@oevery/rsp 3.2.0 → 3.3.0

package/skills/rsp-tdd/SKILL.md

@@ -4,7 +4,7 @@ description: Drive one clear testable behavior owned by a selected RSP Change th
 license: MIT
 metadata:
   author: oevery
-  version: "2026.07.24.1"
+  version: "2026.08.17.1"
 ---
 
 # RSP TDD
@@ -13,10 +13,16 @@ Implement one clear behavior test-first and return evidence to its selected Chan
 
 ## Establish the cycle
 
-Require one selected, ready Change, clear observable behavior, test and production mutation authority, and an executable focused check. Read project authority, current worktree state, and the smallest owning production/test seam.
+Require one selected, ready Change, clear observable behavior, test and production mutation authority, and an executable focused check. Read project authority, current worktree state, and the smallest owning production and test seam.
 
 Stop when behavior or acceptance is unclear, mutation authority is missing, an existing symptom has an unexplained cause, the execution environment is unavailable, or a baseline failure cannot be separated. Return the same Change, evidence, and one needed next action.
 
+## Admit the behavior and test seam
+
+Before RED, confirm that the selected Change, Spec, or project authority owns the observable behavior and that a real production path can reach the proposed seam. Name the consequence, one distinct plausible regression, why existing evidence misses it, and the expected maintenance cost. A test cannot create product authority; an imagined edge state, test-only consumer, wrapper hop, or source-string convention is not sufficient by itself.
+
+If the proposed test would require a new public contract, fallback, compatibility path, or handling for a state the production producer cannot emit, stop for the owning decision instead of encoding the assumption as RED. Explicit user, Change, or project test requirements remain authoritative.
+
 ## RED
 
 Add the smallest focused test at the real owning seam. Run it before production mutation and observe the expected missing or incorrect behavior.
@@ -27,6 +33,8 @@ RED is not a syntax, fixture, environment, or unrelated baseline failure. A pass
 
 Make the minimum production change for GREEN, preserve scope, and rerun the same command. An unexplained failure returns to diagnosis without speculative widening.
 
+Do not widen the product contract merely to satisfy the test. If GREEN requires an unauthorized API, fallback, compatibility behavior, test-only hook, or unreachable-state branch, treat the test premise as invalid or return the missing owner decision; do not mutate production first and rationalize the contract afterward.
+
 ## REFACTOR
 
 Refactor only after GREEN for a concrete local improvement; rerun the focused check. Otherwise skip it.

package/skills/rsp-verify/SKILL.md

@@ -0,0 +1,40 @@
+---
+name: rsp-verify
+description: Verify one selected RSP Change against its declared evidence boundary without mutating product or workflow state.
+license: MIT
+metadata:
+  author: oevery
+  version: "2026.08.21.1"
+---
+
+# RSP Verify
+
+Run one bounded, read-only verification pass for an existing RSP WorkRef. Verify owns evidence collection and the verification result; it is a Discipline, not a router, controller, lifecycle owner, reviewer, or Git delivery capability.
+
+## Require a declared boundary
+
+Require one explicit WorkRef or one unambiguous focus marker, the selected Change or Group context, its declared `Verify` boundary, the comparison baseline, and the authority to run the named checks. Stop when the owner, scope, baseline, required evidence, or environment is ambiguous. A Group Brief is context only; verify the selected executable child or the explicitly named integration boundary.
+
+Read the nearest project instructions, Core or fallback, selected Change and Brief, relevant Specs and Decisions, current diff, blockers, and the smallest production path needed to understand the declared check. Do not invent checks from generic testability or replace an unexplained failure with a guessed assertion; route that case to `rsp-diagnose`.
+
+## Preserve read-only authority
+
+Do not edit product files, Changes, Specs, Decisions, focus markers, archives, configuration, or Git state. Do not start publication, deployment, approval, or human-acceptance actions. Running a declared local test, build, typecheck, lint, browser check, or other environment check is evidence collection only and retains the authority required by that command.
+
+Reuse the invoking Core or Manage contract. Core owns the single outer `ControlOutcome`, route, and continuation; Manage owns execution mode, delegation, result validation, acceptance, lifecycle closeout, and commit eligibility. Verify's canonical result is nested phase evidence, not a peer outer status. Verify does not select worker identity or isolation, derive `review-clean`, or claim `archiveReady`; any identity or independence evidence comes from the host.
+
+## Return one bounded result
+
+Return exactly one canonical result:
+
+- `pass`: every named required check passed within the declared boundary.
+- `fail`: at least one named required check failed.
+- `unavailable`: a required tool, dependency, service, credential, or environment could not be used.
+
+Every result includes `evidence_delta: new | none` to state independently whether the pass, failure, or unavailability produced evidence that changes the next diagnosis or correction. It also includes `boundary: unchanged | changed` to state independently whether the observed owner, paths, baseline, behavior, interface, scope, or authority still matches the declared boundary. Include the WorkRef, lane objective, effective authority, named checks, comparison baseline, observed diff boundary, decisive evidence, omissions, and stop boundary. Preserve exact result and field values as machine-facing values; human-facing narration follows the invoking response-language contract.
+
+`pass` proves only the declared verification boundary. It does not prove semantic review, durable writeback, archive readiness, commit eligibility, publication, deployment, approval, or human acceptance, and it does not grant lifecycle, Git, publication, or acceptance authority. Verify never self-reports worker identity or independence; Manage uses host observations when that distinction is required. Long-running verification remains active while its declared boundary and stop conditions hold; elapsed time, heartbeat, polling, and progress messages do not change the result.
+
+## Stop and return
+
+Stop with the invoking contract's canonical stop reason when evidence is missing, the environment is unavailable, the boundary changes, or a required check cannot be observed. On cancellation, do not begin conflicting work until the command or owned process is observed stopped. Return to Core or Manage with the result, decisive evidence, omissions, and the next owner; never turn an unavailable check into success and never retry without new evidence and safe replay inside the declared boundary.