Use when creating the Bronze Layer Specification for raw/near-raw storage, source fidelity, ingestion metadata, schema drift, replay, quarantine, retention, access control, and lineage before Silver design.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Use when creating the Bronze Layer Specification for raw/near-raw storage, source fidelity, ingestion metadata, schema drift, replay, quarantine, retention, access control, and lineage before Silver design.
des-bronze-layer-design
Purpose
Use this skill to create and validate the Bronze Layer Specification for any data engineering project.
This skill defines how raw or near-raw ingested data will be stored, organized, partitioned, retained, audited, protected, and prepared for downstream Silver transformation.
The Bronze layer should preserve:
source truth;
raw payload or near-raw records;
ingestion evidence;
audit metadata;
source traceability;
schema drift evidence;
replay/backfill capability;
quarantine/rejected data evidence;
raw data security controls.
The goal is to design Bronze as a trustworthy raw evidence layer, not as a cleaned business layer.
In the Phase-Orchestrated Support Model, this phase is not Done when the Bronze Layer Specification is first written.
Phase 09 is Done only when:
Bronze Layer Specification exists
+ Phase 08 artifact and handoff are reviewed
+ Bronze validation work is identified
+ Bronze evidence is collected or waived with reason
+ artifact is revised from evidence
+ Phase 09 Done Gate passes
+ Phase 09 to Phase 10 handoff exists
When To Use
Use this skill when:
Phase 08 Ingestion Specification exists;
Phase 08 handoff exists or the user explicitly accepts the risk of continuing without it;
sources and ingestion patterns are known;
raw/landing storage needs to be designed;
source data must be retained for audit, replay, lineage, or troubleshooting;
schema drift, raw payload retention, partitioning, storage format, access control, retention, or quarantine behavior is unclear;
the workflow router selects Phase 09.
Do not use this skill to design Silver conformance, business cleaning, business deduplication, Gold marts, semantic models, dashboards, APIs, transformations, orchestration implementation, CI/CD workflow files, or code.
Allowed in this phase:
Bronze dataset boundary design
raw preservation design
storage format and organization decision
partitioning design at Bronze level
metadata column definition
schema drift/rescued/quarantine behavior design
replay/backfill support design
retention and access classification
Bronze boundary technical quality expectations
Define Bronze boundary quality checks and operational evidence expectations.
Ask HALT questions for unresolved storage, partition, metadata, retention, access, drift, and quarantine decisions.
Draft the Bronze Layer Specification.
Create the Phase 09 Support Plan.
Collect or reference Phase 09 evidence.
Revise the Bronze Layer Specification using evidence.
Run the Phase 09 Done Gate.
Create the Phase 09 to Phase 10 Handoff.
Update workflow status.
Recommend the next skill.
Do not execute this overview directly. Follow the step files.
Phase-Orchestrated Support Model
Phase 09 uses Bronze validation support work.
The purpose is not Silver cleaning or Gold modeling. The purpose is to make raw/source-aligned storage safe, traceable, replayable, auditable, and useful for downstream transformation.
If evidence is missing, mark the item as Draft, Open, Risk, Deferred, Blocked, Unknown, or Waived with reason.
HALT Policy
This skill must stop when a required decision cannot be safely inferred.
Stop especially when:
upstream ingestion specifications are unknown;
source payloads are blocked.
Detailed HALT checkpoints are defined in steps/.
Guardrails
The agent must not:
treat Bronze as a cleaned business layer;
over-clean or conform source data in Bronze;
remove raw fields without explicit retention and replay decision;
flatten or parse raw payload in a way that destroys source evidence;
choose partition keys without considering cardinality, query/replay patterns, file size, and cost;
hide schema drift;
store sensitive raw data without access policy;
drop bad records without quarantine/evidence;
define business deduplication rules in Bronze unless explicitly limited to ingestion safety;
mark Bronze datasets Ready if mandatory metadata, retention, drift, replay, access, and quarantine policies are missing;
design Silver/Gold tables in this phase;
write pipeline implementation code.
Quality Checklist
Phase 08 Ingestion Specification exists or Draft continuation is explicitly accepted.
Phase 08 handoff exists or missing handoff risk is explicitly accepted.
Each P1/P2 source maps to a Bronze dataset with name, source, and object/feed details.
Raw preservation policy is documented.
Storage format is decided and justified.
File/table organization is documented.
Partitioning strategy is chosen based on replay, maintenance, cost, and cardinality.
Mandatory metadata fields for audit and lineage are defined.
Ingestion audit metadata is defined.
Source traceability metadata is defined.
Schema drift policy is documented.
Replay and backfill support is specified.
Idempotency and deduplication boundary is clear.
Quarantine and rejected data rules are established.
Retention lifecycle policies are set.
Access control and sensitive data rules are defined.
Bronze boundary quality expectations are technical/source-level only.
Phase 09 support plan exists or is explicitly waived with reason.
Phase 09 evidence pack exists or evidence is explicitly waived with reason.
Phase 09 artifact revision notes exist.
Phase 09 Done Gate result is recorded.
Phase 09 to Phase 10 handoff exists.
The artifact does not design detailed Silver/Gold tables, business transformations, semantic models, dashboards, APIs, orchestration implementation, or code.
Anti-Patterns to Avoid
Anti-Pattern
Why It Fails
Applying business logic or conformance in Bronze
Breaks the raw replay guarantee; Silver cannot recover if business rules change.
Dropping raw source fields permanently
Causes irreversible data loss; fields needed later cannot be recovered.
Partitioning by high-cardinality keys
Creates small-file and metadata overhead problems.
Storing raw JSON at scale without a queryable strategy
Can increase scan cost and make downstream processing slow.
Missing metadata columns
Makes lineage, debugging, replay, and audit nearly impossible.
Treating quarantined data as disposable
Loses evidence needed for source debugging and quality analysis.
Giving broad access to raw sensitive data
Bronze often contains unmasked and unstable sensitive fields.
Handoff To The Next Skill
Recommend des-silver-layer-design only after:
Bronze Layer Specification exists
+ Phase 09 support plan exists or is waived with reason
+ Phase 09 evidence pack exists or evidence is waived with reason
+ Phase 09 Done Gate is Pass or Pass with risks
+ Phase 09 to Phase 10 handoff is Ready or Ready with Risks
If the phase is Draft or Blocked, recommend one of:
continue des-bronze-layer-design
return to Step 02 Bronze dataset and storage decisions
resolve HALT question
route back to des-ingestion-design
route back to des-architecture-design
des-wise
des-correct-course