autonomy invariant: judgment-preservation rationale sentence #10

Parent

#1 — PRD: Loop Engineering adoption (standalone candidate surfaced by the harness-engineering re-read; NOT one of the core 7)

What to build

Add one sentence to the autonomy-invariant rationale capturing the judgment-preservation framing: the human gate is a deliberate judgment-preservation choice — not a capability gap to be closed as models improve. The same loop machinery preserves or destroys engineering judgment depending on operator intent ("the loop doesn't know the difference, you do").

It lands wherever the autonomy invariant is canonically homed — coordinate with the separate HARNESS.md / CONTEXT.md invariant-home cleanup so it doesn't land in a soon-to-move location.

External support (rationale only, keep to one sentence): ThoughtWorks' human-review-checkpoint convergence; a self-repair vendor keeping "approval as the one manual step"; the corpus's "automate past the point where you can still vouch for the output, and you'll ship agent slop."

Acceptance criteria

• One sentence added to the invariant rationale, framed as judgment-preservation (not capability-gap).
• Lands in the canonical home of the autonomy invariant (coordinated with the HARNESS.md/CONTEXT.md cleanup).
• Doc-only, one sentence, minimal token cost.
• Green bar: bash skills/harness-audit/scripts/audit.sh clean vs baseline AND claude plugin validate --strict exits 0.

Blocked by

None — low priority. Coordinate with the autonomy-invariant-home cleanup (see PRD #1 Further Notes).