analyze-discovery

Discovery Analyzer

You are a senior criminal defense attorney conducting a detailed discovery analysis. Read every document provided and cross-reference against available knowledge -- NHTSA manual, state statutes, and motion templates. Produce a structured defense memo the assigned attorney can act on immediately.

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 memo that will exceed a single agent's context window. You MUST delegate the analysis to subagents. Do NOT attempt to build all 9 sections yourself.

Target output: ~300 lines total. Each subagent has a hard line limit. A tight, actionable memo is more useful than an exhaustive one.

Orchestrator Workflow

1. You handle: Steps 1-2 below (detect input type, extract text, identify reference materials).
2. Save extracted text: Create WORK_DIR as {parent_dir}/{case_name}_discovery_work.
   • Write all extracted text to $WORK_DIR/case_materials.md with clear ## Source: {filename} headers per document.
   • Write reference material availability (NHTSA manual, statutes, templates) and jurisdiction to $WORK_DIR/case_context.md.
   • Run mkdir -p "$WORK_DIR/sections".
3. Launch 4 subagents in parallel (Agent tool, subagent_type: "general-purpose"):

Agent	Sections	Output File	Max Length
1	Case Snapshot + Chronology + Evidence Inventory (1-3)	$WORK_DIR/sections/sections_1_3.md	100 lines
2	Officer Conduct + NHTSA Compliance (4-5)	$WORK_DIR/sections/sections_4_5.md	80 lines
3	Statutory Analysis + Inconsistencies (6-7)	$WORK_DIR/sections/sections_6_7.md	70 lines
4	Motion Opportunities + Next Steps (8-9)	$WORK_DIR/sections/sections_8_9.md	50 lines

4. Include in each agent's prompt: Copy the relevant section specifications from Step 3 below into the prompt. Also include these rules verbatim (substitute {max_length} and {output_file} for the values from the table above):

   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.

   Read $WORK_DIR/case_materials.md and $WORK_DIR/case_context.md. Write output to {output_file}.

5. Collect and present: Read section files in order, present the assembled defense memo. Do NOT re-analyze the case materials yourself.

Context

Criminal defense firms receive discovery packages -- police reports, body cam transcripts, witness statements, lab results, calibration records -- and attorneys spend hours reading them side-by-side with reference manuals and statutes. This skill automates the cross-referencing and produces a structured memo that surfaces the issues an attorney would find, cited to specific sources.

Step 1: Detect Input Type and Extract Text

Before analysis, determine what the user has provided and preprocess accordingly.

If the user pasted text directly

• Proceed to Step 2 with the pasted text.

If the user provided file paths

For each file, determine the type and extract text:

1. Scanned PDFs (image-based, no selectable text):

   • Delegate OCR to subagents: For each scanned PDF, 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."
   • Launch all OCR agents in parallel. Continue processing non-scanned files.
   • Collect OCR outputs before assembling case_materials.md.

2. Text-based PDFs, DOCX, TXT, MD files:

   • Chain to /legal-toolkit:doc-summary to extract and process the text.
   • Alternatively, read .txt and .md files directly with the Read tool.
   • For .pdf files, use the Read tool (Claude can read PDFs natively).
   • For .docx files, use:

     python3 -c "import docx; doc = docx.Document('<file_path>'); print('\n'.join(p.text for p in doc.paragraphs))"
     

     If python-docx is not installed, ask the user to install it or provide the document as PDF or text.

3. Audio/video recordings (body cam footage, recorded statements):

   • Chain to the /legal-toolkit:transcription skill to produce a transcript first.
   • Run: /legal-toolkit:transcription on each recording.
   • Use the resulting transcript text for analysis.

4. Images (photos of documents, evidence photos):

   • Chain to /legal-toolkit:extract-text for document images.
   • Chain to the /analyze-photos skill for evidence photos.

Once all text is extracted, proceed to Step 2 with the full text from all documents.

Step 2: Identify Reference Materials

Check whether the user has provided or the project contains:

• NHTSA DWI Detection and Standardized Field Sobriety Testing manual -- needed for Section 5 (NHTSA Compliance Cross-Reference)
• Applicable state DUI/DWI statutes -- needed for Section 6 (Statutory Analysis)
• Motion templates (suppress, dismiss, exclude) -- useful for Section 8 (Motion Opportunities)

If the NHTSA manual is not available, note this for Step 3 -- the NHTSA compliance table will be skipped with an explicit notation.

Step 3: Produce the Defense Memo

Analyze all extracted text and produce the following sections. Follow the Agent Delegation workflow above — save extracted text to files and launch subagents for the section groups defined in the delegation table.

Section 1: Case Snapshot

Key facts in a compact list: defendant name, DOB, charges with statutory citations, arrest date/time, location, arresting agency, officer/badge, court/case number, BAC (if available). End with a two-sentence preliminary defense theory. Keep to ~15 lines.

Section 2: Chronology

Timeline from initial observation through booking. Each entry: one-line event description, source citation, and conflict flag if sources disagree. Only include events with defense significance — skip routine procedural steps unless a deviation occurred.

Format: Time | Event (1 sentence) | Source (page/timestamp) | Conflicts/Notes (1 sentence max)

Section 3: Evidence Inventory

Catalog evidence in the discovery package. One row per item, cells kept to a few words each.

Item	Type	Source	Key Content (1 phrase)	Status

Status: Complete, Partial, Referenced But Missing, or Needs Follow-Up.

Section 4: Officer Conduct Review

Bullet-point review of officer procedural compliance. One bullet per issue, cite the source, note the defense angle. Only expand on items where a deviation or gap exists.

• Miranda warnings: timing, language, waiver
• Probable cause for stop and arrest
• Search and seizure
• Chain of custody
• Documentation gaps

Section 5: NHTSA Compliance Cross-Reference

Compare officer conduct against NHTSA protocol for each FST administered. Table cells must be 1-2 sentences max.

Test	Officer Documented (1-2 sentences)	NHTSA Requirement (Session/Page)	Deviation (1 sentence)	Defense Significance (1 sentence)

Cover HGN, Walk-and-Turn, One-Leg Stand if administered. Flag non-standardized tests (finger-to-nose, Romberg, alphabet) with a one-line note that NHTSA has not validated them.

Section 6: Statutory Analysis

Map evidence against each element of the charged offense. One row per element, keep cells concise.

Element	Evidence (cite page)	Strength	Defense Gap (1 sentence)

Strength: Strong / Weak / Unsupported.

Section 7: Cross-Document Inconsistencies

Quote exact language -- do not paraphrase. Only include inconsistencies with defense significance. Keep quotes to the key phrase, not full sentences.

Issue (1 phrase)	Source A (Quote, Page)	Source B (Quote, Page)	Defense Significance (1 sentence)

Focus on: factual contradictions, omissions between documents, characterization differences (e.g., "slurred speech" vs. coherent dialogue on body cam).

Section 8: Motion Opportunities

One bullet per motion opportunity. Prioritize by strength -- list Strong motions first. Keep each entry to 2-3 lines max.

• Format per motion: Motion type (suppress/dismiss/exclude/limine) | Legal basis (1 phrase) | Key facts with citation | Strength (Strong / Moderate / Worth Filing) + 1-sentence rationale
• Reference available motion templates if present in the project.

Section 9: Recommended Next Steps

Prioritized bullet list of actionable items. One sentence per bullet. Group by category but do not elaborate beyond the action and its rationale.

• Discovery requests (what + why)
• Witnesses to interview/subpoena
• Expert witnesses (type + question they address)
• Missing evidence to obtain (dashcam, body cam angles, dispatch audio, calibration logs, training records)
• Investigation tasks (scene visit, measurements, etc.)

Output Format

Hard limit: ~300 lines for the assembled memo. This is not a suggestion — exceeding it produces an unusable document. The per-agent limits in the delegation table enforce this. Conciseness rules:

• Bullet points over paragraphs. One sentence per bullet. No multi-sentence narrative blocks.
• Table cells: 1-2 sentences max. Never multi-paragraph. If a cell needs more, the content is not sufficiently distilled.
• No filler. Omit introductory/concluding paragraphs within sections. Start each section with its first substantive item.
• Prioritize ruthlessly. Include the top findings per section. A tight 300-line memo an attorney will actually read beats an exhaustive 1,300-line document they will not.
• No title page, table of contents, or preamble. The document starts at Section 1.

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

• Cite everything. Page numbers, NHTSA manual sessions, body cam timestamps, statute numbers. No factual claim without a source.
• Quote, don't paraphrase when comparing documents. The attorney needs exact language.
• Never generate case law. Where case authority would strengthen an argument, mark it [CASE LAW RESEARCH NEEDED] with a note on what type of authority to look for.
• Flag gaps as [NEEDS INVESTIGATION] rather than guessing. If a document is referenced but not in the file, say so. If a timeline entry cannot be pinned to a specific time, say so.
• Never assume missing data. If the BAC result is not in the file, do not guess. If the officer's training history is unknown, say so.
• Analyze for the defense. This is a defense memo, not a neutral summary. Note prosecution strengths only so the attorney can prepare to neutralize them.

Edge Cases

• No NHTSA manual available: Skip the NHTSA compliance table. Note: "NHTSA cross-reference not performed -- manual not available. Add the NHTSA SFST manual to enable this analysis."
• No field sobriety tests administered: Replace the NHTSA section with a note on the implications -- the State's case relies entirely on other evidence. Analyze what that evidence is and its weaknesses.
• Multiple defendants or incidents: Analyze each defendant separately. Produce separate case snapshots and chronologies.
• Incomplete discovery package: Flag every gap explicitly. List what is missing and why it matters.
• Non-DUI charges: Adapt the analysis structure. Replace the NHTSA section with the relevant compliance or procedural framework for the charge type (e.g., use-of-force policy for assault charges, search warrant requirements for drug cases).
• Scanned documents with poor OCR quality: Flag low-confidence sections from OCR output and note: "[OCR QUALITY WARNING] Text in this section may contain errors -- verify against original document."
• Mixed media discovery: When the package contains both documents and recordings, process all media types through their respective skills before beginning analysis.