contract-snapshot

Contract Snapshot

A reference skill for the M3-C output_format: table mode. Produces a side-by-side grid of the same questions across N contracts — the in-house lawyer's "compare clauses across N agreements" workflow. Each cell carries a citation back to the source document, and failed extractions render as not found rather than confidently-wrong text.

When this skill applies

Apply when the user wants to compare a small number of well-defined questions across a corpus of similar contracts:

• "What is the term, survival, and governing law across these 5 NDAs we're tracking?"
• "Pull out the payment terms, IP ownership, and termination triggers across these 10 MSAs."
• "For my Q3 portfolio review, I need a grid of these 30 vendor contracts' liability caps."

Do not apply this skill to:

• Single-document review — use the appropriate document-specific skill (nda-review, msa-review-saas, etc.).
• Free-form chat against contracts — that's the regular Chat surface.
• Tasks where the questions aren't well-defined upfront — the column queries must be specific enough to extract from each row's source document; vague queries produce poor cells.

Inputs

The skill takes a set of documents (selected via the Tabular Review UI from a Knowledge Base, a Project, or a free file selection). The four columns above run as Citation Engine-grounded extractions against each document.

To adapt this skill for a different contract type (e.g., MSAs), fork the skill and rewrite the four column queries. Keep them short, specific, and quote-asking — the Citation Engine works best when the model is encouraged to quote rather than paraphrase.

Per-column overrides

This skill demonstrates the two per-column overrides M3-C1 supports:

• ensemble_verification: true on the Survival column. Survival is the load-bearing economic term in confidentiality agreements (a 3-year confidentiality term with a 10-year survival is very different from one with no survival), so cells in this column run through Stage 4 of the Citation Engine cascade — three judges debating whether the cell value is faithful to its citation. Higher cost, higher confidence.
• minimum_inference_tier: 3 on the Governing Law column. The skill-level floor is Tier 2 (commercial inference). Governing-law extraction is the column most likely to surface counterintuitive answers (e.g., a contract drafted under California law but with a Delaware forum-selection clause); routing this column to Tier 3+ avoids the cheapest models' tendency to collapse the two into one answer.

Other columns inherit the skill-level ensemble_verification: false and minimum_inference_tier: 2 defaults — appropriate for the lower-stakes, more-extractive Term and Carveouts columns.

Output format and downstream surfaces

The grid renders in the Tabular Review UI (/lq-ai/tabular/) with sticky-first-row and sticky-first-column. Each cell shows the extracted value + a small confidence chip; click anywhere on the cell to open the existing M2-C2 citation drawer with the source document highlighted at the cited chunk.

From the result view, operators can:

• Export the grid as XLSX — each cell carries its citation as an Excel comment with a clickable link back to the deployment.
• Export as CSV — citations are flattened to sibling {column_name}_citation_url columns.
• Run a bulk operation — e.g., "Redline the Survival column in all rows" runs the nda-review skill against each row's source document with the survival value as context.

Disclaimer (per Decision F)

This skill is a starting point, not a vetted template. The four columns are appropriate for many NDAs but won't be right for every corpus. Before relying on the output of a Tabular Review run on this skill, the user-attorney should:

1. Review the column queries — do they match the questions you actually want answered for this corpus?
2. Spot-check at least one cell per column against the source document — does the extraction faithfully represent the source?
3. Treat any not found cell as a signal to investigate, not as definitive evidence that the clause is absent.

The output is a draft for an in-house lawyer to validate, not a final compliance artifact.

Fork and tune

This skill is intentionally minimal so operators can fork it as a starting point:

# skills/my-org/msa-snapshot/SKILL.md (operator's fork)
output_format: table
columns:
  - name: Payment Terms
    query: What are the payment terms (frequency, days-to-pay, late-fee provisions)?
  - name: IP Ownership
    query: Who owns IP created during the engagement (work-for-hire, license-back, joint)?
  - name: Termination
    query: List each termination right (for cause, for convenience, notice periods).
  # ... more columns

The Tabular UI accepts both saved skills (like this one) and ad-hoc column specs entered directly in the wizard's column step.