A Claude Skill that turns a personal-injury case file — medical records, billing statements, wage-loss documentation, the police report, and the firm’s case notes — into a first-draft settlement demand letter. It builds a dated medical chronology with a record-page citation behind every entry, itemizes special damages from the bills, computes a general-damages range from the firm’s multiplier playbook, and assembles the letter from the firm’s template. Every figure in the draft traces to a source page; every claim the records do not support is listed under “insufficient data” instead of being written into the letter. The attorney edits and signs. The skill never sets the final demand number and never sends.
Attorney review is required. This skill drafts; it does not practice law. The general-damages range it produces is arithmetic on the firm’s own playbook, not a legal valuation. The attorney sets the demand figure, confirms every cited fact against the record, and owns the letter that goes out.
When to use
A plaintiff-side PI firm where paralegals assemble demand packages and a 6-12 page letter takes 3-6 hours of drafting on top of records review. The bottleneck is the first draft, not the legal judgment.
The firm has a written demand-letter template and a damages methodology (multiplier bands by injury severity, or a per-diem rate). The skill structures the draft around the firm’s method; it does not invent one.
Records are already collected and converted to text (OCR’d PDFs, or a records-retrieval export). The skill drafts from records that exist — it does not chase missing bills.
When NOT to use
Setting the demand number. The skill outputs a range from the multiplier playbook. The attorney sets the figure — the number is a strategic call that weighs venue, adjuster, liability strength, and policy limits the skill does not model.
Cases with disputed liability or comparative-fault complexity. When liability is the whole fight, the demand letter is an advocacy document the attorney drafts. The skill is for clear-liability soft-tissue-to-moderate cases where the work is assembling the record, not arguing causation.
Litigation filings. A demand letter is pre-suit. Complaints, motions, and discovery responses are court documents with their own rules — out of scope.
Medical opinion. The skill summarizes what the records say (diagnosis, treatment, provider impressions). It does not infer prognosis, causation, or permanency beyond what a provider wrote. Those are medical opinions; they come from the treating physician or a retained expert.
Records that contain PHI you cannot process under your agreements. Protected health information runs through this skill. Confirm your Claude deployment and your client-engagement terms cover it before any record goes in (see Watch-outs).
Setup
Drop the bundle. Place apps/web/public/artifacts/demand-letter-drafter-claude-skill/SKILL.md into your Claude Code skills directory along with the references/ folder.
Author the firm’s demand template. Replace the skeleton in references/1-demand-letter-template.md with the firm’s actual letter — section order, boilerplate, the demand-and-deadline paragraph, the signature block. The skill assembles into this template; it is the firm’s voice, not the skill’s.
Encode the damages playbook. Fill references/2-damages-playbook.md with the firm’s methodology: multiplier bands by injury severity (the common range is 1.5-5× specials; severe or disfiguring injuries run higher), the per-diem rate if the firm uses one, and any jurisdiction-specific caps or rules the firm tracks. Without this file the skill has no basis for a general-damages range and will refuse to produce one.
Set the chronology schema.references/3-medical-chronology-schema.md defines the columns (date of service, provider, treatment, billed amount, record page) and the citation format. Keep the citation format the firm uses so the attorney can verify against the same page references.
Dry-run on three closed cases. Run the skill on three settled files where the demand is known. Compare its special-damages total to the actual specials, its chronology to the paralegal’s, and its general-damages range to what the firm demanded. Tune the playbook bands until the range brackets the historical demand.
What the skill does
Seven steps. The medical chronology is built before the damages math, because every dollar of specials and every line of the injuries narrative cites a chronology entry, and every chronology entry cites a record page.
Index the record set. Ingest the records, bills, wage documentation, police report, and case notes. Build a page index so every later citation resolves to a specific document and page. Files that are unreadable (failed OCR, image-only) are listed, not silently skipped.
Build the dated medical chronology. One row per date of service: provider, complaint, treatment, billed amount, and the record page it came from. Gaps in treatment are flagged (a 6-week gap between the ER visit and the first follow-up is a fact the adjuster will use; the attorney should know it before the letter goes out, not after).
Itemize special damages. Total the medical bills by provider, add documented wage loss, and add out-of-pocket costs. Each line cites its bill page. The skill flags bills referenced in the records but missing from the file, and liens or balances it cannot reconcile — these go to the attorney as “confirm before sending,” not into the total as a guess.
Draft the liability and facts narrative. From the police report and case notes, draft the how-it-happened section. Statements the report does not support are flagged, not written. Where the report and the client’s account differ, the skill surfaces the conflict rather than picking one.
Draft the injuries and treatment narrative. Grounded in the chronology — diagnosis, course of treatment, provider impressions, current status as the records state it. No prognosis or permanency language unless a provider wrote it; if the firm wants a permanency argument, that is flagged as needing the treating physician’s statement.
Compute the general-damages range. Apply the playbook’s multiplier band to the specials (or the per-diem method, per the playbook). Output is a range with the band shown — “$X-$Y at 2.5-4× specials of $Z” — explicitly labelled a starting point. The skill does not pick a point in the range.
Assemble the draft and the review checklist. Fill the firm’s template. Produce a paired review checklist: every figure with its source citation, every flagged gap, every “insufficient data” item, and the response-deadline line left as a placeholder for the attorney to set. The letter is a draft; the checklist is how the attorney verifies it in minutes instead of re-reading the file.
Cost reality
LLM tokens — a full case file is 40-120k input tokens (records, bills, report) and 4-8k output (chronology + draft). At Claude Sonnet rates that is roughly $0.30-0.90 per demand; on a heavy record set with Opus, $1.50-4. Pennies-to-dollars against the drafting hours it offsets.
Paralegal / attorney time — the win. A demand package is commonly cited at 3-6 hours of paralegal drafting. The skill produces the chronology and first draft in minutes; the paralegal verifies and fills gaps, the attorney edits and sets the number. The realistic saving is the drafting half, not the review half — call it 2-4 hours per demand.
Setup time — 45 minutes plus the firm’s template and playbook authoring (a one-time hour or two that pays back across every case).
Success metric
Draft-to-attorney time — wall-clock from “records collected” to “draft on the attorney’s desk.” Should fall from 1-2 days (in a paralegal’s queue) to same-day.
Specials reconciliation rate — share of demands where the skill’s special-damages total matches the verified total without correction. Aim above 90%; misses point at a records-indexing or bill-parsing gap, not a reason to abandon the draft.
Attorney edit depth — how much of the draft survives to the signed letter. Track it; if the narrative is being rewritten wholesale, the template or the case-selection criteria need tightening, not the skill.
vs alternatives
vs EvenUp Express Demands and similar managed services. EvenUp pairs AI drafting with human drafters and delivers a finished package; marketed turnaround is 1-24 hours, though firms report multi-day delivery in practice. The trade is control and data residency: the managed service holds your PHI and owns the playbook; this skill runs on your own Claude with your template and your damages method, and the draft never leaves your environment. Use the managed service when you want the package handed back finished and are comfortable with the data and turnaround terms; use the skill when you want the draft in-house, today, on your own methodology.
vs Eve and full case-workup platforms. Eve and the litigation-AI platforms work the whole case, not just the demand. If the firm wants case valuation, treatment-gap analytics, and litigation prep as a suite, that is a platform purchase. The skill is one step of that pipeline — the demand draft — for firms not ready to buy the platform.
vs the paralegal drafting from scratch. The status quo. It produces the firm’s voice and judgment but is the slowest path and the one that varies most by who drafted it. The skill standardizes the first draft and the citation discipline; the paralegal’s judgment moves to verification and the edge cases.
Watch-outs
Hallucinated medical facts or figures. A demand letter with an invented diagnosis or an inflated bill is a credibility loss the adjuster will exploit, and a sanctions risk if it reaches a filing. Guard: every chronology entry, every special-damages line, and every factual claim cites a record page; anything the skill cannot cite goes to the “insufficient data” list, not into the letter. The attorney verifies against the cited pages.
The multiplier range read as a valuation. The general-damages band is arithmetic on the firm’s playbook, blind to venue, adjuster history, liability strength, and policy limits. Guard: the output is a labelled range with the band shown, never a single number; the attorney sets the demand.
PHI exposure. The skill processes medical records — protected health information. Guard: confirm the Claude deployment’s data terms (zero-retention / no-training posture, and a BAA where your compliance program requires one) and your client-engagement terms cover AI processing before any record goes in. Run on a deployment that meets the firm’s HIPAA program; do not paste records into a consumer endpoint.
Treatment-gap and liability conflicts buried. A draft that reads clean but omits a treatment gap or a report-vs-client conflict sets the attorney up to be surprised by the adjuster. Guard: gaps and conflicts are surfaced in the review checklist as explicit flags, not smoothed over in the narrative.
Missing bills counted as zero. If a provider’s bill is referenced but absent from the file, totaling only what is present understates specials and the demand. Guard: referenced-but-missing bills are flagged “confirm before sending” and excluded from the total rather than guessed; the paralegal retrieves them before the demand is finalized.
Stale or jurisdiction-wrong playbook. Damages caps, fee rules, and pre-suit notice requirements vary by jurisdiction and change. Guard: the playbook file is the single source the skill cites for its bands; review it against current rules per matter, and treat caps and notice requirements as attorney-confirmed, not skill-asserted.
Stack
The bundle lives at apps/web/public/artifacts/demand-letter-drafter-claude-skill/:
SKILL.md — the skill definition
references/1-demand-letter-template.md — the firm’s demand-letter skeleton to replace with the firm’s actual template
references/3-medical-chronology-schema.md — the chronology table schema and citation format
Tools: Claude. For the structured-extraction pattern behind the chronology step, see contract data extraction. Adjacent tools in the PI cluster: EvenUp, Eve.
---
name: demand-letter-drafter
description: Draft a first-pass personal-injury settlement demand letter from a case file (medical records, bills, wage-loss docs, police report, case notes). Build a cited medical chronology, itemize special damages, compute a general-damages range from the firm's playbook, and assemble the letter into the firm's template. Every figure cites a record page; unsupported claims go to an "insufficient data" list. Attorney review required — never sets the demand number, never sends.
---
# Personal-injury demand letter drafter
## When to invoke
Use this skill when a paralegal or attorney has a collected, text-readable PI case file and wants a first-draft demand letter — a cited medical chronology, itemized special damages, a general-damages range, and a letter assembled into the firm's template.
Do NOT invoke this skill for:
- **Setting the final demand number** — output is a range; the attorney sets the figure.
- **Disputed-liability or comparative-fault cases** — the demand is an advocacy document the attorney drafts; this skill is for clear-liability assembly work.
- **Litigation filings** — complaints, motions, discovery responses are court documents, out of scope.
- **Medical opinion** — the skill reports what providers wrote; it does not infer prognosis, causation, or permanency.
**Attorney review is required on every draft.** The skill assembles and cites; the attorney verifies every cited fact, sets the demand number, and signs.
## Inputs
- Required: `case_file_path` — directory of text-readable records (OCR'd PDFs, exports). Bills, medical records, wage-loss docs, police report, case notes.
- Required: `template` — path to the firm's demand-letter template (`references/1-demand-letter-template.md` is the starting skeleton).
- Required: `damages_playbook` — path to the firm's damages methodology (`references/2-damages-playbook.md`).
- Optional: `chronology_schema` — path to the chronology schema (`references/3-medical-chronology-schema.md`); defaults to the bundled schema.
- Optional: `client_account` — the client's written/recorded account of the incident, for the facts-conflict check.
## Reference files
- `references/1-demand-letter-template.md` — the firm's demand-letter skeleton.
- `references/2-damages-playbook.md` — multiplier bands, per-diem rate, jurisdiction notes.
- `references/3-medical-chronology-schema.md` — chronology columns and citation format.
## Method
Seven steps. Chronology before damages, because every special-damages line and every injuries-narrative sentence cites a chronology entry, and every chronology entry cites a record page.
### 1. Index the record set
Ingest every file. Build a page index keyed by document name + page so later citations resolve. List unreadable files (failed OCR, image-only, password-locked) in an `unprocessed` block — never silently skip them. Stop and report if the bills or the medical records are entirely absent; there is no demand to draft without them.
### 2. Build the dated medical chronology
One row per date of service, per the chronology schema:
| Date | Provider | Complaint / treatment | Billed | Record page |
- Sort by date.
- Flag treatment gaps over the playbook's gap threshold (default 30 days between consecutive visits) — adjusters use gaps to argue the injury resolved.
- Flag any entry where the date, provider, or billed amount is unreadable in the source — cite the page and mark `[illegible — confirm]`.
### 3. Itemize special damages
- **Medical specials** — total billed by provider, citing each bill page. Use billed amounts, not paid/adjusted, unless the playbook says otherwise (jurisdictions differ on the collateral-source and billed-vs-paid question — the playbook controls).
- **Wage loss** — from the documentation (employer letter, pay stubs, tax records), citing the page.
- **Out-of-pocket** — mileage, devices, co-pays, with citations.
- Flag bills referenced in records but missing from the file as `missing — retrieve before sending`. Exclude them from the total. Do not estimate a missing bill.
- Flag liens and unreconciled balances for attorney confirmation.
### 4. Draft the liability and facts narrative
From the police report and case notes, draft how the incident happened. Statements the report does not support are flagged, not written. If `client_account` is provided and it conflicts with the report, surface the conflict — do not pick one side.
### 5. Draft the injuries and treatment narrative
Grounded in the chronology: diagnosis, course of treatment, provider impressions, current status as the records state it. No prognosis, permanency, or causation language beyond what a provider wrote. If the firm wants a permanency argument, emit a flag: `permanency argument requires treating-physician statement — not in record`.
### 6. Compute the general-damages range
Per the playbook:
- **Multiplier method** — apply the severity band to the medical specials. Output `range_low = band_low × specials`, `range_high = band_high × specials`. Show the band.
- **Per-diem method** — if the playbook uses it, multiply the daily rate by the documented recovery period.
Output a labelled range, never a single number:
```
General damages (starting point — attorney sets the figure):
$X – $Y (2.5–4.0× medical specials of $Z, per playbook §severity-moderate)
```
### 7. Assemble the draft and the review checklist
Fill the firm's template. Leave the response-deadline line as a placeholder (`[ATTORNEY: set response deadline — commonly 30 days]`). Produce the review checklist alongside the letter (see Output format) so the attorney verifies in minutes.
## Output format
Two artifacts: the draft letter (firm template, filled) and the review checklist below.
```markdown
# Demand draft — {matter_id} — REVIEW CHECKLIST
⚠️ DRAFT. Attorney sets the demand number and confirms every cited fact. Not sent.
## Figures and citations
| Item | Amount | Source page |
|---|---|---|
| Medical specials — City ER | $4,210 | bills.pdf p.3 |
| Medical specials — Apex PT | $6,880 | bills.pdf p.7-9 |
| Wage loss | $3,100 | wages.pdf p.1 |
| **Special damages total** | **$14,190** | (sum, verified) |
| General damages range | $35,000–$57,000 | playbook §severity-moderate (2.5–4.0×) |
## Flags — resolve before sending
- ⚠️ Treatment gap: 41 days between ER (03-04) and first PT (04-14). records.pdf p.12.
- ⚠️ Missing bill: imaging referenced (records.pdf p.15) — no bill in file. Retrieve.
- ⚠️ Lien: hospital lien noted (records.pdf p.4) — confirm amount.
- ⚠️ Facts conflict: report says "stopped"; client account says "slowing." report.pdf p.2.
## Insufficient data — not written into the letter
- Permanency: no treating-physician permanency statement in record.
- Future medical: no life-care plan or future-treatment estimate in record.
## Provenance
- Case file: {path} ({N} files indexed, {M} unprocessed)
- Template: {path}
- Playbook: {path} §{section}
- Drafted: {ISO timestamp} · Skill v1.0 · {model}
```
## Watch-outs
- **Hallucinated facts/figures.** *Guard:* every figure and factual claim cites a record page; uncitable claims go to "insufficient data," never into the letter.
- **Range read as a valuation.** *Guard:* general damages is always a labelled range with the band shown; the attorney sets the number.
- **PHI exposure.** *Guard:* run only on a Claude deployment whose data terms (zero-retention, BAA where required) meet the firm's HIPAA program. Never a consumer endpoint.
- **Missing bills counted as zero.** *Guard:* referenced-but-absent bills are flagged and excluded, not estimated.
- **Jurisdiction-wrong playbook.** *Guard:* caps, fee rules, and notice requirements are attorney-confirmed; the skill cites the playbook, it does not assert the law.
# Demand-letter template (firm skeleton — replace with the firm's actual letter)
This is a starting skeleton. Replace it with the firm's real demand letter so the skill assembles into the firm's voice, section order, and boilerplate. Keep the `{{placeholder}}` tokens — the skill fills them. Everything else is the firm's prose.
The skill assembles into this template. It does not invent sections. If the firm's letter has a section this skeleton lacks (e.g. a punitive-damages paragraph, a UM/UIM section), add it here with its own placeholders.
---
**{{firm_letterhead}}**
{{date}}
{{adjuster_name}}
{{insurer_name}}
{{insurer_address}}
**RE:** Claim of {{client_name}} · Claim No. {{claim_number}} · Date of loss: {{date_of_loss}} · Your insured: {{insured_name}}
Dear {{adjuster_name}},
This firm represents {{client_name}} for injuries sustained in the {{incident_type}} of {{date_of_loss}}. This letter presents our client's claim and our demand for settlement.
## 1. Facts and liability
{{facts_narrative}}
<!-- Skill fills from police report + case notes. Statements unsupported by the report are flagged in the checklist, not written here. Report-vs-client conflicts are surfaced, not resolved. -->
## 2. Injuries and treatment
{{injuries_narrative}}
<!-- Skill fills from the medical chronology. Diagnosis, course of treatment, provider impressions, current status as the records state it. No prognosis/permanency/causation beyond what a provider wrote. -->
### Treatment chronology
{{chronology_table}}
<!-- The dated medical chronology, per references/3-medical-chronology-schema.md. -->
## 3. Damages
### Special damages
{{special_damages_itemization}}
<!-- Itemized, each line citing its source page. Missing/unreconciled items are flagged in the checklist and excluded here. -->
**Total special damages: {{special_damages_total}}**
### General damages
{{general_damages_narrative}}
<!-- The pain-and-suffering / non-economic narrative. The amount is the attorney's; the skill provides the range in the checklist, not in this paragraph, unless the firm's template states a number here (in which case the attorney inserts it). -->
## 4. Demand
Based on the foregoing, {{client_name}} demands **{{demand_amount}}** to resolve this claim in full.
<!-- {{demand_amount}} is left for the attorney. The skill never fills it. -->
This demand remains open until **{{response_deadline}}**, after which we reserve all rights, including filing suit.
<!-- {{response_deadline}}: attorney sets it. Commonly 30 days; jurisdiction and pre-suit notice rules vary — confirm. -->
Please direct all communication regarding this claim to this office.
Sincerely,
{{attorney_name}}
{{firm_name}}
**Enclosures:** {{enclosures_list}}
<!-- Medical records, bills, wage documentation, photographs, etc. The skill lists what it indexed; the attorney confirms the production. -->
---
## How to adapt this template
1. Paste the firm's actual demand letter over this skeleton.
2. Mark every spot the skill should fill with a `{{placeholder}}` token. The tokens above are the minimum set; add more (UM/UIM, future medicals, lien summary) as the firm's letter requires.
3. Leave `{{demand_amount}}` and `{{response_deadline}}` as placeholders the attorney fills — the skill is configured never to set them.
4. Keep boilerplate (reservation of rights, confidentiality, FRE 408 "for settlement purposes only" legend if the firm uses one) as fixed prose, not placeholders.
# Damages playbook (firm methodology — fill before first use)
The skill computes the general-damages range from this file and nothing else. If a band is not here, the skill refuses to produce a range rather than guessing. Fill every section with the firm's actual methodology and review it against current law per matter.
**This is methodology, not legal advice.** Multiplier bands, caps, and notice rules vary by jurisdiction and change. The attorney confirms the numbers in this file against current rules before they drive a demand.
## 1. Method selection
Choose the firm's default and when it switches:
- **Multiplier method** (default for most soft-tissue to moderate cases): general damages = band × medical specials.
- **Per-diem method** (use when the recovery period is long and well-documented, or when the firm prefers it for a venue): general damages = daily rate × documented recovery days.
State the rule for which method applies to which case type. The skill follows it.
## 2. Multiplier bands by injury severity
Common practice puts the multiplier in the 1.5–5× range, higher for severe or disfiguring injuries. Replace these placeholders with the firm's calibrated bands:
| Severity tier | Definition (firm's criteria) | Band (× specials) |
|---|---|---|
| Minor | Soft-tissue, full resolution < 90 days, no injections/surgery | 1.5 – 2.0 |
| Moderate | Soft-tissue with injections or extended PT, resolution 3–9 months | 2.5 – 4.0 |
| Serious | Fracture, surgery, or documented lasting limitation | 4.0 – 5.0 |
| Severe / catastrophic | Permanent impairment, disfigurement, TBI, spinal | attorney-set; multiplier insufficient |
For the severe/catastrophic tier the skill does NOT output a multiplier range — it flags the case as requiring attorney valuation. A multiplier on specials understates these.
## 3. Per-diem rate
- Daily rate: `${{per_diem_rate}}` (the firm's chosen rate; some firms anchor to the client's daily wage, others to a fixed figure — state which and why).
- Recovery period: documented start (date of loss) to documented end (release from care / maximum medical improvement as the records state it). The skill uses documented dates only — it does not estimate an end date.
## 4. Billed vs. paid
State the firm's position, because it changes the specials total the multiplier runs on:
- [ ] Use **billed** amounts (gross charges).
- [ ] Use **paid/adjusted** amounts.
- [ ] Per jurisdiction — note the rule and the venues where each applies.
Collateral-source rules and billed-vs-paid admissibility vary by jurisdiction. This is an attorney call; the skill applies whatever this section states.
## 5. Jurisdiction notes (attorney-maintained)
The skill cites this section; it does not assert the law. Maintain per venue the firm practices in:
- **Damages caps** — non-economic caps, if any, and the case types they apply to.
- **Pre-suit notice** — required notice periods (e.g. claims against government entities), and the deadline math.
- **Comparative fault** — the rule (pure / modified / contributory) and how the firm reflects it in the demand.
- **Statute of limitations** — for the deadline-awareness flag, not for legal advice.
## 6. Gap threshold
- Treatment-gap flag threshold: `{{gap_days}}` days (default 30). Gaps at or above this are flagged in the checklist for the attorney.
## How to update
Review §2 and §5 at least annually and whenever the firm enters a new venue. The skill's output is only as current as this file; a stale playbook produces a stale range.
# Medical chronology schema (citation format the attorney verifies against)
The chronology is the spine of the demand. Every special-damages line and every sentence of the injuries narrative cites a chronology row; every chronology row cites a record page. Keep the citation format identical to what the firm uses so the attorney verifies against the same references.
## Columns
| Column | Content | Rule |
|---|---|---|
| `Date` | Date of service (YYYY-MM-DD) | Sort ascending. Illegible date → `[illegible — confirm]` + page cite. |
| `Provider` | Treating provider / facility | As named in the record. No normalization that loses the source name. |
| `Complaint / treatment` | The visit's complaint and what was done | From the record only. No inferred diagnosis. |
| `Billed` | Billed amount for the date of service | Per the playbook's billed-vs-paid rule. `[no bill in file]` if records reference it but the bill is absent. |
| `Record page` | `document.pdf p.N` | Every row. A row with no citation is invalid — drop it to the checklist instead. |
## Citation format
`{document_name} p.{page}` — e.g. `records-apex-pt.pdf p.7`. For a page range: `bills.pdf p.7-9`. The skill reuses the exact filenames from the indexed case file so the attorney opens the same document.
## Worked row
| Date | Provider | Complaint / treatment | Billed | Record page |
|---|---|---|---|---|
| 2026-03-04 | City General ER | Neck/back pain post-MVA; X-ray, discharged | $4,210 | bills.pdf p.3 |
| 2026-04-14 | Apex Physical Therapy | Eval + start of PT, 2×/week | $880 | records-apex.pdf p.2 |
## Gaps and flags
- A gap between consecutive `Date` rows at or above the playbook's gap threshold (default 30 days) is flagged in the review checklist with both dates and the page cites bracketing the gap.
- An entry the skill cannot fully read (illegible billed amount, ambiguous provider) is kept in the chronology with the legible fields, the unreadable field marked `[illegible — confirm]`, and the page cited.
## What does NOT go in the chronology
- Prognosis, permanency, causation — unless a provider wrote it, and then it is quoted, not paraphrased into a stronger claim.
- Bills with no corresponding date of service in the records — those go to the checklist as `bill — no matching service date — confirm`, not into a chronology row.