fix: preflight destination skill collisions

skills/process-engine-pattern-author/SKILL.md

@@ -117,6 +117,42 @@ handed to review.
    separate manifest/artifact. Do not prescribe specific tools or mechanisms
    (shell, MCP, wrappers, reloads) — require coherence, not implementation
    doctrine.
+
+   **Destination-skill preflight (generated Turnstone skills).** Once a
+   generated skill's canonical Agent Skills `name` is selected and the
+   intended destination is Turnstone, perform a destination-state preflight
+   **when an authoritative native read-only skill-catalog capability is
+   available in the active generator context** — do not run it before the
+   proposed identity exists, and do not make it another pipeline stage.
+   Record exactly one of:
+   - **PROVEN PRESENT** — authoritative Turnstone catalog evidence shows an
+     existing destination skill whose identity materially collides with the
+     proposed generated skill. Preserve the proposed canonical name, the
+     existing skill identity/name, returned object identity where available,
+     and the evidence source.
+   - **PROVEN ABSENT** — use only when the authoritative Turnstone lookup
+     semantics genuinely support proving absence for the queried skill
+     identity. A generic search returning zero results is NOT automatically
+     PROVEN ABSENT.
+   - **UNPROVEN** — generator lacks the authoritative read surface; the
+     lookup cannot establish absence; the result is ambiguous; returned
+     identity cannot be resolved confidently; or the lookup fails. Never
+     guess.
+   Do not invent a fuzzy-matching or normalization engine: use the generated
+   Agent Skills canonical `name`, authoritative Turnstone skill identity/name
+   fields, and native matching semantics where provided. When returned
+   evidence reveals casing/display-name/identity ambiguity, preserve it — do
+   not silently collapse distinct identities without evidence about their
+   object identities. Broader semantic overlap may be surfaced as a heads-up
+   when materially useful, but is NOT a mandatory catalog-wide ritual.
+   If PROVEN PRESENT, you may propose possible dispositions for later
+   operator consideration (rename the new skill; retain/coexist
+   intentionally; later supersede the old skill; revise/update where
+   supported). You must NOT choose one automatically, and must NOT create,
+   update, enable, disable, delete, or supersede anything. Destination-state
+   evidence informs the draft; it does not grant mutation authority. This
+   scope is generated **skills** only — do not extend to personas, projects,
+   policies, templates, or every Turnstone artifact without evidence.
 8. **Assemble the package** — project/persona/skills/templates/governance as
    one bundle; skill folders mirror the spec layout (SKILL.md + references/ +
    optional scripts/ and assets/) when shipped as repo content.
@@ -162,4 +198,5 @@ handed to review.
 - [ ] Governance artifacts: prompt policy + advisory judge rules (Turnstone-native)
 - [ ] Acceptance criteria written
 - [ ] Executable capability path identified where procedures depend on tools/capabilities; package declarations internally consistent with it (no prescribed tool/mechanism doctrine)
+- [ ] Destination-skill preflight recorded for generated Turnstone skills when authoritative native catalog read is available (PROVEN PRESENT / PROVEN ABSENT / UNPROVEN); no mutation; no automatic disposition
 - [ ] Marked DRAFT, handed to review

skills/process-engine-review/SKILL.md

@@ -49,6 +49,23 @@ the gate.
    executable declaration in the package, REVISE. Review checks internal
    package coherence here; it does not prove runtime availability — that is
    Trial's job.
+
+   **Destination-skill evidence (generated Turnstone skills)**: for each
+   generated Turnstone skill, check the recorded destination-state evidence
+   from Pattern:
+   - **PROVEN ABSENT** — collision dimension satisfied; do not manufacture a
+     collision concern.
+   - **PROVEN PRESENT** — require an explicit, reversible, operator-gated
+     disposition (e.g. rename before deployment; intentional coexistence;
+     create-new then operator-approved disable-old; another supported
+     reversible plan). Review checks the plan; it does NOT execute it.
+     Silent overwrite / accidental duplicate assumptions → REVISE.
+   - **UNPROVEN** — for a generated Turnstone skill intended for deployment,
+     REVISE with the missing destination-state evidence identified. Do not
+     let UNPROVEN silently become PASS merely because Ship later has an
+     operator gate.
+   Review performs zero catalog mutation: no create, update, enable, disable,
+   delete, or supersede. Ship/operator authority remains unchanged.
 3. **Safeguard review** (risk-relevant intents only): per-package safeguards
    present, evidence-named, sized to the domain — no preset doctrine
    (references/safety.md).
@@ -109,6 +126,7 @@ the gate.
 - [ ] Native parse evidence recorded in Review (skill identity, parse invoked, result)
 - [ ] Allowed-field comparison performed against Agent Skills contract; no unexpected/non-spec frontmatter keys present (REVISE if present, regardless of parse result)
 - [ ] Capability-path consistency checked for procedures requiring tools/capabilities (REVISE if no executable declaration)
+- [ ] Destination-skill evidence checked for generated Turnstone skills (PROVEN ABSENT ok; PROVEN PRESENT requires explicit reversible operator-gated disposition; UNPROVEN → REVISE for deployment-bound); zero catalog mutation
 - [ ] Safeguard review completed (risk-relevant intents only)
 - [ ] Coverage check completed against the catalog
 - [ ] Verdict recorded with evidence