Use when the user provides a client alert, regulatory bulletin, law firm memo, or similar legal update and wants the time-sensitive action items, deadlines, and obligations extracted into a checklist organized by deadline. Distinguishes mandatory action items (effective dates of new requirements) from recommended action items (best practices) and informational items (context only). Produces a brief context summary plus deadline-organized checklist with citations to the source alert.
Use when the user provides a client alert, regulatory bulletin, law firm memo, or similar legal update and wants the time-sensitive action items, deadlines, and obligations extracted into a checklist organized by deadline. Distinguishes mandatory action items (effective dates of new requirements) from recommended action items (best practices) and informational items (context only). Produces a brief context summary plus deadline-organized checklist with citations to the source alert.
Extract time-sensitive action items, deadlines, and obligations from a client alert, regulatory bulletin, law firm memo, or similar legal update. Produce a deadline-organized checklist that an in-house lawyer can act on without re-reading the source document.
The skill is operational, not analytical. Action items are stated at the level of granularity the alert supports — typically high-level ("comply with new disclosure requirement by [date]"), not implementation-level ("update privacy policy section 4.2 by [date]"). The user knows their organization better than the skill does; the skill provides the trigger and deadline, not the implementation.
Audience. In-house counsel or compliance analyst with regulatory-intake experience who can audit the extracted checklist against the source alert. Not suitable for unsupervised non-lawyer use; the output is a reviewed working draft, not a deliverable to non-legal stakeholders without lawyer sign-off. Outside counsel and smaller orgs without an in-house team structure should treat functional-owner suggestions ("Legal", "Compliance", "IT") as placeholders and reassign per their own staffing.
When this skill applies
Apply when the user provides a legal-update document and asks for action items, deadlines, or obligations to be extracted. Typical inputs:
Law firm client alerts ("Client Alert: New SEC Cybersecurity Disclosure Rules Take Effect").
Law firm memos on regulatory developments or new statutes.
Industry association updates distilling regulatory changes for members.
Internal compliance bulletins if structured similarly to external alerts.
Do not apply when:
The document is the underlying regulation, statute, or rule itself. Regulations are dense and structured for legal interpretation, not for action-item extraction. Recommend the user start with a client alert summarizing the regulation, or escalate to legal counsel for direct regulatory analysis.
The document is a court opinion or judicial decision. Different shape; would warrant a case-summary skill (deferred enhancement candidate).
The document is a private communication or strategy memo. Different shape; the skill is calibrated for published / shared alerts.
The user wants an analysis of the legal change rather than action items. Action-item extraction is operational. If the user wants to understand the substantive legal change, recommend reading the alert directly or engaging counsel.
Some alerts describe regulatory developments without imposing or describing specific requirements. The skill notes this and produces a "no specific action items extractable" output rather than fabricating actions.
The document is too vague to extract specific actions.
Inputs
The skill requires the document. Optional inputs filter and prioritize the extraction:
organization_context affects applicability. An alert about EU AI Act obligations is highly relevant to an EU-based AI vendor and largely irrelevant to a US-only consumer-goods retailer. The skill uses the organization context to flag items as applicable, conditionally applicable, or not applicable.
relevant_business_areas filters extraction. If the user is focused on data privacy and the alert covers privacy plus employment, the privacy items get full extraction and employment items are noted briefly with a reference to re-run for that focus.
applicable_jurisdictions affects which items are surfaced. An alert covering federal plus multiple states' laws produces different relevance for users in different states.
alert_date affects deadline assessment. Some alerts arrive after some deadlines have passed; the skill should flag past-deadline items separately from forward-looking ones.
When optional inputs are not provided, the skill extracts all items from the alert and flags applicability uncertainty.
Workflow
The workflow has four steps.
Step 1: Document orientation
Before extraction:
Confirm the document is a client alert or similar update. If it's something else (the regulation itself, a court opinion, internal correspondence), stop and route the user.
Identify the alert's date. If not stated, ask the user (the alert_date input) before proceeding — deadline assessment depends on the alert date.
Identify the alert's subject matter (regulatory regime, agency, statute, jurisdiction).
Identify the alert's structure: is it organized by topic, by deadline, by stakeholder? Different structures affect extraction approach.
Note any references to the underlying regulation, prior alerts, or related materials.
Step 2: Item extraction
Walk through the document and extract every item that meets one of these criteria:
Mandatory action item: something the user (or organizations like the user) is required to do, with a deadline. Source: the alert describes a new requirement, a compliance deadline, an effective date, a filing date, or a similar legal obligation.
Recommended action item: something the alert recommends as best practice without it being legally required. Source: the alert uses language like "should consider," "best practice," "we recommend," "prudent companies will."
Informational item: context that affects the user's understanding of the regulatory landscape but doesn't require specific action. Source: the alert describes background, agency posture, enforcement trends, or related developments without imposing requirements.
Conditional action item: action required only under specific circumstances (e.g., "if you process EU residents' data, you must..."). The skill flags the condition and the action.
For each item, capture:
What: the action or obligation, stated in plain language.
By when: the deadline or effective date, with explicit date if available.
Who within the organization: typical functional owner if the alert suggests one (Legal, Compliance, IT, HR, etc.); otherwise marked "Owner: TBD."
Why / source: the underlying legal basis (statute, regulation, agency guidance), with citation to the alert's section.
Conditions / applicability: whether the item applies to all organizations or only specific ones.
Extraction confidence: tag each item per the linguistic patterns graded in reference/extraction_patterns.md.
High — the alert states the action and deadline explicitly with mandatory language ("must", "shall", "is required to", a specified effective/compliance date).
Medium — the alert implies the action or deadline but does not state it directly (e.g., "the rule takes effect for fiscal years ending on or after [date]" requires deriving the filing date from the user's fiscal year).
Low — the alert is ambiguous, conditional on facts not in the document, or uses hedged language ("companies may need to consider…"). Item is extracted but explicitly flagged for lawyer judgment.
Source-fidelity tag: mark each line in the item as either [Extracted] — copied or directly paraphrased from the alert with section citation — or [Inferred] — derived by the skill (e.g., a specific filing deadline computed from an effective date and an assumed filer class, or a functional owner inferred from the topic). Inferred lines must state the assumption and ask the reviewer to confirm.
Step 3: Apply filters and applicability flags
For each extracted item, assess applicability based on optional inputs:
Applicable — the item applies to the user's organization based on organization_context, relevant_business_areas, and applicable_jurisdictions.
Conditionally applicable — the item applies if a specific condition is met (the user has EU users; processes children's data; engages in cross-border data transfers).
Not applicable — the item does not apply to the user based on the inputs provided.
Applicability uncertain — the user's optional inputs are not specific enough to assess; the skill notes the item and flags applicability for user review.
If organization_context is not provided, all items are flagged "applicability uncertain" and the user must assess applicability themselves.
Step 4: Organize by deadline and produce checklist
Sort items by deadline, with separate sections for:
Past-deadline items — deadlines that have already passed at the time of the review. May still warrant action (compliance lookback; remediation; ongoing obligations).
Imminent items — deadlines within 30 days of the review.
Near-term items — deadlines 30 days to 6 months out.
Future items — deadlines beyond 6 months.
Ongoing obligations — items without specific deadlines that represent continuing requirements.
Recommended (no deadline) — best-practice items the alert recommends without specific deadline.
Informational only — items that don't require action.
Output
Produce the report in markdown with this structure:
# Action Items from Client Alert: [Alert title or subject]**Source document:** [alert title and date]
**Alert date:** [date]
**Organization context:** [user-provided, or "not specified"]
**Relevant business areas:** [user-provided, or "all"]
**Applicable jurisdictions:** [user-provided, or "not specified"]
## Context summary
[2-4 sentences. What is the alert about? What regulatory development triggered it? Why does it matter? This is to orient the user without requiring them to read the alert; the action items are below.]
## Mandatory action items
Each item lists, in addition to the substantive fields: **Extraction confidence** (High / Medium / Low) and a per-line **[Extracted]** vs **[Inferred]** tag. Inferred lines state the assumption and request reviewer confirmation.
### Past-deadline items
[If any. Each with: what, deadline (passed), source citation, applicability, recommended remediation, extraction confidence, per-line Extracted/Inferred tags.]
### Imminent (within 30 days)
[Each with: what, deadline (specific date), owner, source citation, applicability, extraction confidence, per-line Extracted/Inferred tags.]
### Near-term (30 days to 6 months)
[Each with: what, deadline, owner, source citation, applicability, extraction confidence, per-line Extracted/Inferred tags.]
### Future (beyond 6 months)
[Each with: what, deadline, owner, source citation, applicability, extraction confidence, per-line Extracted/Inferred tags.]
### Ongoing obligations
[Each with: what, periodicity if applicable, owner, source citation, applicability, extraction confidence, per-line Extracted/Inferred tags.]
## Recommended action items (no specific deadline)
[Each with: what, why recommended, owner, source citation, applicability, extraction confidence, per-line Extracted/Inferred tags.]
## Informational items
[Items the alert flags as context. Brief — usually a bulleted list of 2-5 items that affect the user's understanding without requiring action.]
## Items where applicability is unclear
[Items the skill could not assess for applicability based on the inputs provided. The user must assess these themselves.]
## Items not applicable to user's organization
[If `organization_context` was provided and any items were assessed as not applicable. Brief — these are noted for completeness so the user knows nothing was missed.]
## Source references and follow-ups
[List of: the underlying regulation/statute/guidance referenced in the alert; any prior alerts referenced; suggested follow-up actions (e.g., "obtain the underlying regulation for the legal team's review file"). Brief.]
## Notes on this extraction
[Brief — typically a sentence or two. If the alert was vague, if extraction required interpretation, if items were uncertain — note here. If everything was clear, this section can be a single sentence: "Extraction was straightforward; all items have clear deadlines and applicability."]
The report should match the alert's substance, not pad. A brief alert (1-2 pages, a few action items) produces a brief report. A comprehensive alert (10+ pages, dozens of action items across multiple regimes) produces a long report. Don't pad short alerts; don't truncate substantial ones.
Edge cases and refusals
Alert is too vague to extract specific actions. Produce the context summary and note in "Notes on this extraction" that the alert is informational rather than actionable. Recommend the user obtain a more specific compliance roadmap from outside counsel if the regulatory development warrants action.
Alert covers a regulation that's been updated since the alert was published. If the alert is dated and the user notes the deadline has shifted, the skill should flag this. Without notification, the skill cannot detect this.
Alert is conditional on facts the skill cannot assess ("if your company has more than 250 employees..."). The skill extracts the item and flags it as conditionally applicable, with the condition explicit.
Alert covers many jurisdictions but the user is in only one. The skill filters by applicable_jurisdictions if provided; if not provided, all items are extracted with jurisdiction noted and applicability flagged uncertain.
Alert is duplicative of a prior alert (same regulation re-summarized). The skill extracts items as if the alert were standalone; the user is responsible for identifying duplication with prior compliance work.
Action item the alert recommends seems impractical, illegal, or in conflict with other obligations. The skill extracts the item as the alert states it but notes the concern; this is a flag for the user, not a refusal to extract.
What this skill does not do
Substantive analysis of the regulatory development. The skill extracts what the alert says; it does not opine on whether the alert is correct, whether the regulatory development is well-founded, or whether the action items are sufficient.
Implementation guidance. Action items are at the level the alert supports. The user knows their organization; the skill does not.
Compliance certification. Extracting an action item is not the same as confirming compliance.
Enforcement-risk assessment. The skill does not opine on whether non-compliance is likely to be detected or enforced.
Multi-document synthesis. The skill operates on one alert at a time. Synthesizing across multiple alerts (e.g., comparing different law firms' takes on the same regulatory development) is out of scope; deferred enhancement candidate.
Legal advice. The skill produces an extracted checklist; legal advice on the user's specific circumstances is outside scope.
Reference materials
reference/extraction_patterns.md — patterns for identifying action items, deadlines, and applicability conditions in alert text.
reference/deadline_calibration.md — guidance on deadline categorization (past-deadline, imminent, near-term, future) and edge cases.
examples/example_clear_alert.md — worked example: well-structured alert with clear deadlines.
examples/example_vague_alert.md — worked example: vague alert that produces few extractable items.
examples/example_multi_jurisdiction.md — worked example: alert covering multiple jurisdictions, demonstrating jurisdiction-filtering.
QA Remediation (LegalQuants, 2026-05)
LegalQuants added a targeted remediation pass to address SOME CONCERN gaps surfaced in QA: reconciled the frontmatter author field with the registry (Kevin Keller, per README); added an explicit Audience line (in-house counsel / compliance analyst with regulatory-intake experience); surfaced per-item extraction-confidence bands (High / Medium / Low) tied to the linguistic patterns already graded in reference/extraction_patterns.md; and introduced per-line [Extracted] vs [Inferred] source-fidelity tagging so reviewers can distinguish text copied from the alert from outputs the skill computed (e.g., filing dates derived from effective dates and assumed filer class). Original technical content — workflow steps, applicability flags, edge cases, refusals, and non-goals — is unchanged.
