Investigation Review Hub
Use the Investigation Review Hub as the source of record for eDiscovery analysis.
The hub is a read-only REST API; its base URL is always provided in the task
context.
Core Rules
- Never inspect local files for business evidence. The hub is the only
source of record for matter metadata, subpoena categories, production stats,
custodian sources, review documents, privilege-log data, QC findings,
retention events, and remediation actions.
- Filter everything by matter_id. Every hub response spans all matters;
filter client-side or in SQL with
WHERE matter_id = '...'.
- Follow the answer template exactly. Every task provides an
input/payloads/answer_template.json. Use its field names, types, enums,
and ordering rules. Never invent keys or enum values.
- Use stable hub record IDs. Every finding, issue, risk, source, event,
and privilege entry has a stable identifier (
doc_id, finding_id,
source_id, entry_id, event_id, action_id). Use these IDs in your
output, not synthetic or generated keys.
- Return only the JSON object. No prose outside the JSON.
Workflow
- Read
input/prompt.txt for the task description and matter_id.
- Read
input/payloads/answer_template.json for the exact output schema.
- Read any context payloads (
request_context.json, review_scope.json,
matter_context.json) for category labels and task framing.
- Call
GET /api/schema to understand the data model.
- Pull matter-level data:
GET /api/matters and GET /api/subpoena-categories.
- Pull domain evidence from the relevant endpoints, filtered by
matter_id.
Use POST /api/query with X-API-Key: review-key-017 for joins or aggregations.
- Cross-reference records by stable IDs from the hub.
- Fill the template with sorts and enum choices per its ordering rules.
- Return the single JSON object.
Endpoint Reference
See references/endpoints.md for the full endpoint catalog and data model notes.
Task-Type Patterns
See references/analysis_patterns.md for detailed analysis patterns covering:
- Rolling production gap analysis
- Retention and litigation-hold gap review
- Cross-system remediation dashboard
- Production-readiness review
Key Analysis Rules
Computing unlogged privilege documents
For every privilege entry with issue_type = incomplete_log:
unlogged_count = withheld_count - logged_count.
Classifying retention events
post_hold_loss -> critical risk, requires disclose_preservation_issue.
should_exist_missing -> high risk.
active_system_loss / auto_purged -> medium risk.
policy_destroyed_pre_hold -> low risk, no_action_policy_loss.
Identifying production gaps
production_stats.status = zero_claim_contradicted with matching QC findings -> responsive documents exist but were not produced. Action: recode_and_produce.
production_stats.status = supplement_pending -> incomplete production.
production_stats.status = rolling_review -> still in review.
Identifying source gaps
custodian_sources with issue_tags containing lost -> source_lost.
custodian_sources with not_collected and personal_phone or personal_email -> source_missing, action collect_personal_device.
custodian_sources with archive_available -> remediation path via search_archive.
Privilege analysis
privilege_entries.third_party = 1 -> third_party_waiver, action waiver_assessment_and_disclosure.
privilege_entries.issue_type = over_designated -> documents withheld as privileged that may not be. Action: qc_remediation or privilege_recode_and_log.
QC finding cross-referencing
qc_findings.source_ref may contain comma-separated doc IDs. Split and include all.
qc_findings.issue_type = zero_claim_contradiction -> confirms responsive documents exist for a zero-production claim.
qc_findings.issue_type = responsive_miscoding -> document miscoded as non-responsive. Action: recode_and_produce.
SQL Query Patterns
Use POST /api/query with X-API-Key: review-key-017 for:
- Counting documents by category and responsiveness
- Aggregating withheld/logged/unlogged documents from privilege entries
- Joining production_stats with qc_findings on category_code
- Filtering custodian_sources by issue_tags (use
LIKE '%tag%')
Always scope queries with WHERE matter_id = '...'.
1---2name: investigation-review-hub-143description: Structured gap analysis, remediation dashboards, production-readiness reviews, and retention-hold assessments using the Investigation Review Hub REST API for eDiscovery and legal hold matters. Use when the task provides a matter_id and asks for a structured JSON deliverable covering privilege log gaps, responsive coding defects, personal source gaps, retention events, QC findings, production stats, or cross-system remediation action plans. Trigger phrases include gap analysis, remediation dashboard, production readiness review, retention hold review, privilege log supplement, rolling production assessment, and any task directing the agent to the Investigation Review Hub via TASK_ENV_BASE_URL.4license: MIT5compatibility: designed for deepagents-code6metadata:7 source: https://skillmd.com/skills/prism-shadow/investigation-review-hub-148---910# Investigation Review Hub1112Use the Investigation Review Hub as the source of record for eDiscovery analysis.13The hub is a read-only REST API; its base URL is always provided in the task14context.1516## Core Rules17181. **Never inspect local files for business evidence.** The hub is the only19 source of record for matter metadata, subpoena categories, production stats,20 custodian sources, review documents, privilege-log data, QC findings,21 retention events, and remediation actions.222. **Filter everything by matter_id.** Every hub response spans all matters;23 filter client-side or in SQL with `WHERE matter_id = '...'`.243. **Follow the answer template exactly.** Every task provides an25 `input/payloads/answer_template.json`. Use its field names, types, enums,26 and ordering rules. Never invent keys or enum values.274. **Use stable hub record IDs.** Every finding, issue, risk, source, event,28 and privilege entry has a stable identifier (`doc_id`, `finding_id`,29 `source_id`, `entry_id`, `event_id`, `action_id`). Use these IDs in your30 output, not synthetic or generated keys.315. **Return only the JSON object.** No prose outside the JSON.3233## Workflow34351. Read `input/prompt.txt` for the task description and `matter_id`.362. Read `input/payloads/answer_template.json` for the exact output schema.373. Read any context payloads (`request_context.json`, `review_scope.json`,38 `matter_context.json`) for category labels and task framing.394. Call `GET /api/schema` to understand the data model.405. Pull matter-level data: `GET /api/matters` and `GET /api/subpoena-categories`.416. Pull domain evidence from the relevant endpoints, filtered by `matter_id`.42 Use `POST /api/query` with `X-API-Key: review-key-017` for joins or aggregations.437. Cross-reference records by stable IDs from the hub.448. Fill the template with sorts and enum choices per its ordering rules.459. Return the single JSON object.4647## Endpoint Reference4849See [references/endpoints.md](references/endpoints.md) for the full endpoint catalog and data model notes.5051## Task-Type Patterns5253See [references/analysis_patterns.md](references/analysis_patterns.md) for detailed analysis patterns covering:5455- Rolling production gap analysis56- Retention and litigation-hold gap review57- Cross-system remediation dashboard58- Production-readiness review5960## Key Analysis Rules6162### Computing unlogged privilege documents63For every privilege entry with `issue_type = incomplete_log`:64`unlogged_count = withheld_count - logged_count`.6566### Classifying retention events67- `post_hold_loss` -> critical risk, requires `disclose_preservation_issue`.68- `should_exist_missing` -> high risk.69- `active_system_loss` / `auto_purged` -> medium risk.70- `policy_destroyed_pre_hold` -> low risk, `no_action_policy_loss`.7172### Identifying production gaps73- `production_stats.status = zero_claim_contradicted` with matching QC findings -> responsive documents exist but were not produced. Action: `recode_and_produce`.74- `production_stats.status = supplement_pending` -> incomplete production.75- `production_stats.status = rolling_review` -> still in review.7677### Identifying source gaps78- `custodian_sources` with `issue_tags` containing `lost` -> `source_lost`.79- `custodian_sources` with `not_collected` and `personal_phone` or `personal_email` -> `source_missing`, action `collect_personal_device`.80- `custodian_sources` with `archive_available` -> remediation path via `search_archive`.8182### Privilege analysis83- `privilege_entries.third_party = 1` -> `third_party_waiver`, action `waiver_assessment_and_disclosure`.84- `privilege_entries.issue_type = over_designated` -> documents withheld as privileged that may not be. Action: `qc_remediation` or `privilege_recode_and_log`.8586### QC finding cross-referencing87- `qc_findings.source_ref` may contain comma-separated doc IDs. Split and include all.88- `qc_findings.issue_type = zero_claim_contradiction` -> confirms responsive documents exist for a zero-production claim.89- `qc_findings.issue_type = responsive_miscoding` -> document miscoded as non-responsive. Action: `recode_and_produce`.9091## SQL Query Patterns9293Use `POST /api/query` with `X-API-Key: review-key-017` for:9495- Counting documents by category and responsiveness96- Aggregating withheld/logged/unlogged documents from privilege entries97- Joining production_stats with qc_findings on category_code98- Filtering custodian_sources by issue_tags (use `LIKE '%tag%'`)99100Always scope queries with `WHERE matter_id = '...'`.