@receiz/ai-skills 104.0.0 → 106.0.0

package/README.md

@@ -1,18 +1,20 @@
 # Receiz AI Skills
 
-`receiz-app-builder-skill` is the authoritative v104 workflow for starting,
-integrating, upgrading, diagnosing, and repairing repositories through the typed
-Receiz app contract, shared SDK compiler, CLI, and MCP tools.
+`receiz-build-production-system` is the v106 orchestration skill for substantial
+Receiz applications. Fourteen focused constitutional skills supply architecture,
+domain, law, command, authority, replay, offline, causal, artifact, migration,
+performance, observability, testing, and release contracts. The earlier nine
+domain skills remain packaged for compatibility.
 
 This package teaches AI agents how to understand, verify, build with, and operate Receiz from the existing repository surfaces.
 
-It is published as `@receiz/ai-skills` and is an exact-version dependency of both `@receiz/sdk` and `@receiz/mcp-server`. Installing either package places these nine skill directories under `node_modules/@receiz/ai-skills` for explicit loading by an agent host.
+It is published as `@receiz/ai-skills` and is an exact-version dependency of both `@receiz/sdk` and `@receiz/mcp-server`. Installing either package places all twenty-four skill directories under `node_modules/@receiz/ai-skills` for explicit loading by an agent host.
 
 Receiz is a proof-native artifact system. A Receiz object is not a database row. It is a proof-carrying artifact. The artifact's witnessed history is the truth boundary. Server state, database state, UI state, marketplace state, and model memory are projections only. Old witnessed truth is not stale. If a projection is wrong, rebuild it from object history. If truth is missing, append new truth. Never mutate witnessed truth.
 
 ## SDK, MCP, And Skills
 
-- SDK: `@receiz/sdk` is for building apps. It exposes validators, projections, proof memory, identity artifact helpers, public app-state, public store, commerce, wallet, sports, world, offline queues, webhooks, and sandbox helpers.
+- SDK: `@receiz/sdk` is the universal runtime; `@receiz/sdk/react` is React-only; `@receiz/sdk/compiler` is Node-only; `@receiz/sdk/testing` is browser-safe sandbox and conformance support.
 - MCP: `@receiz/mcp-server` is for agent tool access. It exposes SDK-backed tools for diagnostics, public reads, offline shape inspection, app-state/public-store actions, deterministic previews, and delegated writes. Artifact verification remains SDK `verification.verifyArtifact(file)` and requires continuity.
 - Skills: this package is operating doctrine. It tells an AI which Receiz primitive is active, which source of truth wins, which SDK or MCP rail to use, and what must never be assumed.
 
@@ -20,7 +22,7 @@ Agents may acquire scoped delegated access through the official Receiz Connect/O
 
 ## How Agents Should Use This Package
 
-Start with `receiz-skill-bundle` when the request spans multiple Receiz domains. Use `receiz-proof-skill` for any verification, proof object, vault, profile, pack, card, post, product, listing, receipt, marketplace, or offline artifact question. Use narrower skills only after the primitive is named.
+Start with `receiz-build-production-system` when work spans multiple constitutional domains. Its machine-readable manifest requires the exact active signed registry digest, command-only mutation, independent verification, MCP conformance, and a passing release lock. Use a focused v106 skill for narrower production work. The compatibility router `receiz-skill-bundle` and proof/domain skills remain available for earlier workflows.
 
 Do not invent APIs. Use the SDK and MCP maps inside the skills, then inspect the repo when a requested operation is not listed.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@receiz/ai-skills",
-  "version": "104.0.0",
+  "version": "106.0.0",
   "description": "Authoritative Receiz AI operating skills for proof-native SDK and MCP agents.",
   "type": "module",
   "files": [
@@ -13,7 +13,22 @@
     "receiz-offline-verifier-skill",
     "receiz-proof-skill",
     "receiz-skill-bundle",
-    "receiz-sports-card-skill"
+    "receiz-sports-card-skill",
+    "receiz-architecture",
+    "receiz-domain-builder",
+    "receiz-constitutional-laws",
+    "receiz-command-builder",
+    "receiz-authority-security",
+    "receiz-deterministic-replay",
+    "receiz-offline-first",
+    "receiz-causal-sync",
+    "receiz-portable-artifacts",
+    "receiz-migrations",
+    "receiz-performance",
+    "receiz-observability",
+    "receiz-testing",
+    "receiz-release",
+    "receiz-build-production-system"
   ],
   "keywords": [
     "receiz",

package/receiz-app-builder-skill/SKILL.md

@@ -31,13 +31,15 @@ integration tooling beneath Receiz proof truth; inspection is not verification.
 
 ## Required Behavior
 
+Select the package boundary first: universal runtime from `@receiz/sdk`, React bindings from `@receiz/sdk/react`, Node-only app compiler APIs from `@receiz/sdk/compiler`, and sandbox/conformance support from `@receiz/sdk/testing`.
+
 1. Inspect the explicit repository root with `receiz_project_inspect` or `receiz app inspect`.
 2. Identify intended features and create or update the contract with `defineReceizApp` or `receiz_app_contract_create`.
 3. Compile a read-only plan with `compileReceizAppContract`, `receiz_app_plan`, or `receiz app plan`.
 4. Explain scopes, authority boundaries, findings, and every proposed file.
 5. Obtain explicit confirmation for the exact preview digest before mutation.
 6. Apply the smallest safe change set with `receiz_app_apply` or `receiz app apply`.
-7. Run generated conformance and repository tests.
+7. Run generated conformance, `receiz conformance`, and repository tests.
 8. Reinspect with `receiz_app_check` and report exact evidence.
 9. Use `receiz_app_upgrade` for version-aware upgrades and `receiz_app_explain` for findings.
 10. Report remaining manual work and blocked authority without claiming completion.
@@ -56,18 +58,19 @@ upgrade, diagnosis, and repair routes.
 
 ## MCP Usage Rules
 
-Use only the seven documented tools listed in
+Use the documented tools listed in
 [resources/workflow.md](resources/workflow.md). Inspection, contract proposals,
 planning, checking, upgrade planning, and explanation are read-only.
 `receiz_app_apply` returns a preview first and requires the exact digest as
 explicit confirmation. Every response must retain the MCP authority boundary.
+Use `receiz_app_repair` for confirmed deterministic repair, `receiz_scope_explain` for canonical scopes, `receiz_release_qualify` for read-only release evidence, `receiz_proof_trace` only with canonical artifact bytes for a verdict, and `receiz_webhook_replay` only after signature verification and exact confirmation.
 
 ## SDK Usage Rules
 
-Use `defineReceizApp`, `validateReceizAppContract`,
+Import `defineReceizApp`, `validateReceizAppContract`,
 `compileReceizAppContract`, `inspectReceizProject`,
 `planReceizIntegration`, `checkReceizIntegration`, `planReceizUpgrade`, and
-`explainReceizIntegrationFinding`. Artifact verification remains
+`explainReceizIntegrationFinding` from `@receiz/sdk/compiler`. Artifact verification remains
 `verification.verifyArtifact`; creation remains the SDK's authenticated native
 Record-before-Seal rail.
 
@@ -94,6 +97,7 @@ Read [resources/authority-boundaries.md](resources/authority-boundaries.md).
 Stop when a required delegated permission, identity proof, capability, or user
 decision is unavailable. Report the blocked boundary; do not fabricate it.
 Generated browser code may contain public origins but never server credentials.
+`receiz.extensions.ts` is developer-owned and create-once; regeneration must never overwrite it. Emulator evidence is `sandboxVerified`, never a production verified verdict.
 
 ## Examples
 

package/receiz-app-builder-skill/resources/workflow.md

@@ -7,10 +7,11 @@
 | Add or remove a rail | Update the contract, then plan | Apply only the changed digest |
 | Upgrade | `receiz_app_upgrade` | Apply approved package/generated changes |
 | Diagnose | `receiz_app_check`, `receiz_app_explain` | None until the finding is understood |
-| Repair | Reinspect generated-file drift | Regenerate Receiz-owned files only |
+| Repair | `receiz_app_repair` preview | Exact digest confirmation; regenerate Receiz-owned files only |
 
 After every mutation, run the returned verification commands and reinspect.
 The same repository must produce zero changes on the second run.
+Preserve the developer-owned `receiz.extensions.ts` sibling on every run. Finish with `receiz conformance` or `receiz_release_qualify`.
 
 Do not substitute inspection for verification, publication for settlement, or a
 successful write for proof admission.

package/receiz-architecture/SKILL.md

@@ -0,0 +1,53 @@
+---
+name: receiz-architecture
+description: Use when inspecting or changing Receiz domains, system boundaries, truth hierarchy, state machines, or constitutional architecture.
+---
+
+# receiz-architecture
+
+Map domains, sources of truth, state machines, dependency direction, and cross-domain authority before design or refactoring. Receiz is a proof-native artifact system; this skill operates beneath sealed proof and verified local history.
+
+## Constitutional workflow
+
+1. Inspect the repository and installed SDK version.
+2. Load the active signed registry and applicable constitutional laws.
+3. Identify every authority boundary affected by the change.
+4. Write the implementation contract before implementation.
+5. Implement canonical changes only through SDK commands.
+6. Add law, mutation, replay, and compatibility tests.
+7. Run MCP conformance checks.
+8. Require independent verifier evidence.
+9. Refuse completion until the release lock passes.
+
+Focused outcome: Map domains, sources of truth, state machines, dependency direction, and cross-domain authority before design or refactoring.
+
+## Machine contract
+
+Read [manifest.json](manifest.json) before acting. Its package range, ruleset, active registry digest requirement, laws, allowed tools, forbidden operations, evidence, escalation conditions, and autonomous-authority ceiling are executable constraints. Use plan/permit/execute for any admitted command and obtain explicit confirmation for the exact permit digest. The skill and MCP context are never proof authority.
+
+## Quick reference
+
+| Boundary | Required result |
+|---|---|
+| Stronger truth | Sealed artifact and verified local history remain above SDK, MCP, server, DB, session, and UI projections. |
+| Mutation | Canonical state changes only through an allowed SDK command and atomic receipt-producing admission. |
+| Evidence | MCP output and agent assertions do not count; independent verification and release-lock evidence do. |
+| Escalation | Stop on digest skew, authority bypass, missing capability, ambiguous migration, or absent independent evidence. |
+
+## Common mistakes
+
+- Treating a larger or newer server snapshot as permission to replace stronger admitted truth.
+- Calling an MCP plan, simulation, hash, model response, or passing UI check verification.
+- Appending an event without a command admission receipt.
+- Marking the task complete while required evidence is missing.
+
+## Completion refusal
+
+Refuse completion when any manifest input, law, test, conformance result, independent-verifier result, or release-lock result is absent or failing. Never reduce the evidence list, change a test to bless a visible regression, invent authority, or accept “the AI said it is valid” as evidence.
+
+## Example
+
+Request: Map domains, sources of truth, state machines, dependency direction, and cross-domain authority before design or refactoring.
+
+Return a written implementation contract and read-only plan first. Execute only allowed commands after exact confirmation, then report registry digest, laws, authority, tests, independent evidence, release-lock status, and the exact remaining boundary.
+

package/receiz-architecture/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Receiz Architecture"
+  short_description: "Inspect constitutional architecture"
+  default_prompt: "Use -architecture to inspect a Receiz system's domains, truth hierarchy, and authority boundaries."

package/receiz-architecture/manifest.json

@@ -0,0 +1,57 @@
+{
+  "schema": "receiz.ai-skill-contract.v106",
+  "name": "receiz-architecture",
+  "requires": {
+    "sdk": ">=106.0.0 <107.0.0",
+    "mcp": ">=106.0.0 <107.0.0",
+    "ruleset": "106.0.0",
+    "registryDigest": "bf851c209e807309672c0f466411baa5607ce6b3195fe4eb16755edfeb7f5a1a"
+  },
+  "laws": [
+    "receiz.truth.stronger-first",
+    "receiz.registry.digest-alignment",
+    "receiz.command-only-mutation"
+  ],
+  "allowedTools": [
+    "receiz_architecture_inspect",
+    "receiz_authority_inspect",
+    "receiz_history_trace",
+    "receiz_bypass_scan"
+  ],
+  "forbiddenOperations": [
+    "direct-state-write",
+    "history-rewrite",
+    "authority-bypass"
+  ],
+  "requiredInputs": [
+    "repository-root",
+    "installed-sdk-version",
+    "active-registry-digest",
+    "applicable-laws",
+    "authority-boundaries"
+  ],
+  "requiredEvidence": [
+    "typecheck",
+    "law-coverage",
+    "mutation-tests",
+    "replay-equivalence",
+    "compatibility",
+    "mcp-conformance",
+    "independent-verifier"
+  ],
+  "completionCriteria": [
+    "written-implementation-contract",
+    "constitutional-context-digest-match",
+    "all-required-evidence-pass",
+    "release-lock-pass"
+  ],
+  "escalationConditions": [
+    "authority-bypass-detected",
+    "registry-digest-mismatch",
+    "missing-independent-evidence",
+    "missing-capability",
+    "ambiguous-migration"
+  ],
+  "maximumAutonomousAuthority": "inspect-plan-scaffold-test"
+}
+

package/receiz-authority-security/SKILL.md

@@ -0,0 +1,53 @@
+---
+name: receiz-authority-security
+description: Use when work touches capabilities, identity, PBI, tenancy, ownership, settlement, secrets, delegated access, or authority escalation.
+---
+
+# receiz-authority-security
+
+Audit actors, tenants, capabilities, proof domains, confused-deputy paths, escalation, and cross-boundary effects. Receiz is a proof-native artifact system; this skill operates beneath sealed proof and verified local history.
+
+## Constitutional workflow
+
+1. Inspect the repository and installed SDK version.
+2. Load the active signed registry and applicable constitutional laws.
+3. Identify every authority boundary affected by the change.
+4. Write the implementation contract before implementation.
+5. Implement canonical changes only through SDK commands.
+6. Add law, mutation, replay, and compatibility tests.
+7. Run MCP conformance checks.
+8. Require independent verifier evidence.
+9. Refuse completion until the release lock passes.
+
+Focused outcome: Audit actors, tenants, capabilities, proof domains, confused-deputy paths, escalation, and cross-boundary effects.
+
+## Machine contract
+
+Read [manifest.json](manifest.json) before acting. Its package range, ruleset, active registry digest requirement, laws, allowed tools, forbidden operations, evidence, escalation conditions, and autonomous-authority ceiling are executable constraints. Use plan/permit/execute for any admitted command and obtain explicit confirmation for the exact permit digest. The skill and MCP context are never proof authority.
+
+## Quick reference
+
+| Boundary | Required result |
+|---|---|
+| Stronger truth | Sealed artifact and verified local history remain above SDK, MCP, server, DB, session, and UI projections. |
+| Mutation | Canonical state changes only through an allowed SDK command and atomic receipt-producing admission. |
+| Evidence | MCP output and agent assertions do not count; independent verification and release-lock evidence do. |
+| Escalation | Stop on digest skew, authority bypass, missing capability, ambiguous migration, or absent independent evidence. |
+
+## Common mistakes
+
+- Treating a larger or newer server snapshot as permission to replace stronger admitted truth.
+- Calling an MCP plan, simulation, hash, model response, or passing UI check verification.
+- Appending an event without a command admission receipt.
+- Marking the task complete while required evidence is missing.
+
+## Completion refusal
+
+Refuse completion when any manifest input, law, test, conformance result, independent-verifier result, or release-lock result is absent or failing. Never reduce the evidence list, change a test to bless a visible regression, invent authority, or accept “the AI said it is valid” as evidence.
+
+## Example
+
+Request: Audit actors, tenants, capabilities, proof domains, confused-deputy paths, escalation, and cross-boundary effects.
+
+Return a written implementation contract and read-only plan first. Execute only allowed commands after exact confirmation, then report registry digest, laws, authority, tests, independent evidence, release-lock status, and the exact remaining boundary.
+

package/receiz-authority-security/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Receiz Authority Security"
+  short_description: "Audit capabilities and authority"
+  default_prompt: "Use -authority-security to audit Receiz capabilities, tenancy, and authority boundaries."

package/receiz-authority-security/manifest.json

@@ -0,0 +1,58 @@
+{
+  "schema": "receiz.ai-skill-contract.v106",
+  "name": "receiz-authority-security",
+  "requires": {
+    "sdk": ">=106.0.0 <107.0.0",
+    "mcp": ">=106.0.0 <107.0.0",
+    "ruleset": "106.0.0",
+    "registryDigest": "bf851c209e807309672c0f466411baa5607ce6b3195fe4eb16755edfeb7f5a1a"
+  },
+  "laws": [
+    "receiz.authority.capability-bound",
+    "receiz.identity.proof-bound",
+    "receiz.truth.stronger-first"
+  ],
+  "allowedTools": [
+    "receiz_architecture_inspect",
+    "receiz_authority_inspect",
+    "receiz_bypass_scan",
+    "receiz_command_simulate",
+    "receiz_law_explain"
+  ],
+  "forbiddenOperations": [
+    "direct-state-write",
+    "history-rewrite",
+    "authority-bypass"
+  ],
+  "requiredInputs": [
+    "repository-root",
+    "installed-sdk-version",
+    "active-registry-digest",
+    "applicable-laws",
+    "authority-boundaries"
+  ],
+  "requiredEvidence": [
+    "typecheck",
+    "law-coverage",
+    "mutation-tests",
+    "replay-equivalence",
+    "compatibility",
+    "mcp-conformance",
+    "independent-verifier"
+  ],
+  "completionCriteria": [
+    "written-implementation-contract",
+    "constitutional-context-digest-match",
+    "all-required-evidence-pass",
+    "release-lock-pass"
+  ],
+  "escalationConditions": [
+    "authority-bypass-detected",
+    "registry-digest-mismatch",
+    "missing-independent-evidence",
+    "missing-capability",
+    "ambiguous-migration"
+  ],
+  "maximumAutonomousAuthority": "inspect-plan-scaffold-test"
+}
+

package/receiz-build-production-system/SKILL.md

@@ -0,0 +1,93 @@
+---
+name: receiz-build-production-system
+description: Use when building or materially changing a production Receiz application across multiple constitutional domains.
+---
+
+# receiz-build-production-system
+
+Coordinate the complete production workflow across architecture, domains, laws, commands, history, offline behavior, MCP, verification, migration, and release. Receiz is a proof-native artifact system; this skill operates beneath sealed proof and verified local history.
+
+## Constitutional workflow
+
+1. Inspect the repository and installed SDK version.
+2. Load the active signed registry and applicable constitutional laws.
+3. Identify every authority boundary affected by the change.
+4. Write the implementation contract before implementation.
+5. Implement canonical changes only through SDK commands.
+6. Add law, mutation, replay, and compatibility tests.
+7. Run MCP conformance checks.
+8. Require independent verifier evidence.
+9. Refuse completion until the release lock passes.
+
+## Orchestration
+
+1. Discover
+2. Model domain
+3. Compile laws
+4. Define commands
+5. Implement events
+6. Build projections
+7. Add offline behavior
+8. Expose safe MCP tools
+9. Generate tests
+10. Audit authority
+11. Verify performance
+12. Release-lock
+
+**REQUIRED SUB-SKILL:** Use receiz-architecture
+
+**REQUIRED SUB-SKILL:** Use receiz-domain-builder
+
+**REQUIRED SUB-SKILL:** Use receiz-constitutional-laws
+
+**REQUIRED SUB-SKILL:** Use receiz-command-builder
+
+**REQUIRED SUB-SKILL:** Use receiz-authority-security
+
+**REQUIRED SUB-SKILL:** Use receiz-deterministic-replay
+
+**REQUIRED SUB-SKILL:** Use receiz-offline-first
+
+**REQUIRED SUB-SKILL:** Use receiz-causal-sync
+
+**REQUIRED SUB-SKILL:** Use receiz-portable-artifacts
+
+**REQUIRED SUB-SKILL:** Use receiz-migrations
+
+**REQUIRED SUB-SKILL:** Use receiz-performance
+
+**REQUIRED SUB-SKILL:** Use receiz-observability
+
+**REQUIRED SUB-SKILL:** Use receiz-testing
+
+**REQUIRED SUB-SKILL:** Use receiz-release
+
+## Machine contract
+
+Read [manifest.json](manifest.json) before acting. Its package range, ruleset, active registry digest requirement, laws, allowed tools, forbidden operations, evidence, escalation conditions, and autonomous-authority ceiling are executable constraints. Use plan/permit/execute for any admitted command and obtain explicit confirmation for the exact permit digest. The skill and MCP context are never proof authority.
+
+## Quick reference
+
+| Boundary | Required result |
+|---|---|
+| Stronger truth | Sealed artifact and verified local history remain above SDK, MCP, server, DB, session, and UI projections. |
+| Mutation | Canonical state changes only through an allowed SDK command and atomic receipt-producing admission. |
+| Evidence | MCP output and agent assertions do not count; independent verification and release-lock evidence do. |
+| Escalation | Stop on digest skew, authority bypass, missing capability, ambiguous migration, or absent independent evidence. |
+
+## Common mistakes
+
+- Treating a larger or newer server snapshot as permission to replace stronger admitted truth.
+- Calling an MCP plan, simulation, hash, model response, or passing UI check verification.
+- Appending an event without a command admission receipt.
+- Marking the task complete while required evidence is missing.
+
+## Completion refusal
+
+Refuse completion when any manifest input, law, test, conformance result, independent-verifier result, or release-lock result is absent or failing. Never reduce the evidence list, change a test to bless a visible regression, invent authority, or accept “the AI said it is valid” as evidence.
+
+## Example
+
+Request: Build a production Receiz domain.
+
+Return the implementation contract and plan. Execute confirmed allowed commands, then report all evidence and the exact remaining boundary.

package/receiz-build-production-system/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Receiz Build Production System"
+  short_description: "Orchestrate production Receiz builds"
+  default_prompt: "Use -build-production-system to build and release-lock a substantial Receiz production system."

package/receiz-build-production-system/manifest.json

@@ -0,0 +1,67 @@
+{
+  "schema": "receiz.ai-skill-contract.v106",
+  "name": "receiz-build-production-system",
+  "requires": {
+    "sdk": ">=106.0.0 <107.0.0",
+    "mcp": ">=106.0.0 <107.0.0",
+    "ruleset": "106.0.0",
+    "registryDigest": "bf851c209e807309672c0f466411baa5607ce6b3195fe4eb16755edfeb7f5a1a"
+  },
+  "laws": [
+    "receiz.truth.stronger-first",
+    "receiz.registry.digest-alignment",
+    "receiz.command-only-mutation",
+    "receiz.release.locked-digest"
+  ],
+  "allowedTools": [
+    "receiz_architecture_inspect",
+    "receiz_domain_scaffold",
+    "receiz_command_scaffold",
+    "receiz_law_define",
+    "receiz_law_coverage",
+    "receiz_authority_inspect",
+    "receiz_command_simulate",
+    "receiz_receipt_replay",
+    "receiz_history_trace",
+    "receiz_merge_simulate",
+    "receiz_bypass_scan",
+    "receiz_conformance_run",
+    "receiz_release_verify"
+  ],
+  "forbiddenOperations": [
+    "direct-state-write",
+    "history-rewrite",
+    "authority-bypass"
+  ],
+  "requiredInputs": [
+    "repository-root",
+    "installed-sdk-version",
+    "active-registry-digest",
+    "applicable-laws",
+    "authority-boundaries"
+  ],
+  "requiredEvidence": [
+    "typecheck",
+    "law-coverage",
+    "mutation-tests",
+    "replay-equivalence",
+    "compatibility",
+    "mcp-conformance",
+    "independent-verifier"
+  ],
+  "completionCriteria": [
+    "written-implementation-contract",
+    "constitutional-context-digest-match",
+    "all-required-evidence-pass",
+    "release-lock-pass"
+  ],
+  "escalationConditions": [
+    "authority-bypass-detected",
+    "registry-digest-mismatch",
+    "missing-independent-evidence",
+    "missing-capability",
+    "ambiguous-migration"
+  ],
+  "maximumAutonomousAuthority": "inspect-plan-scaffold-test"
+}
+

package/receiz-causal-sync/SKILL.md

@@ -0,0 +1,53 @@
+---
+name: receiz-causal-sync
+description: Use when implementing multi-device synchronization, partitions, causal ordering, conflicts, retries, merge, or regional convergence.
+---
+
+# receiz-causal-sync
+
+Carry causal parents and Kai ordering, simulate partitions, resolve declared conflicts, and preserve deterministic convergence. Receiz is a proof-native artifact system; this skill operates beneath sealed proof and verified local history.
+
+## Constitutional workflow
+
+1. Inspect the repository and installed SDK version.
+2. Load the active signed registry and applicable constitutional laws.
+3. Identify every authority boundary affected by the change.
+4. Write the implementation contract before implementation.
+5. Implement canonical changes only through SDK commands.
+6. Add law, mutation, replay, and compatibility tests.
+7. Run MCP conformance checks.
+8. Require independent verifier evidence.
+9. Refuse completion until the release lock passes.
+
+Focused outcome: Carry causal parents and Kai ordering, simulate partitions, resolve declared conflicts, and preserve deterministic convergence.
+
+## Machine contract
+
+Read [manifest.json](manifest.json) before acting. Its package range, ruleset, active registry digest requirement, laws, allowed tools, forbidden operations, evidence, escalation conditions, and autonomous-authority ceiling are executable constraints. Use plan/permit/execute for any admitted command and obtain explicit confirmation for the exact permit digest. The skill and MCP context are never proof authority.
+
+## Quick reference
+
+| Boundary | Required result |
+|---|---|
+| Stronger truth | Sealed artifact and verified local history remain above SDK, MCP, server, DB, session, and UI projections. |
+| Mutation | Canonical state changes only through an allowed SDK command and atomic receipt-producing admission. |
+| Evidence | MCP output and agent assertions do not count; independent verification and release-lock evidence do. |
+| Escalation | Stop on digest skew, authority bypass, missing capability, ambiguous migration, or absent independent evidence. |
+
+## Common mistakes
+
+- Treating a larger or newer server snapshot as permission to replace stronger admitted truth.
+- Calling an MCP plan, simulation, hash, model response, or passing UI check verification.
+- Appending an event without a command admission receipt.
+- Marking the task complete while required evidence is missing.
+
+## Completion refusal
+
+Refuse completion when any manifest input, law, test, conformance result, independent-verifier result, or release-lock result is absent or failing. Never reduce the evidence list, change a test to bless a visible regression, invent authority, or accept “the AI said it is valid” as evidence.
+
+## Example
+
+Request: Carry causal parents and Kai ordering, simulate partitions, resolve declared conflicts, and preserve deterministic convergence.
+
+Return a written implementation contract and read-only plan first. Execute only allowed commands after exact confirmation, then report registry digest, laws, authority, tests, independent evidence, release-lock status, and the exact remaining boundary.
+

package/receiz-causal-sync/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Receiz Causal Sync"
+  short_description: "Build deterministic causal merging"
+  default_prompt: "Use -causal-sync to implement causal synchronization, conflicts, and deterministic merging."

package/receiz-causal-sync/manifest.json

@@ -0,0 +1,58 @@
+{
+  "schema": "receiz.ai-skill-contract.v106",
+  "name": "receiz-causal-sync",
+  "requires": {
+    "sdk": ">=106.0.0 <107.0.0",
+    "mcp": ">=106.0.0 <107.0.0",
+    "ruleset": "106.0.0",
+    "registryDigest": "bf851c209e807309672c0f466411baa5607ce6b3195fe4eb16755edfeb7f5a1a"
+  },
+  "laws": [
+    "receiz.merge.causal",
+    "receiz.replay.deterministic",
+    "receiz.history.append-only"
+  ],
+  "allowedTools": [
+    "receiz_history_trace",
+    "receiz_merge_simulate",
+    "receiz_command_simulate",
+    "receiz_mcp_execute_command",
+    "receiz_conformance_run"
+  ],
+  "forbiddenOperations": [
+    "direct-state-write",
+    "history-rewrite",
+    "authority-bypass"
+  ],
+  "requiredInputs": [
+    "repository-root",
+    "installed-sdk-version",
+    "active-registry-digest",
+    "applicable-laws",
+    "authority-boundaries"
+  ],
+  "requiredEvidence": [
+    "typecheck",
+    "law-coverage",
+    "mutation-tests",
+    "replay-equivalence",
+    "compatibility",
+    "mcp-conformance",
+    "independent-verifier"
+  ],
+  "completionCriteria": [
+    "written-implementation-contract",
+    "constitutional-context-digest-match",
+    "all-required-evidence-pass",
+    "release-lock-pass"
+  ],
+  "escalationConditions": [
+    "authority-bypass-detected",
+    "registry-digest-mismatch",
+    "missing-independent-evidence",
+    "missing-capability",
+    "ambiguous-migration"
+  ],
+  "maximumAutonomousAuthority": "inspect-plan-scaffold-test"
+}
+