daily-fix: bash -n pin test for Step 10d TG fenced blocks

Overview / Motivation

Auto-filed by /daily 2026-07-29 (problem sweep; emitting source: parked formal candidate on #1790 (implementer r1, sweep fp 781f787dccc2, ts 2026-07-29T09:16:51Z)).

Goal

Pin the bash-parseability of the Step 10d TG fenced blocks so a comment/formatting edit cannot re-break copy-paste executability without failing a test.

Workflow gap

• Bug observed: Orchestrators execute these fenced blocks verbatim; no mechanical check pins their bash parseability — the existing substring pins are structurally blind to this regression class (#1790).
• Why it is a workflow gap: grep -n '_tg_blocks' tests/test_step10d_guard3.py -> helper exists (L1392); the file's only bash -n assertion (L1242) is scoped to the post-merge guard region's single fence, not the TG blocks (read in context 2026-07-30, this run).
• Confidence (emitter): medium
• verified-at-filing: Per the candidate's diff sketch: a test iterating _tg_blocks(_skill_text()), substituting <N> -> dummy id, asserting bash -n rc==0 per block.

Proposed change (refine in planning)

tests/test_step10d_guard3.py

Scope / surfaces

• Primary target: `

Provenance

• workflow_fix_target: tests/test_step10d_guard3.py

• fingerprint: daf77c9ac3eb

• Origin park: #1790 events 2026-07-29T09:16:51Z (fingerprint 781f787dccc2); verbatim candidate block lives in that marker. `

• Grep the workflow surface for the pattern before editing and update every hit.

Constraints / invariants

• Workflow-surface only; scripts/workflow_lint.py --check-asks passes; ruff clean on touched files.