Composition — --hard and --auto

--hard and --auto are both propagated flags. They may coexist. The gate fires at exactly one named point; everywhere else, each flag's normal semantics apply.

Propagation graph

/cheese --hard <input>
        │
        ├──► /mold --hard            (passes --hard through to /cook)
        │       │
        │       └──► /cook --hard    (passes --hard through to /press)
        │
        └──► /cook --hard            (or invoked directly)
                │
                └──► /press --hard
                        │
                        └──► /age --hard
                                │
                                └──► /cure --hard   ← gate fires here
                                        │
                                        └──► /hard-cheese <slug>

Upstream skills (cheese, mold, cook, press, age) are pure pass-through for --hard. They never invoke the gate themselves. /cure is the only pipeline skill that calls /hard-cheese.

The matrix

Invocation	Gate fires?	When	Notes
/hard-cheese <slug> standalone	Yes	Immediately.	No pipeline state required. Always idempotent up to the freshness check.
/cure --hard <slug> interactive	Conditionally	Only when the user picks the share-for-review handoff option.	Picking "re-review" or "stop" → no gate.
/cure --auto --hard --stake medium+	Yes	When cure reads next: done from the age slug it just invoked (chain-clean or two-cure-pass cap reached). Fires from that cure frame before returning to the caller.	This is the single sanctioned --auto puncture point.
/cure --auto (no --hard)	No	n/a	Existing behaviour preserved.
/cook --hard (without continuing through cure)	No	n/a	The flag is set but the pipeline stops before the firing point. No gate, no artifact.
/cheese --hard <input>	No (at routing)	Router only propagates the flag; whatever target it dispatches to carries --hard.	Gate still fires at the eventual /cure invocation, if any.

The single puncture point

--hard punctures --auto exactly once, at the end of /cure --auto --hard's final pass. Everywhere else, --auto's skip-handoff semantics apply: cook does not pause, press does not pause, age does not pause, the intermediate cure passes do not pause. Only the terminal moment — when the chain is about to exit and the user is about to share the code — triggers the gate.

This is intentional. Vibecheck's faithful analogue is "before code is applied" — singular, not per-step. Punching gates into every phase of the auto chain would create fatigue and dilute the gate's signal.

Non-TTY guard

If /hard-cheese detects it is running without an interactive input stream (no human can respond to the vibecheck prompt), it fails closed and aborts. The puncture only makes sense when a human is in the loop. A vacuous "auto-pass" with no human present would defeat the entire mechanism.

Auto-driven CI pipelines should not pass --hard. If they do, the gate aborts with a clear error: "--hard requires an interactive TTY; remove --hard or run interactively".

Flag precedence summary

• --auto without --hard: chain runs forward without pause. Unchanged from today.
• --hard without --auto: gate fires at cure's interactive share-for-review handoff (if selected).
• --auto --hard: chain runs forward without pause through all intermediate steps; gate fires once at the terminal moment; user must clear it; chain then exits.
• --auto --hard on a non-TTY: aborts with the non-TTY error above.

There is no silent precedence. The only point where one flag overrides the other is named here.