Conciseness

Mostly efficient — the routing table, safety tiers, and troubleshooting table are dense with non-obvious RDS-specific facts Claude would not reliably know ("RDS Oracle is VPC-only", "no SYS/SYSTEM logins", NAE/return-path NACL rules) — but the ~60-line "Rubric-Critical Facts to Always Surface" section duplicates content already in troubleshooting.md, compute-runtime.md, and cman-proxy.md (e.g. the DPI-1047 thin-mode fix and the ORA-12170 checklist appear in both places), and meta-commentary like "the rubric greps for these" and "a common misdiagnosis that the rubric catches" earns no tokens for the reader. Version-pinned examples ("claude-sonnet-4-20250514", "19.0.0.0.ru-2024-01") are time-sensitive and not placed in a deprecated/old-patterns section. Not 4: the duplication and rubric-gaming filler are more than "minor instances of over-explanation"; not 2 because nothing explains concepts Claude already knows and the core reference-routing content is tight.

Actionability

Concrete, executable commands appear throughout: `aws ec2 authorize-security-group-ingress --group-id sg-123 --protocol tcp --port 1521 --source-group sg-456`, `nc -zv <rds-endpoint> 1521`, `aws rds describe-db-instances --query 'DBInstances[0].Endpoint'`, `SELECT value FROM v$parameter WHERE name = 'service_names'`, plus five bundled diagnostic scripts. Not 5: the body itself contains no complete copy-paste connection-code example (deferred to the language references, which is acceptable per the code-vs-instruction note but leaves minor gaps), and some directives like "Name 'eksctl', 'OIDC', 'iamserviceaccount' explicitly" tell the agent what to mention rather than giving the runnable command.

Workflow Clarity

A clear three-stage sequence (Verify Dependencies → Classify and Route → Execute Workflow) with genuine validation checkpoints for the destructive operations the skill guards: explicit user-confirmation gates before any create/modify, downtime warnings before instance-class/engine changes, a refuse-and-offer-assessment path for deletes/reboots, and a dependency check that must precede code generation. The troubleshooting table gives error→fix feedback loops. Not 5: some checkpoints are implicit rather than looped — e.g. after running test_connectivity.sh there is no explicit "if this fails, go here" branch in the body, and parameter validation states formats but no re-prompt loop.

Progressive Disclosure

Good structure: the body routes to twelve real one-level-deep reference files via a symptom→file table (all verified to exist and be substantive), instructs "Load only the matching reference", and lists the five scripts with links. Not 5: the inlined "Rubric-Critical Facts" section is exactly the detailed per-scenario content that belongs in troubleshooting.md/compute-runtime.md/cman-proxy.md (anchor-3 symptom present), and the handoff section reaches two levels deep into another skill's tree (`aws-database-selection/references/handoff-contract.md` → `workload-primary-artifact.schema.json`). These are minor organization gaps rather than structural ones, so it sits at 4.

Specificity

Opens with concrete actions ("Diagnoses and resolves Amazon RDS for Oracle connectivity, authentication, networking, and driver troubleshooting") and then enumerates specific concrete scenarios and tools ("connecting a Python Lambda to RDS Oracle in a VPC with pooling and cold-start optimization", "EKS pods... via the Secrets Manager CSI driver with IRSA", "ORA-12170 cross-VPC timeouts", "DPI-1047", "CMAN on EC2 as a proxy with HA across two AZs"). Coverage is comprehensive and nothing is vague. Not 4 because the minor gaps (no mention of e.g. ORA-12541/ORA-12514 in the description itself) are negligible against the breadth of concrete capability named.

Completeness

Both "what" ("Diagnoses and resolves...") and "when" ("Applicable to any RDS-for-Oracle question including...") are explicitly present, so it clears the anchor-4 bar, but the when-clause is padded into a very long enumeration rather than concise explicit trigger guidance, and the trailing "Covers..." sentence adds keyword laundry-list material that dilutes rather than sharpens the when. Not 5: the guideline "do not reward verbosity; concise and clear beats long and padded" — the when guidance is explicit but buried at the end of ~180 words, so it does not read as the crisp dual what+when statement of the anchor-5 example.

Trigger Term Quality

Natural user vocabulary is comprehensively covered with synonyms and product-specific tokens users would actually type: error codes ("ORA-12170", "DPI-1047"), tool names ("SQL Developer/DBeaver/Toad/SQLcl"), APIs ("port 1521", "VPC peering", "Transit Gateway", "SSM port forwarding"), and driver keywords ("python-oracledb thin vs thick mode", "init_oracle_client"). Not 4: no common natural term or synonym is observably missing for this domain.

Distinctiveness Conflict Risk

Clear niche with distinct triggers: every phrase is scoped to "Amazon RDS for Oracle" specifically, and it even disambiguates the nearest neighbor products ("RDS Proxy does NOT support RDS Oracle"). Minimal conflict risk with general Oracle, EC2, or other database skills. Not 4: there is no meaningful overlap risk — the trigger terms (ORA-12170 from EC2, CMAN on EC2, RDS Oracle endpoints) are unambiguous for this one skill.