@bookedsolid/reagent 0.3.0 → 0.4.0

package/agents/engineering/design-system-developer.md

@@ -65,7 +65,7 @@ CONSTRAINTS:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/design-systems-animator.md

@@ -73,7 +73,7 @@ CONSTRAINTS:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/devops-engineer.md

@@ -138,7 +138,7 @@ Discover the project's branching strategy from existing branches and configurati
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/drupal-integration-specialist.md

@@ -202,7 +202,7 @@ CONSTRAINTS:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/drupal-specialist.md

@@ -99,7 +99,7 @@ You are the Drupal specialist for this project. The team has deep Drupal roots a
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/engineering-manager-frontend.md

@@ -109,7 +109,7 @@ You lead the frontend team to build beautiful, fast, accessible UIs.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/frontend-specialist.md

@@ -54,7 +54,7 @@ Follow existing patterns in the codebase for:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/infrastructure-engineer.md

@@ -50,7 +50,7 @@ Adapt your patterns to what the project actually uses.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/lit-specialist.md

@@ -57,7 +57,7 @@ Web components in SSR require special handling:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/migration-specialist.md

@@ -100,7 +100,7 @@ CREATE INDEX CONCURRENTLY idx_table_column
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/ml-engineer.md

@@ -90,7 +90,7 @@ DON'T USE THIS AGENT FOR:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/mobile-engineer.md

@@ -164,7 +164,7 @@ You are the Mobile Engineer Lead for this project. You lead mobile strategy, dev
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/motion-designer-interactive.md

@@ -91,7 +91,7 @@ You create motion that users remember.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/nextjs-specialist.md

@@ -119,7 +119,7 @@ export async function POST(request: NextRequest) {
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/open-source-specialist.md

@@ -82,7 +82,7 @@ You are the Open Source specialist for this project. You advise on OSS strategy,
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/performance-engineer.md

@@ -78,7 +78,7 @@ Adapt your patterns to what the project actually uses.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/performance-qa-engineer.md

@@ -90,7 +90,7 @@ DON'T USE THIS AGENT FOR:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/pr-maintainer.md

@@ -66,7 +66,7 @@ git -C /path/to/.worktrees/branch-name merge origin/dev
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/principal-engineer.md

@@ -71,7 +71,7 @@ Key integration challenge: Ensure all rendering strategies coexist correctly wit
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/privacy-engineer.md

@@ -84,7 +84,7 @@ DON'T USE THIS AGENT FOR:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/qa-engineer.md

@@ -149,7 +149,7 @@ it('dispatches click event when clicked', async () => {
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/security-engineer.md

@@ -122,7 +122,7 @@ Adapt your patterns to what the project actually uses.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/security-qa-engineer.md

@@ -83,7 +83,7 @@ DON'T USE THIS AGENT FOR:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/senior-backend-engineer.md

@@ -291,7 +291,7 @@ When assigned backend work, you assess the domain (API, auth, data pipeline, med
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/senior-database-engineer.md

@@ -43,7 +43,7 @@ You execute the Database Architect's vision with precision and reliability.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/senior-frontend-engineer.md

@@ -106,7 +106,7 @@ DON'T USE THIS AGENT FOR:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/senior-product-manager-platform.md

@@ -20,7 +20,7 @@ Core product/marketing role supporting growth and user experience.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/senior-technical-project-manager.md

@@ -42,7 +42,7 @@ You keep infrastructure projects on track.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/site-reliability-engineer-2.md

@@ -43,7 +43,7 @@ You support the SRE Lead in maintaining 99.9%+ uptime.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/solutions-architect.md

@@ -59,7 +59,7 @@ When evaluating technology:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/sre-lead.md

@@ -114,7 +114,7 @@ You are the guardian of platform reliability.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/staff-engineer-platform.md

@@ -219,7 +219,7 @@ Every system you design supports the project's users. Every standard you set pro
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/staff-software-engineer.md

@@ -51,7 +51,7 @@ CONSTRAINTS:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/storybook-specialist.md

@@ -133,7 +133,7 @@ CONSTRAINTS:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/supabase-specialist.md

@@ -86,7 +86,7 @@ const channel = supabase
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/technical-project-manager.md

@@ -41,7 +41,7 @@ You ensure product features ship on time and on quality.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/technical-writer.md

@@ -120,7 +120,7 @@ Remember: You are creating authoritative reference documentation. Every page mus
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/test-architect.md

@@ -84,7 +84,7 @@ CONSTRAINTS:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/typescript-specialist.md

@@ -84,7 +84,7 @@ declare namespace astroHTML.JSX {
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/ux-researcher.md

@@ -26,7 +26,7 @@ UX Researcher, reporting to Senior Product Designer.
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/agents/engineering/vp-engineering.md

@@ -63,7 +63,7 @@ Every deliverable must pass ALL gates before merge:
 2. **Never trust LLM memory** — Verify current state via tools, git, and file reads. Programmatic project memory (`.claude/MEMORY.md`, `.reagent/`) is OK
 3. **Verify before claiming** — Check actual state (build output, test results, git status) before reporting status
 4. **Validate dependencies** — Verify packages exist (`npm view`) before installing; check version compatibility
-5. **Graduated autonomy** — Respect reagent L0-L4 levels from `.reagent/policy.yaml`
+5. **Graduated autonomy** — Respect reagent L0-L3 levels from `.reagent/policy.yaml`
 6. **HALT compliance** — Check `.reagent/HALT` before any action; if present, stop immediately
 7. **Audit awareness** — All tool invocations may be logged; behave as if every action is observed
 

package/dist/cli/commands/init.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"init.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/init.ts"],"names":[],"mappings":"AASA,wBAAgB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,IAAI,CAuH5C"}
+{"version":3,"file":"init.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/init.ts"],"names":[],"mappings":"AASA,wBAAgB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,IAAI,CA0H5C"}

package/dist/cli/commands/init.js

@@ -62,9 +62,11 @@ export function runInit(args) {
     }
     // Step 8: Policy
     results.push(...installPolicy(targetDir, profileName, profile, dryRun));
-    // Step 9: Agent team
+    // Step 9: Gateway config
+    results.push(...installGatewayConfig(targetDir, dryRun));
+    // Step 10: Agent team
     results.push(...installAgents(targetDir, dryRun));
-    // Step 10: Claude commands
+    // Step 11: Claude commands
     results.push(...installClaudeCommands(targetDir, dryRun));
     // Summary
     console.log('');
@@ -91,7 +93,7 @@ export function runInit(args) {
     if (!dryRun) {
         console.log('\n✓ reagent init complete');
         console.log('\nCommit these files (safe to commit):');
-        console.log('  git add .cursor/rules/ .husky/ .claude/commands/ CLAUDE.md .reagent/policy.yaml && git commit -m "chore: add reagent zero-trust config"');
+        console.log('  git add .cursor/rules/ .husky/ .claude/commands/ CLAUDE.md .reagent/policy.yaml .reagent/gateway.yaml && git commit -m "chore: add reagent zero-trust config"');
         console.log('');
         console.log('Do NOT commit (gitignored — stays on your machine):');
         console.log('  .claude/hooks/');
@@ -369,6 +371,53 @@ notification_channel: ""
     }
     return [{ file: '.reagent/policy.yaml', status: 'installed' }];
 }
+function installGatewayConfig(targetDir, dryRun) {
+    const PKG_VERSION = getPkgVersion();
+    const reagentDir = path.join(targetDir, '.reagent');
+    const gatewayPath = path.join(reagentDir, 'gateway.yaml');
+    // Idempotent: skip if gateway.yaml already exists
+    if (fs.existsSync(gatewayPath)) {
+        return [{ file: '.reagent/gateway.yaml', status: 'skipped' }];
+    }
+    if (!dryRun) {
+        fs.mkdirSync(reagentDir, { recursive: true });
+        const content = `# .reagent/gateway.yaml — generated by @bookedsolid/reagent v${PKG_VERSION}
+# Defines downstream MCP servers that reagent proxies through its middleware chain.
+#
+# Usage:
+#   reagent serve
+#
+# Each server entry spawns a child process and communicates over stdio MCP.
+# Tool names are namespaced as: <server-name>__<tool-name>
+#
+# Tier classification (convention-based, overridable):
+#   read:        get_*, list_*, search_*, query_*, read_*, fetch_*, check_*, health_*, describe_*, show_*, count_*
+#   destructive: delete_*, drop_*, purge_*, remove_*, destroy_*, ban_*, kick_*, revoke_*, truncate_*
+#   write:       everything else (default)
+
+version: "1"
+
+# To add a server, replace "servers: {}" with:
+#
+# servers:
+#   my-server:
+#     command: npx
+#     args: ['-y', 'my-mcp-server@latest']
+#     env:
+#       API_KEY: '\${MY_API_KEY}'
+#     tool_overrides:
+#       dangerous_action:
+#         tier: destructive
+#       admin_delete:
+#         tier: destructive
+#         blocked: true
+
+servers: {}
+`;
+        fs.writeFileSync(gatewayPath, content, 'utf8');
+    }
+    return [{ file: '.reagent/gateway.yaml', status: 'installed' }];
+}
 function installAgents(targetDir, dryRun) {
     const agentsSrcDir = path.join(PKG_ROOT, 'agents');
     const agentsDestDir = path.join(targetDir, '.claude', 'agents');