chati-dev 4.5.16 → 4.5.27

package/framework/tasks/orchestrator-resume.md

@@ -20,21 +20,18 @@ criteria:
 Restore session context from previous work, determine pipeline position, and prepare to continue.
 
 ## Prerequisites
-- Session file exists at `.chati/session.yaml`
+- Installed provider-native entrypoint and deterministic router
 - Handoff artifacts in `.chati/artifacts/handoffs/`
 - Constitution at `chati.dev/constitution.md`
 - Project configuration at `chati.dev/config.yaml`
 
 ## Steps
 
-### 1. Locate Session File
-Check for existing session at `.chati/session.yaml`:
-- If found: Load and validate
-- If not found: This is a new session, initialize instead
-- If corrupted: Attempt recovery from handoffs
+### 1. Activate Through the Router
+Run the provider-native entrypoint, which invokes `chati-router.js --harness <active-harness>`. Consume only the returned JSON. If `ok` is false, report the exact error and stop. Never locate, read or reconstruct `.chati/session.yaml` directly.
 
 ### 2. Load Session State
-Parse the YAML session file and extract:
+Use the validated `session` object returned by the router to extract:
 - Project metadata (name, type, language)
 - Current mode (planning/build/deploy)
 - Pipeline position (phase, last agent)
@@ -121,16 +118,15 @@ If last agent is complete but handoff is absent:
 ## Error Handling
 
 ### Session File Not Found
-If `.chati/session.yaml` doesn't exist:
+If the router reports that no session exists:
 - Check if this is initial installation
 - Look for backup in `.chati/backups/`
 - Offer to initialize new session
 
-### Session File Corrupted
-If YAML parsing fails:
-- Try JSON parse (in case of format change)
-- Load most recent backup
-- Escalate to user with corruption details
+### Session State Rejected
+If the router rejects session state:
+- Report the exact router error and relevant diagnostic detail
+- Stop without fallback parsing, backup substitution or reconstruction
 
 ### Handoff Chain Broken
 If handoff sequence has gaps:
@@ -155,7 +151,7 @@ If pipeline position is ambiguous:
 ```yaml
 session_state:
   loaded_at: "2026-02-13T10:30:00Z"
-  session_file: ".chati/session.yaml"
+  session_source: "router_validated_context"
   session_age_hours: 8
   project:
     name: "HealthCare Portal"

package/framework/tasks/orchestrator-route.md

@@ -20,7 +20,7 @@ criteria:
 Analyze user input, classify intent, determine pipeline position, and route to the appropriate agent.
 
 ## Prerequisites
-- Session state loaded from `.chati/session.yaml`
+- Validated session state returned by `chati-router.js status`
 - Constitution available at `chati.dev/constitution.md`
 - Agent definitions in `chati.dev/agents/`
 - User input captured
@@ -34,7 +34,7 @@ Extract the core intent from user message. Identify:
 - Context changes (e.g., "actually, let's change the database")
 
 ### 2. Load Session Context
-Read `.chati/session.yaml` to understand:
+Run `node chati.dev/orchestrator/chati-router.js status` and consume only its validated JSON. If `ok` is false, report the exact error and stop. Never read `.chati/session.yaml` directly. Use the returned context to understand:
 - Current mode (planning/build/deploy)
 - Last completed agent
 - Pipeline position
@@ -91,7 +91,7 @@ Dispatch to the selected destination:
 - For deviations: Trigger deviation protocol
 
 ### 10. Log Routing Action
-Update session state with routing information:
+Execute the router-returned action so the CLI session manager records:
 - Timestamp
 - Source (user input)
 - Target (agent/action)

package/framework/tasks/orchestrator-spawn-terminal.md

@@ -23,7 +23,7 @@ Coordinate parallel execution of independent tasks across multiple terminal inst
 ## Prerequisites
 - Multi-terminal IDE support (Claude Code, VS Code, Cursor, Windsurf)
 - Task definitions with `parallelizable: true` flag
-- Session state at `.chati/session.yaml`
+- Router-validated canonical session context
 - Agent definitions in `chati.dev/agents/`
 
 ## Steps

package/framework/tasks/orchestrator-status.md

@@ -21,15 +21,15 @@ criteria:
 Generate a comprehensive status report of the current project, showing pipeline progress, agent completion, and next steps.
 
 ## Prerequisites
-- Session file at `.chati/session.yaml`
+- Installed `chati-router.js` with the `status` command
 - Agent definitions in `chati.dev/agents/`
 - Constitution at `chati.dev/constitution.md`
 - Artifacts in `.chati/artifacts/`
 
 ## Steps
 
-### 1. Load Session State
-Read `.chati/session.yaml` to get current project state:
+### 1. Load Validated Session State
+Run `node chati.dev/orchestrator/chati-router.js status` and consume only its validated JSON. If `ok` is false, report the exact error and stop. Never read `.chati/session.yaml` directly. Use the returned fields for:
 - Project metadata (name, type, language)
 - Current mode and phase
 - Agent statuses with timestamps
@@ -124,7 +124,7 @@ If no activity in last 7+ days:
 ## Error Handling
 
 ### Session File Missing
-If `.chati/session.yaml` doesn't exist:
+If the router reports that no session exists:
 - Report "No active session"
 - Suggest running `/chati` to initialize
 - Check for backup session files
@@ -141,11 +141,11 @@ If handoff or artifact files can't be read:
 - Continue with available data
 - Log warning for investigation
 
-### Corrupted Session Data
-If session YAML is malformed:
-- Attempt partial parsing
-- Report known good data
-- Flag corruption clearly
+### Unsafe or Corrupted Session Data
+If the router rejects session state:
+- Report its exact error and diagnostic detail
+- Run the router `doctor` command only when instructed by the orchestration contract
+- Stop without partial parsing, reconstruction or direct file access
 
 ## Output Format
 

package/framework/tasks/orchestrator-suggest-mode.md

@@ -21,7 +21,7 @@ criteria:
 Analyze session state to determine if mode transition conditions are met, calculate confidence, and either suggest to user or autonomously execute the transition.
 
 ## Prerequisites
-- Session state at `.chati/session.yaml`
+- Router-validated canonical session context
 - Constitution Article XI at `chati.dev/constitution.md`
 - Agent completion data
 - Quality scores from QA agents

package/framework/tasks/qa-impl-consolidate.md

@@ -95,7 +95,7 @@ Compile the final QA implementation report synthesizing all testing, analysis, a
     - Include quick-reference guide to QA reports
 
 12. **Execute Handoff**
-    - Update session.yaml with QA implementation completion
+    - Return QA implementation completion in the handoff; the orchestrator persists it through the session manager
     - Log handoff decision (to devops, dev, or orchestrator)
     - If autonomous gate (PASS verdict), trigger next agent automatically
     - If manual approval needed, prompt orchestrator
@@ -133,7 +133,7 @@ Compile the final QA implementation report synthesizing all testing, analysis, a
 - If autonomous gate should trigger but handoff fails, log error
 - Do not proceed automatically
 - Escalate to orchestrator for manual handoff
-- Document failure in session.yaml
+- Return the failure in the structured handoff so the orchestrator records it canonically
 
 ## Output Format
 

package/framework/tasks/qa-impl-performance-test.md

@@ -37,7 +37,7 @@ Execute performance benchmarks to measure CLI startup time, agent execution time
 2. **Define Performance Metrics**
    - **CLI Startup Time**: Time from `npx chati-dev` invocation to first output
    - **Agent Execution Time**: Time per agent task (avg, p50, p95, p99)
-   - **File Operations**: Read/write time for session.yaml, config.yaml
+   - **File Operations**: Canonical session-manager read/write time and config.yaml access
    - **YAML Parsing**: Time to parse task definitions, config files
    - **Memory Usage**: Peak memory during CLI execution, agent workflows
    - **State Operations**: Time for state read, write, validation
@@ -62,8 +62,8 @@ Execute performance benchmarks to measure CLI startup time, agent execution time
    - Calculate agent execution time distribution (avg, p50, p95, p99)
 
 6. **Execute File Operation Benchmarks**
-   - Benchmark session.yaml read (cold and warm cache)
-   - Benchmark session.yaml write (atomic write with backup)
+   - Benchmark the canonical session reader (cold and warm cache)
+   - Benchmark session-manager atomic persistence and backup behavior
    - Benchmark config.yaml read and merge
    - Benchmark task definition loading (all .md files in chati.dev/tasks/)
    - Measure filesystem sync overhead
@@ -374,7 +374,7 @@ threshold_compliance:
 
   file_operations:
     threshold: 50ms
-    max_current: 35ms (session.yaml write)
+    max_current: 35ms (canonical session-manager write)
     status: PASS
     margin: 15ms (30% under threshold)
 

package/framework/tasks/qa-impl-regression-check.md

@@ -50,7 +50,7 @@ Execute regression tests to ensure new changes have not broken existing function
    - **Test status changes**: Tests that were passing now fail, or vice versa
    - **Output differences**: stdout/stderr changes, different error messages
    - **Behavior changes**: Different execution paths, timing differences
-   - **File artifact changes**: session.yaml structure, config format, generated files
+   - **File artifact changes**: canonical session schema, config format, generated files
 
 5. **Categorize Differences**
    - **True regressions**: Unintended breakage of existing functionality
@@ -208,7 +208,7 @@ regressions:
       severity: HIGH
       type: functional
       feature: Session state persistence
-      description: "session.yaml no longer includes mode_transitions array"
+      description: "canonical session schema no longer includes mode_transitions array"
       baseline_behavior: "mode_transitions[] populated with audit trail"
       current_behavior: "mode_transitions[] empty or missing"
       root_cause: "State schema update removed array initialization"
@@ -258,7 +258,7 @@ intentional_changes:
 
   - id: CHANGE-003
     feature: State schema version
-    description: "session.yaml schema version updated to 2.0"
+    description: "canonical session schema version advanced to 2.0"
     baseline: "version: 1.0"
     current: "version: 2.0"
     justification: "Schema breaking change for mode governance"
@@ -336,7 +336,7 @@ smoke_tests:
     command: npx chati-dev init
     status: PASS
     execution_time: 8.2s
-    notes: "Completed successfully, session.yaml created"
+    notes: "Completed successfully, canonical session state created by the session manager"
 
   agent_handoff:
     workflow: "greenfield-wu -> brief -> detail"

package/framework/tasks/qa-impl-sast-scan.md

@@ -104,7 +104,7 @@ Run static analysis security testing (SAST) to identify security vulnerabilities
     - List all CRITICAL and HIGH findings with details
     - Provide overall gate status (PASS, FAIL, CONDITIONAL)
     - Include remediation roadmap
-    - Update session.yaml with scan results
+    - Return scan results in the handoff; the orchestrator persists them through the session manager
 
 ## Decision Points
 

package/framework/tasks/qa-impl-test-execute.md

@@ -105,7 +105,7 @@ Execute the complete test suite (unit and integration tests) for the implemented
 
 12. **Log Results and Next Steps**
     - Save test-results.yaml to session
-    - Update session.yaml with test execution status
+    - Return test execution status in the handoff; the orchestrator persists it through the session manager
     - If all tests pass and coverage meets targets: proceed to SAST scan
     - If failures exist: flag for dev agent remediation, then re-run
 

package/framework/tasks/qa-impl-verdict.md

@@ -108,7 +108,7 @@ Issue the final QA verdict by synthesizing all QA implementation results (tests,
 
 12. **Log Verdict and Update Session**
     - Save qa-verdict.yaml with complete verdict
-    - Update session.yaml with QA implementation status
+    - Return QA implementation status in the handoff; the orchestrator persists it through the session manager
     - If PASS (autonomous gate), trigger handoff to qa-impl-consolidate
     - If FAIL, flag for orchestrator to return to dev
     - If CONDITIONAL, prompt orchestrator for user decision
@@ -145,7 +145,7 @@ Issue the final QA verdict by synthesizing all QA implementation results (tests,
 - If autonomous gate should trigger but handoff fails, log error
 - Do not proceed to next agent automatically
 - Escalate to orchestrator for manual handoff
-- Document failure in session.yaml
+- Return the failure in the structured handoff so the orchestrator records it canonically
 
 ## Output Format
 

package/framework/tasks/qa-planning-consolidate.md

@@ -101,7 +101,7 @@ Compile all QA planning outputs into a final qa-plan.yaml document, calculate an
     - Document any open questions or assumptions
 
 12. **Log Completion and Next Steps**
-    - Update session.yaml with qa-planning completion status
+    - Return qa-planning completion status in the handoff; the orchestrator persists it through the session manager
     - Log overall score and autonomous transition decision
     - If autonomous transition approved, trigger dev agent
     - If manual approval needed, prompt orchestrator for user input
@@ -137,7 +137,7 @@ Compile all QA planning outputs into a final qa-plan.yaml document, calculate an
 - If overall score < 90%, compilation succeeds but transition fails
 - Generate detailed gap analysis with prioritized action items
 - Escalate to orchestrator for user decision (remediate or override via deviation protocol)
-- Log decision and rationale in session.yaml
+- Return the decision and rationale in the structured handoff so the orchestrator records it through the session manager
 
 ## Output Format
 
@@ -234,7 +234,7 @@ executive_summary:
 handoff_to_dev:
   priority_1_tests:
     - Mode transition validation (state management, orchestrator)
-    - session.yaml write operations (state management)
+    - canonical session-manager operations (state management)
     - Version migration execution (upgrade system)
 
   priority_2_tests:

package/framework/tasks/qa-planning-coverage-plan.md

@@ -31,7 +31,7 @@ Create a detailed test coverage plan with specific targets for each module, comp
    - Identify all modules: orchestrator, agents (13), state management, file operations, CLI, validators, parsers
    - List core utilities: YAML handlers, i18n, error handling, logging
    - Note external interfaces: MCP tools, IDE integration, package managers
-   - Document data structures: session.yaml schema, config.yaml, task definitions
+   - Catalog data structures: canonical session schema, config.yaml, task definitions
 
 2. **Categorize Components by Testability**
    - **High Testability**: Pure functions (validators, parsers, formatters, calculators)
@@ -47,7 +47,7 @@ Create a detailed test coverage plan with specific targets for each module, comp
 
 4. **Set Module-Specific Targets**
    - **Orchestrator**: 85% (critical routing and mode governance logic)
-   - **State Management**: 90% (session.yaml read/write, consistency checks)
+   - **State Management**: 90% (canonical session-manager reads/writes and consistency checks)
    - **Agent Task Execution**: 75% (task loading, validation, execution)
    - **File Operations**: 80% (read/write/create, permission handling, error recovery)
    - **Validators**: 90% (schema validation, input sanitization)

package/framework/tasks/qa-planning-gate-define.md

@@ -49,7 +49,7 @@ Define quality gates and enforcement thresholds for each stage of the chati.dev
    - **Enforcement**: Orchestrator validation before mode transition
 
 4. **Define Build Phase Gates**
-   - **Implementation Complete**: All tasks marked done in session.yaml
+   - **Implementation Complete**: All tasks reported complete in the router-validated task state
    - **Test Execution**: All unit + integration tests passing
    - **SAST Clean**: Zero critical vulnerabilities, high vulnerabilities < 3
    - **No Regressions**: All regression checks passing vs baseline
@@ -80,7 +80,7 @@ Define quality gates and enforcement thresholds for each stage of the chati.dev
    - **Automated**: CI/CD pipeline checks, pre-commit hooks
    - **Agent-Enforced**: Orchestrator mode governance, qa-planning consolidation score
    - **Manual**: Human review for override decisions, complex risk assessment
-   - **Audit Trail**: All gate passes/failures logged in session.yaml mode_transitions
+   - **Audit Trail**: All gate passes/failures returned in the handoff for canonical session-manager persistence under `mode_transitions`
 
 9. **Create Gate Bypass Procedures**
    - **Deviation Protocol**: Orchestrator-managed with user confirmation
@@ -183,7 +183,7 @@ pipeline_stages:
     gates:
       - name: Implementation Complete
         threshold: All tasks done
-        enforcement: session.yaml validation
+        enforcement: canonical session-manager validation
         bypass: deviation_protocol
 
       - name: Test Suite
@@ -267,7 +267,7 @@ bypass_procedures:
     trigger: orchestrator
     requires: user confirmation
     documentation: [reason, risk assessment, mitigation plan]
-    audit: logged in session.yaml mode_transitions
+    audit: returned in the handoff for canonical session-manager persistence under mode_transitions
     expiration: 24 hours
 
   manual_approval:

package/framework/tasks/qa-planning-risk-matrix.md

@@ -30,7 +30,7 @@ Build a comprehensive risk matrix mapping all features, components, and workflow
 1. **Enumerate All System Features**
    - Core workflows: greenfield pipeline, brownfield pipeline, mode transitions
    - Agent operations: task execution, handoffs, validation
-   - State management: session.yaml CRUD, consistency checks, concurrency
+   - State management: canonical session-manager CRUD, consistency checks, concurrency
    - File operations: create, read, write, delete, permissions
    - CLI commands: /chati, npx chati-dev *, help, status
    - Configuration: config.yaml loading, merging, validation
@@ -156,7 +156,7 @@ features:
     rationale: Invalid mode transitions can corrupt workflow state
     mitigation:
       - Strict validation in mode-governance.js
-      - Audit trail in session.yaml mode_transitions
+      - Audit trail persisted by the canonical session manager under `mode_transitions`
       - Rollback mechanism for failed transitions
     testing_priority: 1
     coverage_target: 95%
@@ -194,7 +194,7 @@ features:
     gate_enforcement: critical
 
   # STATE MANAGEMENT
-  - feature: session.yaml write operations
+  - feature: canonical session-manager write operations
     area: state_management
     impact: critical
     likelihood: high
@@ -534,7 +534,7 @@ testing_priorities:
     gate_enforcement: blocker
     features:
       - Mode transition validation
-      - session.yaml write operations
+      - canonical session-manager write operations
       - Concurrent state access
       - File write with permissions
       - Version migration execution

package/framework/tasks/qa-planning-test-strategy.md

@@ -47,7 +47,7 @@ Define a comprehensive testing strategy covering unit, integration, end-to-end,
    - **Manual Tests**: IDE compatibility, internationalization, accessibility
 
 4. **Identify Risk Areas**
-   - State management corruption (session.yaml consistency)
+   - State management corruption (canonical session consistency through the session manager)
    - File system operations (permissions, race conditions)
    - Agent handoff logic (transition validation)
    - Mode governance enforcement (unauthorized transitions)
@@ -169,7 +169,7 @@ testing_pyramid:
 
 risk_areas:
   critical:
-    - State corruption in session.yaml
+    - Canonical session corruption reported by the router
     - Mode governance bypass
     - File system race conditions
   high:

package/node_modules/@chati/browser-capability/src/index.js

@@ -5,6 +5,7 @@ const TOOLS = new Set(['playwright-cli', 'playwright-mcp', 'chrome-devtools-mcp'
 const CLASSIFICATIONS = new Set(['internal', 'confidential', 'client-restricted', 'personal-restricted']);
 const ACTION_CONFIRMATIONS = new Set(['required', 'not-required-by-policy']);
 const ARTIFACT_KINDS = new Set(['screenshot', 'dom', 'network', 'trace', 'download', 'console']);
+const EVIDENCE_CLASSES = new Set(['delegation-only', 'browser-simulation', 'real-browser']);
 
 function fail(code, message, details = {}) { throw new ContractError(code, message, details); }
 function object(value, code, label) {
@@ -70,7 +71,7 @@ export function assertBrowserCapabilityForTask(task = {}) {
   return policy;
 }
 
-function validateLaunchRequest(input, policy) {
+export function validateBrowserLaunchRequest(input, policy) {
   object(input, 'INVALID_BROWSER_REQUEST', 'browser launch request');
   string(input.browser_session_id, 'INVALID_BROWSER_REQUEST', 'browser_session_id');
   string(input.task_id, 'INVALID_BROWSER_REQUEST', 'task_id');
@@ -117,7 +118,7 @@ export class BrowserCapability {
 
   preflight(input = {}) {
     const policy = validateBrowserPolicy(input.policy);
-    const request = validateLaunchRequest(input, policy);
+    const request = validateBrowserLaunchRequest(input, policy);
     const manifestInput = {
       schema_version: 1,
       manifest_type: 'BrowserEvidenceManifest',
@@ -129,6 +130,7 @@ export class BrowserCapability {
       allowed_domain: request.hostname,
       profile_isolation: policy.profile_isolation,
       classification: request.classification,
+      evidence_class: 'browser-simulation',
       artifact_policy: policy.artifact_policy,
       action_confirmation: policy.action_confirmation,
       artifacts: [],
@@ -163,3 +165,11 @@ export class BrowserCapability {
     return this.#issue(output);
   }
 }
+
+export function validateBrowserEvidenceClass(value) {
+  string(value, 'INVALID_BROWSER_EVIDENCE_CLASS', 'browser evidence_class');
+  if (!EVIDENCE_CLASSES.has(value)) {
+    fail('INVALID_BROWSER_EVIDENCE_CLASS', 'browser evidence_class must distinguish delegation, simulation, or real browser execution');
+  }
+  return value;
+}

package/node_modules/@chati/planning/src/index.js

@@ -9,6 +9,16 @@ function fail(code, message, details = {}) {
   throw new ContractError(code, message, details);
 }
 
+function trustedGitEnvironment(source = process.env) {
+  const environment = Object.fromEntries(Object.entries(source).filter(([key]) => !key.startsWith('GIT_')));
+  return {
+    ...environment,
+    GIT_CONFIG_NOSYSTEM: '1',
+    GIT_CONFIG_GLOBAL: process.platform === 'win32' ? 'NUL' : '/dev/null',
+    GIT_TERMINAL_PROMPT: '0',
+  };
+}
+
 function assertObject(value, code, label) {
   if (!value || typeof value !== 'object' || Array.isArray(value)) fail(code, `${label} must be an object`);
 }
@@ -47,6 +57,9 @@ function validateTask(task, knownTaskIds) {
   assertStringArray(task.acceptance_criteria_refs, 'MISSING_ACCEPTANCE_CRITERIA', 'task.acceptance_criteria_refs', { allowEmpty: false });
   assertImmutableRef(task.execution_scope_ref, 'INVALID_HANDOFF_TASK', 'task.execution_scope_ref');
   assertImmutableRef(task.routing_constraints_ref, 'INVALID_HANDOFF_TASK', 'task.routing_constraints_ref');
+  if (task.required_independence_level !== undefined && !['A', 'B', 'C'].includes(task.required_independence_level)) {
+    fail('INVALID_HANDOFF_TASK', 'task.required_independence_level must be A, B or C');
+  }
   assertObject(task.execution_binding, 'TASK_EXECUTION_BINDING_REQUIRED', 'task.execution_binding');
   for (const field of ['provider_id', 'harness_id', 'model_id', 'reasoning_configuration']) {
     assertString(task.execution_binding[field], 'TASK_EXECUTION_BINDING_REQUIRED', `task.execution_binding.${field}`);
@@ -91,7 +104,8 @@ export function buildCanonicalManifest(handoff) {
   assertObject(handoff, 'INVALID_HANDOFF', 'handoff');
   const { seal, ...rest } = handoff;
   assertObject(seal, 'INVALID_HANDOFF', 'seal');
-  const { manifest_digest, ...sealWithoutDigest } = seal;
+  const sealWithoutDigest = { ...seal };
+  delete sealWithoutDigest.manifest_digest;
   return canonicalize({ ...rest, seal: sealWithoutDigest });
 }
 
@@ -152,7 +166,9 @@ export function verifyLocalGitCommit({ git_dir, git_commit_ref }) {
   assertString(git_commit_ref, 'INVALID_GIT_COMMIT_REF', 'git_commit_ref');
   if (!GIT_OBJECT_ID.test(git_commit_ref)) fail('INVALID_GIT_COMMIT_REF', 'git_commit_ref must be an immutable Git object id');
   try {
-    execFileSync('git', ['-C', git_dir, 'cat-file', '-e', `${git_commit_ref}^{commit}`], { stdio: 'ignore' });
+    execFileSync('git', ['-C', git_dir, 'cat-file', '-e', `${git_commit_ref}^{commit}`], {
+      stdio: 'ignore', env: trustedGitEnvironment(),
+    });
   } catch {
     fail('GIT_COMMIT_NOT_FOUND', 'sealed git_commit_ref is unavailable in the supplied local repository', { git_commit_ref });
   }
@@ -181,6 +197,7 @@ export function assertDispatchable({ handoff, task_id, completed_task_ids = [],
     manifest_digest: validated.seal.manifest_digest,
     git_commit_ref: validated.seal.git_commit_ref,
     task_id: task.task_id,
+    execution_binding: task.execution_binding,
   }));
 }
 
@@ -196,16 +213,19 @@ export function validateAttemptBinding({ attempt, handoff }) {
   for (const field of ['attempt_id', 'handoff_id', 'handoff_version', 'manifest_digest', 'git_commit_ref', 'task_id']) {
     assertString(attempt[field], 'INVALID_ATTEMPT_BINDING', field);
   }
+  assertObject(attempt.execution_binding, 'INVALID_ATTEMPT_BINDING', 'execution_binding');
   if (attempt.handoff_id !== validated.handoff_id || attempt.handoff_version !== validated.version || attempt.manifest_digest !== validated.seal.manifest_digest || attempt.git_commit_ref !== validated.seal.git_commit_ref) {
     fail('ATTEMPT_HANDOFF_DRIFT', 'attempt is not bound to this exact sealed handoff', { attempt_id: attempt.attempt_id });
   }
-  if (!validated.tasks.some((task) => task.task_id === attempt.task_id)) fail('ATTEMPT_TASK_DRIFT', 'attempt task is not part of bound handoff');
+  const task = validated.tasks.find((candidate) => candidate.task_id === attempt.task_id);
+  if (!task) fail('ATTEMPT_TASK_DRIFT', 'attempt task is not part of bound handoff');
+  if (sha256(attempt.execution_binding) !== sha256(task.execution_binding)) fail('ATTEMPT_EXECUTION_BINDING_DRIFT', 'attempt execution binding differs from the sealed task');
   return deepFreeze(canonicalize(attempt));
 }
 
 function legacyValue(value, label) {
   if (typeof value === 'string' && value.trim() !== '') return value;
-  fail('INVALID_LEGACY_PLANNING_ARTIFACT', `${label} must provide an immutable reference`);
+  return fail('INVALID_LEGACY_PLANNING_ARTIFACT', `${label} must provide an immutable reference`);
 }
 
 function legacyCriteriaRefs(task, taskId) {

package/node_modules/@chati/provider-registry/src/index.js

@@ -98,6 +98,17 @@ export function validateInstallation(installation, snapshot, { bindingMap = DEFA
         throw new ContractError('INVALID_INSTALLATION', 'allowed_reasoning_configurations must be an array of non-empty strings');
       }
     }
+    const hasExecutablePath = binding.executable_path !== undefined;
+    const hasExecutableDigest = binding.executable_sha256 !== undefined;
+    if (hasExecutablePath !== hasExecutableDigest) {
+      throw new ContractError('INVALID_INSTALLATION', 'provider executable path and digest must be selected together');
+    }
+    if (hasExecutablePath) {
+      assertString(binding.executable_path, 'INVALID_INSTALLATION', 'executable_path');
+      if (!binding.executable_path.startsWith('/') || !/^[a-f0-9]{64}$/.test(binding.executable_sha256 || '')) {
+        throw new ContractError('INVALID_INSTALLATION', 'provider executable binding must contain an absolute path and SHA-256 digest');
+      }
+    }
     selectedHarnesses.add(binding.harness_id);
     selectedBindings.set(`${binding.provider_id}:${binding.harness_id}`, binding);
   }