What problem does it solve? When you fan out 2+ parallel research or review lanes, each returns its own finding blocks, and merging them by hand produces truncated claims, broken cross-references, duplicated findings, and reports that fail strict renderers. This Skill defines the lane output contract and the exact assembly procedure to produce one clean, evidence-graded record. ## Core Features & Use Cases - Verbatim block extraction and truncation: Lifts ### F# — claim [GRADE] blocks from authoritative full summaries (never truncated delegation transcripts) and cuts each block at embedded ## Still open headers so the renderer captures all findings. - Consolidation and renumbering: Dedupes cross-cutting findings into a single section with provenance notes, moves ACCEPTED items to a register, renumbers sequentially, and runs a stale cross-reference pass. - Render-gated verification: Uses the renderer as a format gate that refuses ungraded claims and READ/MEASURED findings without Evidence lines, plus a checklist covering grade mix, severity counts, and section structure. - Use Case: After dispatching three parallel security review lanes via Hermes, pull the full summaries from the Hermes state DB, merge 41 findings into one report, renumber them, and render to a temp directory to confirm the provenance chart counts every finding. ## Quick Start Assemble the findings from my three parallel review lanes into a single evidence-graded report and verify it renders cleanly.
