: /logs [seconds], usage: /document <name>).
• SSH command mode honours that: pkgs/id/src/world_ssh.rs:667-672 renders {message}\r\nworld> and the session continues.
• The join pump does the opposite: pkgs/id/src/commands/world.rs:1054-1056 maps WorldLine::Message(message) to anyhow::bail!("{message}"), and the pump loop calls pump_command_frame(line)? (commands/world.rs:1007), so the error breaks the writer loop, closes the connection, and ends id world join with a failure exit.

One mistyped /imput … or /logs soon in --command-mode therefore disconnects the user from the world instead of showing the usage line and a fresh prompt. The pump's own test even enshrines the divergence: assert!(pump_command_frame("/nope").is_err()).

2. /logs [seconds] is parsed, validated — and silently discarded by both consumers

• world_line.rs:29, 78-87 — WorldLine::Logs { seconds: Option<u64> }; a non-numeric argument is a usage error, so the grammar promises the argument means something.
• SSH: world_ssh.rs:846 maps WorldLine::Logs { .. } to WorldCommand::Logs, discarding seconds (the handler then prints the "event tailing is not available in command mode" stub regardless of the argument).
• Pump: commands/world.rs:1051 maps WorldLine::Logs { .. } to {"type": "logs"} with no seconds field.

So /logs 5 and /logs 5000 behave identically everywhere, with no hint to the user. Either the frame/handler needs to carry the value, or the grammar should stop accepting (and validating) an argument nothing uses.

Why it matters

The whole point of #546 / F11.5 is one grammar with one behaviour per surface; a shared parser whose error and argument semantics differ per consumer re-creates the divergence the refactor exists to remove, and the pump case is user-hostile: a recoverable typo becomes a disconnect (and a nonzero exit for scripted/piped use of --command-mode).

Evidence

• PR id: Phase 11 foundations — F11.10 module split (world.rs, world_session.rs) #808 (OPEN, branch id/phase-11); file:line on the PR branch as fetched this run (gh api .../contents/...?ref=id/phase-11). Not on main yet — fix-before-merge.
• Dedup: id: one line grammar for SSH command mode, the id world join pump and CLI verbs (--command-mode) #546 is the feature tracker this PR implements; id: F12.16 id world doctor #559/id: F12.1 apply wiring — module install + schedules reconciliation #807-style searches plus "pump command" found no issue covering the Message//logs behaviour. The PR body itself notes "/logs in command mode is a documented stub" for SSH — it does not note the pump's session-killing Message handling or that the pump sends a bare logs frame.

Acceptance

• In --command-mode, a line that parses to WorldLine::Message is shown to the user and the pump keeps running (matching SSH command mode); only /quit / EOF / a real I/O error ends the session. The pump test asserts the message path does not error and a following valid line still produces its frame.
• /logs [seconds] either takes effect identically on both surfaces (value carried into the logs request / stub message states what is ignored), or world_line rejects the argument with a usage message that says it is unsupported — in both cases, SSH and the pump behave the same.
• A parity test runs the same line corpus through both consumers' mappings (the structural-guard style PR id: Phase 11 foundations — F11.10 module split (world.rs, world_session.rs) #808 already uses for the projection) so the two mappings cannot silently drift again.