Trigger: reviewer transport, advisory transport, review adapter, lens prompt/schema, OpenCode reviewer plugin, Codex reviewer, ReviewProviderContract. Enforce the shared Go advisory reviewer transport contract.
SkillScore84Maintained
Automatic
◆Coding pack
Last verified 9/29/2026
What it changes
Passed 18 of 21 test cases with the skill, and 3 of 21 without it.
Without skill12%
With skill84%
Checks it passed that the model alone missed
Prohibits building parallel Go contract systems.
Requires all benchmark journeys to remain unchanged.
Enforces a 1000-line limit per PR slice for cutover changes.
and 22 more in the full report
18 fixed3 passed either way4 still failing0 broke
Pass-rate lift
+72 pp
Tokens
-12%
Agent turns
0%
Response time
-55%
Model gemini-3.6-flashMeasured 9/29/2026v121 of 25 cases passed with skill · 21 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 18 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 18
Prompt
Before releasing a transport adapter slice, what benchmark verification step is required, and what outcome must be demonstrated across the journey corpus?
What was checked, without and with the skill
Checked
Without skill
With skill
Enforces running the benchmark journey corpus diff on every slice.
Missed
Passed
Requires all benchmark journeys to remain unchanged.
Missed
Passed
case-23 · gemini-3.6-flash · 9/29/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md44 lines3 files
Activation Contract
Load when changing how any runtime (Claude Code, OpenCode, Codex, future) invokes an RDD reviewer, refuter, or fix validator, or when touching reviewer prompts, lens evidence, result schemas, adapters, or transport capability policy.
Hard Rules
One Go provider contract owns binding, frozen evidence, lens mandate, prompt, result schema, byte budget, JSON extraction, and admission. Start from existing native sources (reviewLensContextBlock, ReviewerResultSchema, RunReviewCaptureResult, CaptureAdmittedReviewerResult); never build a parallel system.
Adapters only invoke the reviewer and return raw bytes plus a transport error. An adapter must never parse bindings, rebuild prompts, copy schemas, apply local budgets, parse JSON, capture or preserve results, hold retry state, or decide blocking.
Runtime output is advisory: it cannot create authority, mutate review state, mint a receipt, or open a gate. Only Go admission does.
Byte budget refuses before invocation; never truncate. Malformed, truncated, or incomplete input never becomes PASS, complete, or a clean receipt.
No OpenCode restart, child isolation, special session, or OPENCODE_DISABLE_* variables. An ordinary running session is sufficient.
The RDD semantic pipeline is untouchable: lens selection, severity meanings, candidate causality, refuter adjudication, bounded correction, verification, terminal receipt, and read-only review-context evaluation. These are review-lifecycle concerns only; ordinary repository policy owns delivery.
Refuter and fix validator consume the same provider contract. A validator that could not inspect produced no verdict — surface a blocked decision; never record it as failed (that irreversibly consumes the single correction).
Capability advertisement requires shared-contract conformance plus organic runtime proof. Codex stays unadvertised until both positive and fail-closed proofs pass.
Transport failures never consume a correction budget, duplicate a capture, or strand a lineage.
Historical receipts stay readable; never rewrite provider_command or runtime_interception descriptors.
Decision Gates
Condition
Action
Logic fits "prompt/evidence/schema/validation"
It belongs in the Go contract, never an adapter.
Adapter cannot expose raw final output
Runtime stays typed unavailable; no artifact-parser workaround.
Cutover slice exceeds 1000 lines/PR
Chained PRs on the tracker; advertisement flip lands last.
Shared prompt changes any semantic outcome
STOP; regression against baseline blocks the slice.
Coordinate with the deferred Wave 7 switch-removal finding first.
Execution Steps
Freeze the boundary: contract-level tests around current native prompt/evidence/schema/admission.
Extract the Go provider contract; differential-test against current admission.
Add adapter-minimality static guards (no binding types, prompt text, schema copies, budgets, parsers, capture, retry in adapters) — from the first adapter slice, not last.
Cut Claude + OpenCode over without an advertised intermediate path; delete the OpenCode plugin and isolation machinery in the same release slice.
Prove Codex organically before advertising; retire historical transport code last.
Gate every slice on the bench journey corpus diff: all journeys unchanged.
Output Contract
Report per slice: contract surface touched, adapter minimality guard status, bench corpus diff result, semantic regression result, capability advertisement state, and any unresolved maintainer decision (e.g. receipt provider_contract descriptor).