fix: require native activation of routed stages

skills/process-engine-core/SKILL.md

@@ -74,6 +74,32 @@ and governance surfaces.
      validate → shortlist with provenance → operator chooses (never just
      re-ask for material)
    - unclear → ask one clarifying question (never guess scope).
+
+   **Routing decision vs native stage activation.** Routing names where the
+   work belongs; it does NOT by itself make that stage govern. A Process
+   Engine stage is not active merely because another stage names, routes to,
+   or hands off to it. Before executing any routed Process Engine stage,
+   ensure the canonical target stage is **proven natively active**:
+   1. select the correct stage;
+   2. if the exact canonical target skill is not already proven active in the
+      current execution context, **natively activate/load it** through
+      Turnstone's native skill mechanism; if it is already proven active, do
+      not reload it merely for ceremony;
+   3. confirm the intended skill actually governs (loaded/applied skill
+      identity/content evidence, not memory or prose);
+   4. only then execute that stage.
+   "Already active" means the exact canonical target stage is proven to
+   govern the current execution context — never the model remembering it
+   loaded earlier, the stage name appearing in prose, another Process Engine
+   stage being active, an old/stale version having been loaded, or a prior
+   workstream having loaded it. Do not emulate an unloaded stage from memory,
+   Core summaries, references, or prior context. If native activation fails
+   or cannot be established, do not silently continue as that stage — report
+   the missing stage activation and stop at that transition until the
+   intended skill can govern. Use the native activation mechanism available
+   in the active context; the behavioral requirement is that the canonical
+   stage skill governs, not a specific API call. Do not invent a state
+   machine or custom router.
 7. **Load standards checklist** (references/standards.md) and the generation
    basis (references/best-practices.md — full Osmani catalog index) and apply
    them to every step.

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

@@ -180,7 +180,16 @@ handed to review.
 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.
-9. **Hand off** — mark DRAFT, hand the package to `process-engine-review`.
+9. **Hand off** — mark DRAFT and hand the package to `process-engine-review`.
+   Naming Review does not make it active: before Review executes, ensure the
+   exact canonical `process-engine-review` skill is **proven natively active**
+   in the current execution context; if it is not already proven active,
+   **natively activate/load it** through Turnstone's native skill mechanism
+   and confirm the canonical Review skill governs. If it is already proven
+   active, do not reload it merely for ceremony. Do not emulate Review from
+   this skill, Core, references, or prior context. If the native activation
+   cannot be established, stop at the handoff and report the missing stage
+   activation — do not proceed as if Review ran.
 
 ## Examples
 - "I want a skill that writes release notes" → package: project + persona

skills/process-engine-review/SKILL.md

@@ -103,15 +103,36 @@ the gate.
 8. **Operator gate**: present verdict and the draft; only operator sign-off
    advances to trial/ship. Log the verdict.
 
+**Native stage activation on transitions.** Naming the next stage does not
+make it active. Before any stage executes after a transition, ensure the
+exact canonical target Process Engine skill is **proven natively active** in
+the current execution context; if it is not already proven active,
+**natively activate/load it** through Turnstone's native skill mechanism and
+confirm it governs. If it is already proven active, do not reload it merely
+for ceremony. "Already active" means the exact canonical target stage is
+proven to govern the current execution context — not memory, stage-name
+prose, another PE stage being active, a stale version, or a prior workstream.
+Do not emulate an unloaded stage from this skill, Core, references, or prior
+context. If activation cannot be established, stop at the transition and
+report the missing stage activation — do not silently continue as the target
+stage.
+
 ## REVISE loop (formal)
 
 1. **Diagnose** — the verdict names each specific finding (what, where, why).
-2. **Rewrite** — pattern-author revises the artifact against the findings,
-   preserving intent.
-3. **Audit** — the artifact returns to review; the re-review confirms each
-   finding is addressed and the result is materially better than before.
-   Regression trial re-runs after any change that affects its performance
-   or scope (process-engine-trial).
+2. **Rewrite** — before Pattern performs the rewrite, ensure the exact
+   canonical `process-engine-pattern-author` skill is **proven natively
+   active**; if not already proven active, **natively activate/load it**.
+   Only the canonical Pattern skill may revise the artifact against the
+   findings, preserving intent.
+3. **Audit** — before re-review, ensure the exact canonical
+   `process-engine-review` skill is **proven natively active**; if not
+   already proven active, **natively activate/load it**. The re-review
+   confirms each finding is addressed and the result is materially better
+   than before. Regression trial re-runs after any change that affects its
+   performance or scope — before that regression runs, ensure the exact
+   canonical `process-engine-trial` skill is **proven natively active**; if
+   not, **natively activate/load it**.
 
 ## Examples
 - Generated package review: frontmatter spec-valid? acceptance criteria

skills/process-engine-triage/SKILL.md

@@ -32,6 +32,14 @@ categorizes feedback, proposes process revisions, and gates them through review.
    - DESIGN (experience flaw) → draft a process/artifact revision → review
    - IDEA (new capability) → log to roadmap; propose to operator
    - SAFETY (risk/scope/safety-adjacent report) → highest priority; review immediately
+   Routing names where the work belongs; it does not make the target stage
+   active. Before any routed Process Engine stage executes (pattern-author
+   fix, review, trial regression), ensure the canonical target stage skill is
+   **proven natively active**; if not already proven active, **natively
+   activate/load it** through Turnstone's native skill mechanism and confirm
+   it governs — do not emulate the stage from this skill or prior context.
+   If activation cannot be established, stop at the transition and report the
+   missing stage activation.
    Each report gets: severity (blocker / major / minor / nit), reproduction
    status (reproduced / not-reproduced / unknown), and deduplication (link to
    an existing report if it is the same defect). Stars/forks are **adoption

skills/process-engine-trial/SKILL.md

@@ -80,6 +80,21 @@ per-package spec.
    in Trial until evidence is complete. FAIL routes to pattern-author
    (revisions) or review (re-verify).
 
+**Native stage activation on transitions.** Naming the next stage does not
+make it active. For FAIL requiring package revision, ensure the appropriate
+canonical stage skill is **proven natively active** (`process-engine-pattern-author`
+for rewrites or `process-engine-review` for re-verification) before that
+stage executes; if not already proven active, **natively activate/load it**.
+For PASS, do not execute Ship merely because Ship is named: present/retain
+the operator Ship gate, and only once Ship is separately authorized should
+`process-engine-ship` be **proven natively active** (activated/loaded if not
+already) for deployment work. "Already active" means the exact canonical
+target stage is proven to govern the current execution context — not memory,
+stage-name prose, another PE stage, a stale version, or a prior workstream.
+Do not emulate an unloaded stage from this skill, Core, references, or prior
+context; if activation cannot be established, stop at the transition and
+report the missing stage activation.
+
 ## Examples
 - Case: out-of-scope request → input outside the package's declared domain →
   expected: decline and route, never guess. Actual recorded.