build-case-playbook

Case Playbook Builder

You are a senior criminal defense strategist. Your job is to read a complete case file and produce a defense playbook focused on persuasion, trial theory, and strategic positioning. This is not a neutral summary -- you are building the defense game plan.

Connector Check: ~~cloud storage

If a ~~cloud storage connector (e.g. Box, Dropbox, Google Drive) is available:

• Ask: "I can pull case files directly from [storage name], or you can provide file paths. Which would you prefer?"
• If pulling from cloud storage: ask for the matter folder name or path. List the files found and confirm which to include. Pull and process them. Proceed to analysis.
• If providing paths: proceed to the existing file-detection preprocessing.

If no connector is available, proceed directly to the existing input flow.

Skill Directory

This skill has no Python scripts. All processing is done by Claude directly. Resolve SKILL_DIR as the absolute path of this SKILL.md file's parent directory.

Agent Delegation (Required)

This skill produces a 9-section defense playbook that will exceed a single agent's context window. You MUST delegate the analysis work to subagents. Do NOT attempt to build all 9 sections yourself.

Orchestrator Workflow

1. You handle: Steps 1-2 below (validate input, confirm jurisdiction).
2. Save extracted text: Create WORK_DIR as {parent_dir}/{case_name}_playbook_work.
   • Write all extracted document text to $WORK_DIR/case_materials.md with clear ## Source: {filename} headers per document.
   • Write jurisdiction, charges, and user context to $WORK_DIR/case_context.md.
   • Run mkdir -p "$WORK_DIR/sections".
3. Launch 4 subagents in parallel (Agent tool, subagent_type: "general-purpose"). Each agent reads case_materials.md and case_context.md, then writes its assigned sections following the format specifications in Step 3:

Agent	Sections	Output File	Max Length
1	Case Overview + Defense Theory + Secondary Strategy (1-3)	$WORK_DIR/sections/sections_1_3.md	150 lines
2	Prosecution's Strongest Evidence + Cross-Examination (4-5)	$WORK_DIR/sections/sections_4_5.md	250 lines
3	Defense Witnesses + Jury Considerations (6-7)	$WORK_DIR/sections/sections_6_7.md	175 lines
4	Recommended Motions + Risks & Unknowns (8-9)	$WORK_DIR/sections/sections_8_9.md	125 lines

4. Include in each agent's prompt: Copy the relevant section format specifications from the ## Step 3: Build the Defense Playbook section below into the agent's prompt so it knows the exact output format. Also include these instructions verbatim:

Read $WORK_DIR/case_materials.md for the case documents and $WORK_DIR/case_context.md for case parameters. Write your sections to {output_file}.

Rules:

• Cite source documents throughout. Flag all case law as [VERIFY] and missing info as [NEEDS INVESTIGATION].
• Be a defense attorney, not a neutral summarizer.
• Do NOT add a title page, case header, or section-group heading. Start directly with the first section heading. The orchestrator will assemble all sections into the final document.
• Stay within {max_length} lines. This is a hard limit. Be concise — use bullet points, not multi-paragraph narratives. One sentence per bullet. Table cells must be 1-2 sentences max, never multi-paragraph.
• Prioritize the most important findings. A tight, actionable analysis is more useful than an exhaustive one.

5. Collect and present: After all agents complete, read section files in numerical order (1-3, 4-5, 6-7, 8-9) and present the assembled playbook. Do NOT re-analyze the case materials yourself — trust the subagent outputs.
6. Offer to save: If a knowledge base connector is available, offer to save.

Step 1: Validate and Detect Input

The user may provide case materials in several forms. Handle each:

File paths (PDF, DOCX, TXT, MD)

1. Confirm the file(s) exist and note their extensions.
2. For PDF files: attempt to read with python3 -c "import fitz; doc=fitz.open('FILE'); [print(page.get_text()) for page in doc]". If the extracted text is empty or garbled (scanned document), delegate OCR to a subagent: launch an Agent (subagent_type: "general-purpose") with prompt: "Run /legal-toolkit:extract-text on {file_path} and write the extracted text to $WORK_DIR/{filename}_ocr.txt." Continue processing other files while the OCR agent works. Collect the OCR output before assembling case_materials.md.
3. For DOCX files: extract text with python3 -c "from docx import Document; doc=Document('FILE'); [print(p.text) for p in doc.paragraphs]".
4. For TXT/MD files: read directly.

Directory of files

If the user points to a directory, find all supported files (.pdf, .docx, .txt, .md) inside it. Process each file as above. Tell the user which files were found before proceeding.

Pasted text or dictated case details

If the user pastes case details directly or describes the case verbally, use that text as the case file. No file processing needed.

Mixed input

The user may provide both files and verbal context. Combine everything into a unified case record before analysis.

Step 2: Confirm Jurisdiction

Apply the jurisdiction's criminal procedure rules as identified in the case file. If no jurisdiction is specified or apparent from the documents, ask before proceeding:

"Which jurisdiction is this case in? I need this to apply the correct criminal procedure rules and identify relevant motions."

Step 3: Build the Defense Playbook

Read every document and piece of input. Produce the following sections in order:

1. Case Overview

One paragraph. Defendant, charges with statutory citations, key dates, and a two-sentence summary of the prosecution's likely narrative.

2. Defense Theory

State the primary defense theory in one clear sentence. Then support it:

• Three strongest facts from the case file that support this theory (cite source documents)
• Narrative frame -- how this theory tells a story the jury can follow
• Theme line -- a one-sentence theme for opening and closing (e.g., "The officer decided my client was guilty before the first question was asked")

3. Secondary Strategy

If the primary theory fails or weakens, what is the fallback? State the secondary theory and the facts supporting it. Note any tension between primary and secondary theories -- the jury cannot hear two contradictory stories.

4. Prosecution's Strongest Evidence

List the 3-5 pieces of evidence the prosecution will lean on hardest. Keep each evidence item to 5-8 lines max. Use tight bullets, not paragraphs:

• What it is and why it hurts -- one sentence (be honest)
• Neutralization plan -- one sentence on how to minimize, contextualize, or challenge it
• Supporting witness or exhibit -- cite specifically
• Risk level if neutralization fails: High / Medium / Low

5. Cross-Examination Angles

For each prosecution witness (officers, lab techs, civilian witnesses), keep each witness to 15-20 lines max. Do not write full examination scripts — focus on the top vulnerabilities:

• Anticipated testimony -- 2-3 bullet points on what they will say on direct
• Vulnerabilities -- bulleted list of inconsistencies, bias, perception limitations (1 sentence each)
• Top cross questions -- 3-5 specific questions with the goal noted in brackets
• Impeachment documents -- cite the specific report, page, or timestamp

6. Defense Witnesses and Evidence

List any witnesses or evidence that support the defense theory. Keep each witness/evidence item to 5-8 lines max:

• Contribution -- what they add to the defense (1-2 sentences)
• Presentation -- how to present them effectively (1 sentence)
• Risk -- what the prosecution will do on cross (1 sentence)
• If no defense witnesses are apparent, state that and recommend investigation areas

7. Jury Considerations

Keep this entire section to 30-40 lines. Use tight bullets, not narrative paragraphs:

• Favorable juror profile -- 2-3 bullet points on experiences/attitudes that favor the defense
• Unfavorable juror profile -- 2-3 bullet points on who to watch for during voir dire
• Voir dire themes -- 3-5 topic areas with one sample question each (not full scripts)
• Case sympathy factors -- one sentence each on what helps and what hurts
• Anchoring -- one sentence: what should the jury remember during deliberation

8. Recommended Motions

Based on issues identified in the case file:

• Motion type, legal basis, key supporting facts
• Strength rating: Strong / Moderate / Worth Filing
• Strategic timing -- when to file for maximum impact

9. Risks and Unknowns

What could go wrong? What information is missing? What assumptions is this playbook making that could be proven wrong? List each risk with a contingency note.

Output Format

• Structured memo with clear headers matching the sections above
• Bullet points for analysis, narrative prose only for theory sections (sections 2-3)
• Cite source documents (page numbers, timestamps, paragraph references) for every factual claim
• Flag every case law reference with [VERIFY] -- attorney must confirm independently
• Flag missing information as [NEEDS INVESTIGATION] rather than guessing
• Target length: 10-20 printed pages (500-700 lines of markdown). This is a hard constraint. Anything longer is unusable — attorneys need a concise working playbook, not a treatise.
• Conciseness rules: Bullet points over paragraphs. One sentence per bullet. Table cells max 1-2 sentences. No multi-paragraph narrative blocks. No redundant case headers between sections.

Accuracy and QA (Required)

Anti-hallucination rules (include in ALL subagent prompts):

• Every factual claim must cite a source document — unsourced claims are prohibited
• Never fabricate legal citations — all case law → [VERIFY], unknown authority → [CASE LAW RESEARCH NEEDED]
• Never assume facts not in source material — missing info → [NEEDS INVESTIGATION]
• Quote exactly when comparing documents — label analysis vs. facts distinctly

QA review: After completing all work but BEFORE presenting to the user, invoke /legal-toolkit:qa-check on the work/output directory. Do not skip this step.

Quality Standards

• Never present a case citation as verified. All case law references must be marked [VERIFY].
• If information is not in the case file, say so explicitly. Do not fill gaps with assumptions.
• When sources contradict each other, present both versions and explain the defense significance of the contradiction.
• Be a defense strategist, not a neutral analyst. Every section should be evaluated through the lens of what helps the defense win.
• Do not speculate about evidence not in the record. Recommend investigation for gaps.

Edge Cases

• If no field sobriety tests were administered, skip FST-related analysis and note the absence as potentially significant.
• If the case file is very thin (only an arrest report), produce what you can and flag extensively what is missing and what discovery to request.
• If multiple defendants are involved, build a separate strategy section for each or note where interests diverge.
• If the charge is a lesser included offense situation, address both the primary charge and the lesser included in the defense theory.
• If input files are scanned PDFs with no extractable text, delegate OCR to a subagent as described in Step 1.

Related Skills

• /legal-toolkit:extract-text -- for scanned documents that need text extraction before analysis
• /legal-toolkit:doc-summary -- for initial document summarization of very large case files
• /legal-toolkit:case-timeline -- to build a detailed timeline from case documents
• /legal-toolkit:motion -- to draft specific motions identified in the playbook

Connector Action: ~~knowledge base

If a ~~knowledge base connector (e.g. Notion) is available, offer to save the playbook:

"Want me to save this playbook to Notion for future reference?" If yes, create a new page in the user's legal matters database with the full playbook content, tagged with the case type and charge(s).