A pack of structured Claude prompts for MSA (Master Services Agreement) negotiation — first-pass redline against the firm’s playbook, counter-position drafting against the counterparty’s redlines, fallback-position laddering (preferred → acceptable → walk-away), and counterparty-style reading (do their redlines suggest aggressive vs cooperative posture). Every prompt is calibrated against a firm-specific MSA playbook and produces structured output the contracts manager edits before sending. Replaces the “open the playbook PDF, read, write redlines manually” loop with anchored prompts that surface the same playbook positions consistently across every negotiation.
When to use
The firm has a written MSA playbook with named positions per clause (preferred / acceptable / walk-away). Without the playbook the prompts have nothing to anchor against.
Volume of MSA negotiations is high enough that consistency across negotiations matters (typically >5 active MSAs at any time).
The contracts manager or in-house counsel reviews and edits the prompt outputs before sending. The pack is decision support; the human is the negotiation surface.
When NOT to use
Fully novel deal structures that the playbook doesn’t cover (e.g. first-of-its-kind partnership, joint development arrangement). The pack works against established patterns; novel deals need counsel from scratch.
Replacing counsel’s strategic judgment. The pack drafts; counsel decides which positions to take, which battles to pick, when to escalate.
Adversarial negotiations where counterparty posture has shifted to litigation-prep. The pack is for commercial negotiation, not pre-litigation positioning.
Auto-sending counter-redlines. The pack produces the draft; the human sends.
Setup
Drop the bundle. Place apps/web/public/artifacts/msa-negotiation-prompt-pack/msa-negotiation-prompt-pack.md somewhere your contracts team can read (Notion, internal wiki, a Claude project’s knowledge files).
Author the firm’s MSA playbook. Per common clause (limitation of liability, indemnification, IP ownership, termination, confidentiality, payment terms, warranties, SLA), document: preferred position, acceptable position, walk-away position, with rationale for each level.
Create a Claude project per active negotiation. Drop the playbook + the counterparty’s draft + any prior round of redlines as project knowledge.
Run the prompts in sequence. First-pass redline → counter-position drafting (after counterparty responds) → fallback laddering (when stuck) → posture reading (anytime).
Review and edit before sending. The contracts manager owns voice and judgment; the prompts give structure and consistency.
R1. Read the counterparty’s MSA draft and produce a redline per clause against the firm’s playbook. For each redline, cite the playbook position (preferred / acceptable / walk-away) and the specific replacement language.
R2. Identify “missing clauses” — clauses standard in the firm’s MSA template that don’t appear in the counterparty’s draft. Flag each with the recommended insertion location and language.
C1. Read the counterparty’s response to the firm’s redlines. For each clause where they pushed back, draft the firm’s counter-position with rationale grounded in the playbook.
C2. Identify which counterparty positions the firm should accept (within the playbook’s “acceptable” range) vs. push back on (outside the range). The output is a triage table the contracts manager uses to scope the next round.
Tier 3 — Fallback laddering
F1. Given a clause where the firm and counterparty are stuck, produce the playbook’s fallback positions (preferred → acceptable → walk-away) and a rationale for each step. Includes the “horse-trading” question: what would the counterparty have to give to take this position?
F2. Given a stuck negotiation overall, surface the cross-clause trades that might unblock — the counterparty wants A and the firm wants B; can we trade?
Tier 4 — Counterparty posture reading
P1. Read the counterparty’s redlines and characterize their negotiation posture: aggressive (large redlines on liability and IP), cooperative (limited redlines, willing to accept playbook), conservative (defending standard terms with minor edits), uninformed (redlines that suggest they don’t understand the implications).
P2. Compare the counterparty’s posture across rounds. Trends matter: a counterparty that opened cooperative and shifted aggressive may have brought in outside counsel mid-negotiation, suggesting escalation risk.
Cost reality
LLM tokens per prompt invocation — typically 10-30k input (playbook + counterparty draft + skill instructions) and 2-5k output. Per-prompt cost ~$0.10-0.30.
Per-negotiation total cost — running all 8 prompts across the negotiation lifecycle is ~$1-3.
Contracts-manager / counsel time — the win. First-pass redline by hand is 60-90 minutes per MSA; with R1 + R2 it’s 15-30 minutes including review. Across a quarter with 20 active MSAs, this is meaningful.
Setup time — 20 minutes once. Playbook authoring is the binding cost (typically a one-time 20-40 hour project that pays back over months).
Success metric
Cycle-time per round — wall-clock from “counterparty sent draft” to “firm sends redlines back.” Should drop from 3-7 days to under 2 days.
Cross-negotiation consistency — at quarterly review, does the firm’s position on a given clause look consistent across active negotiations? Inconsistency was the most common pre-pack failure mode.
Counsel-edit rate per output — share of prompt outputs the counsel edits substantially. Should sit at 20-40%; below 10% means the counsel is rubber-stamping; above 60% means the playbook anchors are too vague.
vs alternatives
vs Spellbook / Harvey / DraftWise MSA modules. Those products integrate redlining into Word and offer firm-specific playbooks. Pick them if your contracts team lives in Word; the install + license cost is meaningful but the in-flow UX is hard to beat.
vs ChatGPT-style “redline this MSA against this playbook.” Generic chat returns paragraph feedback. The pack is structurally different: per-clause output with playbook citation, fallback laddering, posture reading.
vs hand-authored redlines. Right for the highest-stakes MSAs (M&A, JV) where counsel’s judgment dominates. The pack earns its setup cost on the everyday vendor and customer MSAs.
Watch-outs
Playbook drift.Guard: prompts cite playbook positions explicitly; missing or stale playbook positions surface as “playbook doesn’t cover this” rather than fabricated positions.
Counterparty-confidential terms.Guard: MSA drafts contain commercial-confidential terms. Use API access with zero-retention configuration; don’t paste into shared chat surfaces.
Auto-send drift.Guard: prompts produce drafts; the contracts manager sends. Pack does not output executable / sendable content.
Strategic-context gap.Guard: prompts reflect the playbook, not the strategic context (e.g. “this customer is the firm’s largest expansion target — accept worse terms”). Counsel injects strategic context before sending.
Posture-reading bias.Guard: P1 / P2 are heuristics. The contracts manager weights against actual interaction history; the prompt’s posture call is a starting point.
Stack
The bundle lives at apps/web/public/artifacts/msa-negotiation-prompt-pack/:
msa-negotiation-prompt-pack.md — the eight prompts, paste-ready
Tools: Claude. The output drops into Word redline format, your CLM, or back into the negotiation thread.
# MSA Negotiation — Eight Prompts for Claude
A pack of structured prompts for MSA negotiation: first-pass redline, counter-position drafting, fallback laddering, counterparty-style reading. Each prompt is calibrated against the firm's MSA playbook and produces structured output the contracts manager edits before sending.
## How to use this pack
1. Create a Claude project per active negotiation: `msa-<counterparty>-<matter-id>`.
2. Drop the firm's MSA playbook in as project knowledge. Drop the counterparty's current draft. Drop any prior round's redlines.
3. Save each prompt below as a saved prompt within the project.
4. Run R1 + R2 on the counterparty's first draft. Run C1 + C2 after their response. Run F1/F2 when stuck. Run P1/P2 anytime for posture context.
5. Edit before sending. The contracts manager owns voice and judgment; prompts give structure and consistency.
## MSA playbook input shape
The pack assumes a playbook with this shape (one entry per common clause):
```yaml
clause: limitation_of_liability
preferred:
position: "1× annual fees, exclude consequential damages, carve out IP indemnity / breach of confidentiality / data breach for 5×"
rationale: "Aligned with industry standard for SaaS at our scale; carves preserve recovery on highest-risk exposures."
acceptable:
position: "2× annual fees, similar carves"
rationale: "Acceptable when the deal warrants flexibility on cap level; carve preservation is non-negotiable."
walk_away:
position: "Liability cap below 1× fees, OR no carves on IP/confidentiality/data"
rationale: "Below 1× cap signals the firm cannot recover even direct damages on most matters; missing carves means breaches that should be uncapped become capped."
notes: "Counterparties often anchor at 12 months of fees; we counter to 'fees paid in 12 months prior to claim,' which is functionally the same but maps to our preferred."
```
If a clause isn't in the playbook, the pack flags it as "playbook does not cover" rather than asserting a position.
---
# Tier 1 — First-pass redline
## R1. Redline counterparty's MSA against firm playbook
```
Role: You are a contracts attorney redlining an MSA draft against the
firm's playbook.
Context: The MSA playbook is loaded as project knowledge. Read every
clause in the playbook before redlining.
Input: The counterparty's MSA draft.
Task: For each clause in the counterparty's draft, identify whether the
counterparty's position falls within the firm's preferred / acceptable /
walk-away range, OR is outside the playbook's coverage.
For each clause, output:
- Clause name (per playbook)
- Counterparty's position (1-2 sentence summary)
- Playbook tier (preferred / acceptable / walk-away / outside-playbook)
- Recommended redline language (if not preferred)
- Rationale citing the playbook entry
Things to avoid:
- Inventing positions not in the playbook.
- Generic "this should be tighter" without specific replacement language.
- Asserting a position when the playbook says walk-away (instead, flag
for counsel decision on whether to walk).
- Skipping clauses because "they're standard" — every clause is reviewed
against the playbook.
Output format: Markdown table with columns Clause | Counterparty | Tier |
Recommended redline | Rationale.
```
## R2. Identify missing clauses
```
Role: You are a contracts attorney comparing the counterparty's MSA draft
against the firm's standard MSA template.
Context: The firm's standard MSA template is in project knowledge.
Input: The counterparty's MSA draft.
Task: Identify clauses present in the firm's standard template but absent
from the counterparty's draft. For each missing clause, recommend an
insertion location and provide the language.
For each missing clause, output:
- Clause name
- Why the firm wants it (rationale from playbook)
- Recommended insertion location (which section / after which existing
clause)
- Replacement language
Things to avoid:
- Recommending insertion of clauses the firm's playbook tier as
"negotiable" if the counterparty's draft already covers the topic
differently — that's a redline to existing language, not an insertion.
- Recommending more than 5 insertions in one round; pick the top 5 by
risk, defer the rest.
Output format: Markdown list with the four bullets per missing clause.
```
---
# Tier 2 — Counter-position drafting
## C1. Draft counter-positions to counterparty's response
```
Role: You are a contracts attorney drafting the firm's counter-position
to the counterparty's response on prior redlines.
Context: Playbook in project knowledge. The firm's prior redlines are in
project knowledge. The counterparty's response is the input.
Input: The counterparty's response to the firm's redlines.
Task: For each clause where the counterparty pushed back on the firm's
redline, draft a counter-position. The counter-position is one of:
- Hold firm at preferred (if counterparty's pushback is unsupported)
- Move to acceptable (if counterparty's pushback has merit and the
move stays within playbook range)
- Trade for a concession on another clause (if the counter-position
needs leverage)
For each clause, output:
- Clause name
- Counterparty's pushback (1 sentence summary)
- Recommended counter (with specific replacement language if a redline)
- Rationale grounded in playbook
Things to avoid:
- Capitulating without naming what was traded (every move down the
playbook should name the trade or the rationale).
- Asserting positions that escalate beyond playbook walk-away.
- Generic "we accept your changes" or "we maintain our position" —
always name the specific clause language.
Output format: Markdown table.
```
## C2. Triage counterparty positions: accept vs push back
```
Role: You are a contracts attorney scoping the next negotiation round.
Context: Playbook in project knowledge.
Input: The counterparty's response.
Task: Triage every clause in the counterparty's response into one of
three buckets:
- ACCEPT — counterparty's position is within firm's preferred or
acceptable range; no further negotiation needed.
- PUSH BACK — counterparty's position is outside acceptable; the firm
should counter.
- ESCALATE — counterparty's position is at or past walk-away; the
contracts manager should escalate to counsel before further
negotiation.
Output: A triage table the contracts manager uses to scope this round.
For each clause, output:
- Clause name
- Bucket (accept / push back / escalate)
- One-sentence justification
Things to avoid:
- Treating every clause as push-back (that's how negotiations stall).
Accept the acceptable.
- Escalating clauses that fall within the playbook's walk-away
threshold but the firm has accepted in past deals (note the past
acceptance and recommend acceptance with rationale).
Output format: Markdown table grouped by bucket.
```
---
# Tier 3 — Fallback laddering
## F1. Per-clause fallback ladder
```
Role: You are a contracts attorney constructing a fallback ladder for
a stuck clause negotiation.
Context: Playbook in project knowledge.
Input: The clause name + the counterparty's current position + any prior
exchanges on this clause.
Task: Lay out the firm's fallback ladder from preferred to walk-away.
For each rung, name:
- The position (specific language)
- Rationale
- The "horse-trade question" — what would the counterparty have to
give to take this rung?
Things to avoid:
- Inventing a rung not in the playbook.
- Skipping rungs (every step should be a defensible move).
- Overly creative trades that don't reflect the firm's actual
priorities.
Output format: Markdown list, ordered preferred → acceptable → walk-away.
```
## F2. Cross-clause trade surfacer
```
Role: You are a contracts attorney looking for cross-clause trades to
unblock a stuck negotiation.
Context: Playbook + current state of all clauses in project knowledge.
Input: A note from the contracts manager describing the negotiation's
stuck state.
Task: Identify pairs of clauses where the firm has flexibility (could
move down the playbook ladder) AND the counterparty has flexibility
(based on their prior signals — e.g. they accepted firm's preferred on
clause X without resistance, suggesting they don't value that clause).
For each candidate trade, output:
- Clause the firm gives ground on
- Clause the counterparty gives ground on
- Why the trade balances (rationale grounded in both parties' apparent
priorities)
Things to avoid:
- Trades where the firm gives more than it gets (the trade is
one-sided).
- Trades that involve clauses outside playbook coverage (those need
counsel before being offered).
- Trades that involve walk-away rungs.
Output format: Markdown list of candidate trades.
```
---
# Tier 4 — Counterparty posture reading
## P1. Characterize counterparty's negotiation posture
```
Role: You are a contracts attorney characterizing the counterparty's
negotiation posture from their redlines.
Context: Counterparty's redlines in project knowledge.
Input: The counterparty's redline document.
Task: Characterize the posture as one of:
- Aggressive — large redlines on liability / IP / indemnification;
multiple walk-away-tier positions.
- Cooperative — limited redlines; willing to accept playbook-preferred
on most clauses; minor edits only.
- Conservative — defending standard terms with minor edits; not seeking
to change the firm's positions but holding their own.
- Uninformed — redlines that suggest the counterparty doesn't
understand the implications (e.g. requesting language that is
contradictory or ambiguously worded).
- Mixed — different postures across different clause categories.
For the chosen posture, output:
- Three pieces of evidence from the redlines
- The implication for the firm's negotiation approach
- The escalation signal (does this posture call for counsel review
before next round?)
Things to avoid:
- Reading posture from the counterparty's identity rather than their
redlines (a Fortune 500 company might be cooperative; a startup
might be aggressive).
- Missing mixed posture by collapsing to a single label.
Output format: Markdown.
```
## P2. Cross-round posture trend
```
Role: You are a contracts attorney tracking the counterparty's posture
across rounds.
Context: All prior rounds' redlines + posture characterizations in
project knowledge.
Input: The current round's redlines.
Task: Compare the current round's posture to prior rounds. Surface trends:
- Aggressive → cooperative trend (counterparty is closing; the firm
can hold positions).
- Cooperative → aggressive trend (counterparty has brought in outside
counsel mid-negotiation, OR something has changed in the deal
context — escalation risk).
- Stable cooperative or stable conservative (negotiation is progressing
normally).
- Stable aggressive (counterparty is committed to the positions;
consider whether the deal is workable).
For the trend, output:
- The trend label
- Three pieces of evidence comparing prior rounds
- The recommended response posture
Things to avoid:
- Inferring strategy without evidence (e.g. "they're trying to delay"
without redline patterns to support).
- Recommending escalation on every small posture shift.
Output format: Markdown.
```