@rudderhq/agent-runtime-gemini-local 0.2.1 → 0.2.2-canary.0

package/skills/skill-optimizer/references/universal-optimization-lens.md

@@ -0,0 +1,73 @@
+# Universal Optimization Lens
+
+Use this when a skill optimization is complex, cross-domain, or likely to recur.
+
+## 1. Purpose and scope
+
+- What durable job does this skill own?
+- What is explicitly outside scope?
+- What user role or team does it serve?
+- Does the skill preserve its original identity after the patch?
+
+## 2. Triggering and boundaries
+
+- What should trigger the skill?
+- What near-miss queries should not trigger it?
+- Which other skills might compete with it?
+- Does the description include enough context without becoming keyword spam?
+
+## 3. Inputs and assumptions
+
+- What required inputs must be discovered before work begins?
+- What source of truth should be used?
+- What units, locale, time zone, standard, or policy applies?
+- What should happen when information is missing?
+
+## 4. Workflow and decision rules
+
+- Which steps must happen before others?
+- Where are the branch points?
+- What are stop conditions?
+- Which rules should be deterministic rather than left to judgment?
+
+## 5. Tools and authority
+
+- Which tools are required?
+- Which operations are read-only vs write actions?
+- Where is dry-run needed?
+- What needs explicit approval?
+
+## 6. Outputs and interfaces
+
+- What artifact should be produced?
+- What template or schema should be followed?
+- What links, citations, IDs, or files must be included?
+- Who consumes the output next?
+
+## 7. Quality and evaluation
+
+- What does success mean?
+- What validation cases prove the change?
+- What regression case prevents old behavior from breaking?
+- Is there a deterministic verifier or checklist?
+
+## 8. Safety, privacy, and policy
+
+- What sensitive data could appear?
+- What regulated advice or consequential decision is involved?
+- What needs consent, review, or audit trail?
+- What data should be minimized, redacted, or excluded?
+
+## 9. Failure and recovery
+
+- What should the skill do when blocked?
+- What retries are safe?
+- What rollback or cleanup path is required?
+- What partial result is useful?
+
+## 10. Maintainability
+
+- Is the core `SKILL.md` concise?
+- Should details move into references or scripts?
+- Are examples realistic?
+- Is there a changelog and version note?