Investigation Review Hub
Workflow
- Read the task prompt and every task-local payload, especially
answer_template.json and any request_context.json, review_scope.json, or matter_context.json.
- Treat
answer_template.json as the output contract. Preserve required top-level keys, item keys, enum values, nullability, ordering rules, and numeric precision exactly. Return one JSON object only.
- Use only the task-local payloads and the running Investigation Review Hub. Do not inspect environment source, database files, generated data files, hidden manifests, task answers, or evaluator files.
- Determine the hub base URL from the task. If the prompt uses
<TASK_ENV_BASE_URL>, use the environment access information supplied with the task only to get the base URL. Use any task-provided SQL API header exactly as given.
- Pull all hub evidence for the requested
matter_id. Prefer scripts/fetch_hub_matter.py when SQL access is available. Run it from the skill package directory, or replace the script path with the resolved path to the skill resource:
python scripts/fetch_hub_matter.py --base-url "$TASK_ENV_BASE_URL" --matter-id "$MATTER_ID" --api-key "$API_KEY" > hub.json
If SQL is unavailable, call the listed REST endpoints and filter rows by matter_id locally.
Hub Evidence Model
Expect these recurring evidence families:
- Matter metadata: agency, investigation type, issued date, hold date, status.
- Subpoena/request categories: category codes, labels, date ranges, topics.
- Production stats: produced, withheld, responsive, nonresponsive, status, zero-claim notes.
- Custodian sources: source type, status, event date, post-hold flag, category impacts, issue tags.
- Review documents: responsiveness, privilege status, produced status, category, issue tags, summaries.
- Privilege entries: withheld/logged counts, issue type, third-party flag, category, notes.
- QC findings: issue type, severity, document count, affected category, source reference.
- Retention events: event date, hold date, policy section, volume, status, affected categories, source ref.
- Remediation actions: action type, priority, severity, owner, target refs, due days.
Parse category-impact fields and issue-tag fields as structured lists when possible; otherwise split conservatively on commas or semicolons. Keep stable hub IDs exactly as emitted.
Materiality Filter
Hub tables can contain noisy routine rows, metadata gaps, family-member records, similar labels from other matters, and ordinary collection variance. Do not promote a row merely because a status is imperfect or a count differs. Treat an item as material when it is supported by the task focus plus at least one strong signal:
- a retention, source, privilege, QC, document, or remediation row has an issue type/tag/note that matches the requested review;
- a remediation action targets the record;
- a QC finding or reviewed document directly contradicts a production status or zero-production claim;
- the row is an available archive/source that the prompt asks to report as a remediation path;
- related rows across tables support the same category, source, or privilege defect.
Ignore routine, closed, duplicate, metadata-only, and unrelated-category rows unless the template explicitly asks to report them.
Classify Material Issues
Use the template enum names, but map evidence consistently:
- Post-hold destruction or loss:
post_hold_loss, preservation_failure, source_lost, destroyed, critical or high, usually disclose_preservation_issue.
- Pre-hold policy destruction:
policy_destroyed_pre_hold, low, normally no_action_policy_loss unless the template asks only for open production blockers.
- Missing required record:
should_exist_missing or missing_required_record, action locate_missing_record.
- Uncollected personal or side-channel source:
personal_source_gap, collection_gap, not_collected or partial_collection, action collect_personal_device or source-specific collection. Require a material issue tag, action, or cross-table support; do not escalate ordinary tracker variance.
- Available archive or retained source: include in retained/available-source sections when it is a remediation path; classify as
archive_available, available_archive, available, or preserved_available as the template permits, action search_archive or collect_archive, and do not treat as an irretrievable loss by itself.
- Responsive material coded nonresponsive, zero-production contradiction, or QC responsiveness finding: classify as
responsiveness_miscode, responsive_miscoding, or zero_claim_contradiction; count supporting documents; action recode_and_produce.
- Privilege-log gap: issue type, notes, QC, or action indicate incomplete log. Then calculate
unlogged_count = max(withheld_count - logged_count, 0). Do not infer an escalated log gap from withheld/logged differences alone when the row is labeled clean, family mismatch, routine, or outside the prompt scope.
- Third-party privilege issue: third-party flag/name or notes show disclosure to a non-privileged recipient; classify as
third_party_waiver, production impact privilege_exposure, action waiver_assessment_and_disclosure.
- Privilege miscoding or over-designation: QC or privilege notes show privileged documents coded nonprivileged, business-only counsel-copy overdesignation, downgrade, or recode needs; action
privilege_recode_and_log or qc_remediation according to the template enums.
Include only material issues requested by the prompt and schema. Keep closed/no-gap categories out unless the template explicitly requests complete coverage.
Build Output Sections
- Risk or issue ledgers: anchor each item on the most specific stable hub ID. Add supporting
source_refs or record_refs for documents, QC findings, privilege entries, retention events, and sources. Use 0 for non-applicable counts when the schema says integer; use null only where the schema allows it.
- Category coverage/statuses: aggregate all open material blockers per category. Choose the status and production impact that best reflect the highest-severity blocker; use a multi-blocker enum when available. Sort supporting refs and category lists.
- Retained or available sources: report sources that remain remediation paths or limit loss.
limits_loss_for_categories should be the intersection of affected categories and categories the source can remediate.
- Metrics: calculate from the selected evidence, not from prose. Count unique events/sources/categories where the metric name asks for events, sources, or categories. Do not double-count the same issue through both a QC finding and its source document unless the metric asks for documents.
- Actions: use hub remediation actions when they match the schema. Otherwise synthesize a minimal grouped action plan from the material issues, ordered by urgency.
Ranking and Owners
Rank preservation disclosures for post-hold losses first, then waiver/exposure issues, recode-and-produce blockers, privilege-log supplementation, collection of missing sources, archive searches, and low-risk documentation/no-action items. Prefer owner values already present in remediation actions; otherwise map actions to common owners:
- disclosure or readiness hold: outside counsel or litigation counsel
- forensic or personal-device collection: forensics, eDiscovery vendor, or client IT
- review coding/QC: review QC or review vendor
- privilege log, waiver, recode: privilege team or privilege counsel
- records retention or missing business records: records management or compliance audit
Always choose an enum value allowed by the template.
Final Checks
- Validate every enum, required key, count, boolean,
null, and list against answer_template.json.
- Apply all ordering rules from the template, including sorting IDs and category codes inside arrays.
- Ensure
production_ready or rolling_production_ready is false when any material open blocker remains.
- Remove analysis notes and return exactly one JSON object with no Markdown or prose.
1---2name: investigation-review-hub-53description: Use for legal investigation review tasks that require structured JSON from an Investigation Review Hub, including production gap analysis, retention and litigation-hold loss review, SEC/DOJ remediation dashboards, privilege-log/QC production-readiness reviews, and cross-system action plans based on hub endpoints plus task-local answer templates.4metadata:5 source: https://skillmd.com/skills/prism-shadow/investigation-review-hub-56---78# Investigation Review Hub910## Workflow11121. Read the task prompt and every task-local payload, especially `answer_template.json` and any `request_context.json`, `review_scope.json`, or `matter_context.json`.132. Treat `answer_template.json` as the output contract. Preserve required top-level keys, item keys, enum values, nullability, ordering rules, and numeric precision exactly. Return one JSON object only.143. Use only the task-local payloads and the running Investigation Review Hub. Do not inspect environment source, database files, generated data files, hidden manifests, task answers, or evaluator files.154. Determine the hub base URL from the task. If the prompt uses `<TASK_ENV_BASE_URL>`, use the environment access information supplied with the task only to get the base URL. Use any task-provided SQL API header exactly as given.165. Pull all hub evidence for the requested `matter_id`. Prefer [scripts/fetch_hub_matter.py](scripts/fetch_hub_matter.py) when SQL access is available. Run it from the skill package directory, or replace the script path with the resolved path to the skill resource:1718```bash19python scripts/fetch_hub_matter.py --base-url "$TASK_ENV_BASE_URL" --matter-id "$MATTER_ID" --api-key "$API_KEY" > hub.json20```2122If SQL is unavailable, call the listed REST endpoints and filter rows by `matter_id` locally.2324## Hub Evidence Model2526Expect these recurring evidence families:2728- Matter metadata: agency, investigation type, issued date, hold date, status.29- Subpoena/request categories: category codes, labels, date ranges, topics.30- Production stats: produced, withheld, responsive, nonresponsive, status, zero-claim notes.31- Custodian sources: source type, status, event date, post-hold flag, category impacts, issue tags.32- Review documents: responsiveness, privilege status, produced status, category, issue tags, summaries.33- Privilege entries: withheld/logged counts, issue type, third-party flag, category, notes.34- QC findings: issue type, severity, document count, affected category, source reference.35- Retention events: event date, hold date, policy section, volume, status, affected categories, source ref.36- Remediation actions: action type, priority, severity, owner, target refs, due days.3738Parse category-impact fields and issue-tag fields as structured lists when possible; otherwise split conservatively on commas or semicolons. Keep stable hub IDs exactly as emitted.3940## Materiality Filter4142Hub tables can contain noisy routine rows, metadata gaps, family-member records, similar labels from other matters, and ordinary collection variance. Do not promote a row merely because a status is imperfect or a count differs. Treat an item as material when it is supported by the task focus plus at least one strong signal:4344- a retention, source, privilege, QC, document, or remediation row has an issue type/tag/note that matches the requested review;45- a remediation action targets the record;46- a QC finding or reviewed document directly contradicts a production status or zero-production claim;47- the row is an available archive/source that the prompt asks to report as a remediation path;48- related rows across tables support the same category, source, or privilege defect.4950Ignore routine, closed, duplicate, metadata-only, and unrelated-category rows unless the template explicitly asks to report them.5152## Classify Material Issues5354Use the template enum names, but map evidence consistently:5556- Post-hold destruction or loss: `post_hold_loss`, `preservation_failure`, `source_lost`, `destroyed`, `critical` or `high`, usually `disclose_preservation_issue`.57- Pre-hold policy destruction: `policy_destroyed_pre_hold`, `low`, normally `no_action_policy_loss` unless the template asks only for open production blockers.58- Missing required record: `should_exist_missing` or `missing_required_record`, action `locate_missing_record`.59- Uncollected personal or side-channel source: `personal_source_gap`, `collection_gap`, `not_collected` or `partial_collection`, action `collect_personal_device` or source-specific collection. Require a material issue tag, action, or cross-table support; do not escalate ordinary tracker variance.60- Available archive or retained source: include in retained/available-source sections when it is a remediation path; classify as `archive_available`, `available_archive`, `available`, or `preserved_available` as the template permits, action `search_archive` or `collect_archive`, and do not treat as an irretrievable loss by itself.61- Responsive material coded nonresponsive, zero-production contradiction, or QC responsiveness finding: classify as `responsiveness_miscode`, `responsive_miscoding`, or `zero_claim_contradiction`; count supporting documents; action `recode_and_produce`.62- Privilege-log gap: issue type, notes, QC, or action indicate incomplete log. Then calculate `unlogged_count = max(withheld_count - logged_count, 0)`. Do not infer an escalated log gap from withheld/logged differences alone when the row is labeled clean, family mismatch, routine, or outside the prompt scope.63- Third-party privilege issue: third-party flag/name or notes show disclosure to a non-privileged recipient; classify as `third_party_waiver`, production impact `privilege_exposure`, action `waiver_assessment_and_disclosure`.64- Privilege miscoding or over-designation: QC or privilege notes show privileged documents coded nonprivileged, business-only counsel-copy overdesignation, downgrade, or recode needs; action `privilege_recode_and_log` or `qc_remediation` according to the template enums.6566Include only material issues requested by the prompt and schema. Keep closed/no-gap categories out unless the template explicitly requests complete coverage.6768## Build Output Sections6970- Risk or issue ledgers: anchor each item on the most specific stable hub ID. Add supporting `source_refs` or `record_refs` for documents, QC findings, privilege entries, retention events, and sources. Use 0 for non-applicable counts when the schema says integer; use `null` only where the schema allows it.71- Category coverage/statuses: aggregate all open material blockers per category. Choose the status and production impact that best reflect the highest-severity blocker; use a multi-blocker enum when available. Sort supporting refs and category lists.72- Retained or available sources: report sources that remain remediation paths or limit loss. `limits_loss_for_categories` should be the intersection of affected categories and categories the source can remediate.73- Metrics: calculate from the selected evidence, not from prose. Count unique events/sources/categories where the metric name asks for events, sources, or categories. Do not double-count the same issue through both a QC finding and its source document unless the metric asks for documents.74- Actions: use hub remediation actions when they match the schema. Otherwise synthesize a minimal grouped action plan from the material issues, ordered by urgency.7576## Ranking and Owners7778Rank preservation disclosures for post-hold losses first, then waiver/exposure issues, recode-and-produce blockers, privilege-log supplementation, collection of missing sources, archive searches, and low-risk documentation/no-action items. Prefer owner values already present in remediation actions; otherwise map actions to common owners:7980- disclosure or readiness hold: outside counsel or litigation counsel81- forensic or personal-device collection: forensics, eDiscovery vendor, or client IT82- review coding/QC: review QC or review vendor83- privilege log, waiver, recode: privilege team or privilege counsel84- records retention or missing business records: records management or compliance audit8586Always choose an enum value allowed by the template.8788## Final Checks8990- Validate every enum, required key, count, boolean, `null`, and list against `answer_template.json`.91- Apply all ordering rules from the template, including sorting IDs and category codes inside arrays.92- Ensure `production_ready` or `rolling_production_ready` is false when any material open blocker remains.93- Remove analysis notes and return exactly one JSON object with no Markdown or prose.