ooligo
claude-skill

Subpoena triage with Claude

Difficulty
intermediate
Setup time
30min
For
legal-ops · in-house-counsel · paralegal
Legal Ops

Stack

A Claude Skill that triages an incoming subpoena (or third-party document request, or law-enforcement request) — extracts the issuer, jurisdiction, custodian, response deadline, requested document categories, and privilege exposure; classifies whether the firm has standing to object; estimates scope (number of custodians implicated, expected document volume) — and produces a structured intake report the in-house counsel uses to scope the response. Replaces the paralegal’s first-pass read of a 12-page subpoena with a 10-minute review of a structured report. Always escalates to counsel; never auto-responds, never auto-acknowledges service.

When to use

  • The firm receives subpoenas or law-enforcement requests at a frequency where intake triage is itself a bottleneck (typically >5 per month).
  • The firm has a written subpoena response procedure (named counsel owners, document custodian map, privilege-log practices). The skill structures the intake; the procedure is the substance.
  • The firm distinguishes between subpoena types (civil third-party, grand-jury, regulatory, foreign legal-process) — different objection paths and timelines apply, and the skill needs the firm’s classification rubric.

When NOT to use

  • Auto-acknowledging service. Service rules vary by jurisdiction; in some, acknowledging service triggers the response clock. The counsel decides on acknowledgement.
  • Replacing counsel’s privilege analysis. The skill flags privilege exposure (privileged communications likely in scope, work-product doctrine implicated). The counsel does the analysis.
  • Foreign letters rogatory or Hague Convention requests. Different procedure, different timeline, different counsel involvement. The skill flags but does not handle.
  • Grand jury subpoenas. Often confidential; the skill should not write to standard tracking systems for these — counsel determines storage posture per matter.
  • Replacing a legal-hold issuance. A subpoena triggers a hold for the relevant custodians; the skill flags the trigger but the litigation hold orchestration is the issuance flow.

Setup

  1. Drop the bundle. Place apps/web/public/artifacts/subpoena-triage-claude-skill/SKILL.md into your Claude Code skills directory.
  2. Author the firm’s classification rubric. Per subpoena type, document: who owns the response, which jurisdiction’s rules apply, default timeline, default counsel routing. Template in references/1-subpoena-classification.md.
  3. Configure the custodian-map source. The skill estimates which custodians are implicated; that requires a current map. Pull from HRIS / e-discovery platform / legal-ops registry.
  4. Set the privilege-flag rules. When a subpoena’s scope likely overlaps with attorney-client communications, the skill flags it. The flagging rules vary by firm (in-house GC vs outside-counsel posture). Template in references/2-privilege-flag-rules.md.
  5. Dry-run on closed subpoenas. Process three closed subpoenas. Compare the skill’s classification, scope estimate, and privilege flags to what the counsel actually did.

What the skill does

Six steps. Classification before scope estimation, because scope depends on what type of subpoena it is.

  1. Extract metadata. Issuer (court, agency, requesting party), case caption, service date, response deadline, jurisdiction. Halt if any of these can’t be determined — counsel needs all five before triage proceeds.
  2. Classify by type. Civil third-party / grand-jury / regulatory (SEC, DOJ, state AG) / foreign legal-process / law-enforcement. Per the firm’s rubric, route to the appropriate counsel owner and timeline.
  3. Extract requested document categories. The subpoena’s requests are typically itemized; the skill extracts each and tags with the firm’s document-category taxonomy (contracts, communications, financials, technical specs, HR records, etc.). Flag overbroad requests (request asking for “all documents related to X” without temporal or scope limit).
  4. Estimate custodian implications. From the document categories and the matter context, identify which custodians (named individuals or roles) are likely to hold responsive documents. Output: list of custodians, NOT a request to issue a hold (that’s the next step, counsel-approved).
  5. Flag privilege exposure. Where requested categories overlap with likely-privileged communications (legal advice, work product, joint-defense communications), flag with paired guard (“flagging because category ‘all communications with counsel’ implicates attorney-client privilege; recommend privilege log preparation as part of response”).
  6. Emit triage report. Structured Markdown for the counsel + JSONL audit log per intake.

Cost reality

  • LLM tokens — typically 8-15k input (subpoena PDF text + skill instructions) and 2-4k output. ~$0.10-0.20 per subpoena.
  • Counsel / paralegal time — the win. Manual triage of a complex subpoena is 1-2 hours. Reviewing the skill’s report is 15-30 minutes. The bigger time saver is on consistency — every subpoena goes through the same structure.
  • Setup time — 30 minutes plus per-firm rubric authoring.

Success metric

  • Time-to-counsel-routing — should drop from 4-24 hours (manual paralegal triage in queue) to under 30 minutes.
  • Counsel re-classification rate — share of subpoenas the counsel reclassifies after the skill’s output. Should be under 15%; above that the rubric needs tuning.
  • Privilege-flag coverage — at retrospective review, share of subpoenas where privilege issues actually arose vs. share where the skill flagged them. Misses are the signal to update the privilege rules.

vs alternatives

  • vs paralegal first-pass review. Paralegal review remains the right call for the subpoenas where context-heavy judgment matters (e.g. a subpoena referencing a matter the paralegal has been working). The skill complements rather than replaces — it handles volume while paralegals handle nuance.
  • vs e-discovery platform intake (Relativity, Everlaw, Logikcull). Those handle the response-execution side; the skill handles the intake-triage side. Use both.
  • vs counsel reads every subpoena cold. The most accurate path; also the slowest. The skill earns its setup cost on volume.

Watch-outs

  • Citation hallucination on legal classifications. Guard: every classification cites the firm’s rubric section; classifications without a citable rule are flagged “rubric does not cover” rather than asserted.
  • Privilege exposure missed. Guard: the privilege-flag rules are conservative — over-flag rather than under-flag. The counsel tightens later as patterns emerge.
  • Service-acknowledgement implication. Guard: the skill explicitly notes “skill processed; service NOT acknowledged” in the audit log. Counsel handles acknowledgement.
  • Confidentiality of grand-jury subpoenas. Guard: the skill’s classification step routes grand-jury subpoenas to a separate, restricted-access destination per the firm’s procedure. Counsel reviews before the skill writes to any standard tracking system.
  • Foreign legal-process drift. Guard: foreign letters rogatory and Hague Convention requests trigger a “skill cannot triage — escalate to international counsel” output rather than miscategorizing as a domestic subpoena.
  • Custodian estimate over-estimate. Guard: the custodian list is a starting point for the counsel; over-estimates are flagged as such. The skill does not auto-issue holds.

Stack

The bundle lives at apps/web/public/artifacts/subpoena-triage-claude-skill/:

  • SKILL.md — the skill definition
  • references/1-subpoena-classification.md — per-firm classification rubric
  • references/2-privilege-flag-rules.md — privilege-flag rule template

Tools: Claude. For follow-on hold issuance, see litigation hold orchestration. For document-collection orchestration, see evidence collection for ediscovery.

Related: legal intake, privilege review, ediscovery.

Files in this artifact

Download all (.zip)