diff --git a/cycode/cli/apps/ai_guardrails/ides/base.py b/cycode/cli/apps/ai_guardrails/ides/base.py
index ab5958b4..ace69ec2 100644
--- a/cycode/cli/apps/ai_guardrails/ides/base.py
+++ b/cycode/cli/apps/ai_guardrails/ides/base.py
@@ -47,6 +47,16 @@ class HookDecision:
     user_message: str | None = None
     agent_message: str | None = None
 
+    @property
+    def model_message(self) -> str:
+        """What a tool deny tells the model: the findings the user sees, then the agent's instructions."""
+        return '\n'.join(message for message in (self.user_message, self.agent_message) if message)
+
+    @property
+    def permission_reason(self) -> str:
+        """A PreToolUse permissionDecisionReason: a deny reason goes to the model, an ask reason to the user."""
+        return self.model_message if self.action == DecisionAction.DENY else self.user_message or ''
+
     @classmethod
     def allow(cls, event_type: AiHookEventType) -> 'HookDecision':
         return cls(action=DecisionAction.ALLOW, event_type=event_type)
diff --git a/cycode/cli/apps/ai_guardrails/ides/claude_code.py b/cycode/cli/apps/ai_guardrails/ides/claude_code.py
index c70c363f..ca126d84 100644
--- a/cycode/cli/apps/ai_guardrails/ides/claude_code.py
+++ b/cycode/cli/apps/ai_guardrails/ides/claude_code.py
@@ -397,8 +397,13 @@ def build_hook_response(self, decision: HookDecision) -> dict:
         if decision.event_type == AiHookEventType.PROMPT:
             if decision.action == DecisionAction.ALLOW:
                 return {}
-            # Both DENY and (unexpected) ASK on prompts collapse to a block.
-            return {'decision': 'block', 'reason': decision.user_message or ''}
+            # Both DENY and (unexpected) ASK on prompts collapse to a block. Claude Code echoes the blocked
+            # prompt (secret included) under the reason unless suppressed; the reason already locates the finding.
+            return {
+                'decision': 'block',
+                'reason': decision.user_message or '',
+                'hookSpecificOutput': {'hookEventName': 'UserPromptSubmit', 'suppressOriginalPrompt': True},
+            }
 
         # FILE_READ / MCP_EXECUTION → hookSpecificOutput shape.
         if decision.action == DecisionAction.ALLOW:
@@ -412,7 +417,7 @@ def build_hook_response(self, decision: HookDecision) -> dict:
             'hookSpecificOutput': {
                 'hookEventName': 'PreToolUse',
                 'permissionDecision': decision.action.value,  # 'deny' or 'ask'
-                'permissionDecisionReason': decision.user_message or '',
+                'permissionDecisionReason': decision.permission_reason,
             }
         }
 
diff --git a/cycode/cli/apps/ai_guardrails/ides/codex.py b/cycode/cli/apps/ai_guardrails/ides/codex.py
index a41e6f12..cb8e6deb 100644
--- a/cycode/cli/apps/ai_guardrails/ides/codex.py
+++ b/cycode/cli/apps/ai_guardrails/ides/codex.py
@@ -319,7 +319,7 @@ def build_hook_response(self, decision: HookDecision) -> dict:
             'hookSpecificOutput': {
                 'hookEventName': 'PreToolUse',
                 'permissionDecision': decision.action.value,  # 'deny' or 'ask'
-                'permissionDecisionReason': decision.user_message or '',
+                'permissionDecisionReason': decision.permission_reason,
             }
         }
 
diff --git a/cycode/cli/apps/ai_guardrails/ides/copilot.py b/cycode/cli/apps/ai_guardrails/ides/copilot.py
index 90b71028..01a1738c 100644
--- a/cycode/cli/apps/ai_guardrails/ides/copilot.py
+++ b/cycode/cli/apps/ai_guardrails/ides/copilot.py
@@ -474,7 +474,8 @@ def build_hook_response(self, decision: HookDecision) -> dict:
             'hookSpecificOutput': {
                 'hookEventName': 'PreToolUse',
                 'permissionDecision': decision.action.value,  # 'deny' or 'ask'
-                'permissionDecisionReason': decision.user_message or '',
+                # Copilot VS Code shows a deny reason to the user rather than the model, which is harmless
+                'permissionDecisionReason': decision.permission_reason,
             }
         }
 
diff --git a/cycode/cli/apps/ai_guardrails/ides/cursor.py b/cycode/cli/apps/ai_guardrails/ides/cursor.py
index 4fbf070e..0dab8c5e 100644
--- a/cycode/cli/apps/ai_guardrails/ides/cursor.py
+++ b/cycode/cli/apps/ai_guardrails/ides/cursor.py
@@ -149,10 +149,13 @@ def build_hook_response(self, decision: HookDecision) -> dict:
         # FILE_READ / MCP_EXECUTION → permission shape
         if decision.action == DecisionAction.ALLOW:
             return {'permission': 'allow'}
+        # beforeReadFile has no agent_message; its deny user_message is what reaches the model
+        file_read_deny = decision.event_type == AiHookEventType.FILE_READ and decision.action == DecisionAction.DENY
         return {
             'permission': decision.action.value,  # 'deny' or 'ask'
-            'user_message': decision.user_message or '',
-            'agent_message': decision.agent_message or '',
+            'user_message': decision.model_message if file_read_deny else decision.user_message or '',
+            # The agent does not see user_message, and needs the findings (their sha) to act on its instructions
+            'agent_message': decision.model_message,
         }
 
     def build_session_payload(self, raw_payload: dict) -> AIHookPayload:
diff --git a/cycode/cli/apps/ai_guardrails/scan/handlers.py b/cycode/cli/apps/ai_guardrails/scan/handlers.py
index 5d29129f..81c699db 100644
--- a/cycode/cli/apps/ai_guardrails/scan/handlers.py
+++ b/cycode/cli/apps/ai_guardrails/scan/handlers.py
@@ -10,9 +10,11 @@
 
 import json
 import os
+import sys
 import threading
 from collections.abc import Callable
 from dataclasses import dataclass
+from gettext import ngettext
 from typing import TYPE_CHECKING, Any, NamedTuple
 
 import typer
@@ -46,12 +48,53 @@
 HandlerFn = Callable[[typer.Context, AIHookPayload, dict], HookDecision]
 
 
+def _ignore_commands(shas: tuple[str, ...]) -> list[str]:
+    """Ready-to-run ignore commands, one per finding, in the order the findings are listed.
+
+    Names the binary running this hook, which may not be on PATH (e.g. MDM installs). Repo scope: it lands in
+    the cwd's .cycode/config.yaml, which matches where the hook reads it as long as the agent runs it.
+    """
+    executable = sys.executable if getattr(sys, 'frozen', False) else sys.argv[0]
+    if ' ' in executable:
+        executable = f'"{executable}"'
+    return [f'`{executable} ignore --by-sha {sha}`' for sha in shas]
+
+
+def _with_ignore_commands(shas: tuple[str, ...], one: str, many: str) -> str:
+    """`one` formatted with the single command, or `many` followed by the commands as a list; '' without shas."""
+    commands = _ignore_commands(shas)
+    if len(commands) == 1:
+        return one.format(command=commands[0])
+    if commands:
+        return many + ''.join(f'\n  {command}' for command in commands)
+    return ''
+
+
+def _false_positive_user_hint(shas: tuple[str, ...]) -> str:
+    # The agent never saw the blocked prompt, so the command has to be complete for the developer to pass on
+    return _with_ignore_commands(
+        shas,
+        one='Not a real secret? Ask your code agent to run {command}, then resend.',
+        many='Not real secrets? Ask your code agent to run the command for each one, then resend:',
+    )
+
+
+def _false_positive_agent_hint(shas: tuple[str, ...]) -> str:
+    # Ignoring is the developer's call; the agent may run it for them, but only once they confirm
+    return 'Tell the user what was blocked.' + _with_ignore_commands(
+        shas,
+        one=' Only if the user confirms it is a false positive, run {command} and retry.',
+        many=' Only for a finding the user confirms is a false positive, run its command and retry:',
+    )
+
+
 class ScanOutcome(NamedTuple):
     """What one guardrail scan came back with; the verdict is the server's, which applied the tenant's floors."""
 
     violation_summary: str | None = None
     scan_id: str | None = None
     verdict: GuardrailsMode | None = None
+    shas: tuple[str, ...] = ()
 
 
 NO_SCAN = ScanOutcome()
@@ -96,7 +139,10 @@ def handle_before_submit_prompt(ctx: typer.Context, payload: AIHookPayload, poli
             block_reason = SECRETS_BLOCK_REASON_BY_EVENT_TYPE[AiHookEventType.PROMPT]
             if scan_outcome.verdict == GuardrailsMode.BLOCK:
                 outcome = AIHookOutcome.BLOCKED
-                user_message = f'Remove secrets before sending. {scan_outcome.violation_summary}'
+                secrets = ngettext('secret', 'secrets', len(scan_outcome.shas))
+                user_message = f'{scan_outcome.violation_summary}\nRemove the {secrets} and resend.'
+                if scan_outcome.shas:
+                    user_message += f'\n{_false_positive_user_hint(scan_outcome.shas)}'
                 return HookDecision.deny(AiHookEventType.PROMPT, user_message)
             outcome = AIHookOutcome.WARNED
         return HookDecision.allow(AiHookEventType.PROMPT)
@@ -164,14 +210,15 @@ def handle_before_read_file(ctx: typer.Context, payload: AIHookPayload, policy:
                 block_reason = SECRETS_BLOCK_REASON_BY_EVENT_TYPE[AiHookEventType.FILE_READ]
                 if scan_outcome.verdict == GuardrailsMode.BLOCK:
                     outcome = AIHookOutcome.BLOCKED
-                    user_message = f'Cycode blocked reading {file_path}. {scan_outcome.violation_summary}'
+                    user_message = (
+                        f'Cycode blocked reading {file_path} because it contains a secret.\n'
+                        f'{scan_outcome.violation_summary}'
+                    )
                     return HookDecision.deny(
-                        AiHookEventType.FILE_READ,
-                        user_message,
-                        'Secrets detected; do not send this file to the model.',
+                        AiHookEventType.FILE_READ, user_message, _false_positive_agent_hint(scan_outcome.shas)
                     )
                 outcome = AIHookOutcome.WARNED
-                user_message = f'Cycode detected secrets in {file_path}. {scan_outcome.violation_summary}'
+                user_message = f'Cycode detected secrets in {file_path}.\n{scan_outcome.violation_summary}'
                 return HookDecision.ask(
                     AiHookEventType.FILE_READ,
                     user_message,
@@ -282,7 +329,7 @@ def _handle_arg_scan(
                 return HookDecision.deny(
                     feature.event_type,
                     feature.deny_message(scan_outcome.violation_summary),
-                    feature.deny_agent_message,
+                    f'{feature.deny_agent_message} {_false_positive_agent_hint(scan_outcome.shas)}',
                 )
             outcome = AIHookOutcome.WARNED
             return HookDecision.ask(
@@ -360,9 +407,11 @@ def handle_before_mcp_execution(ctx: typer.Context, payload: AIHookPayload, poli
         policy,
         _ArgScanFeature(
             event_type=AiHookEventType.MCP_EXECUTION,
-            deny_message=lambda v: f'Cycode blocked MCP tool call "{tool}". {v}',
+            deny_message=lambda v: (
+                f'Cycode blocked MCP tool call "{tool}" because its arguments contain a secret.\n{v}'
+            ),
             deny_agent_message='Do not pass secrets to tools. Use secret references (name/id) instead.',
-            ask_message=lambda v: f'Allow MCP tool call "{tool}"? {v}',
+            ask_message=lambda v: f'Allow MCP tool call "{tool}"?\n{v}',
             ask_agent_message='Possible secrets detected in tool arguments; proceed with caution.',
             scan_enabled=get_policy_value(policy, 'mcp', 'scan_args', default=True),
         ),
@@ -478,11 +527,14 @@ def _perform_scan(
     if not local_scan_result:
         return NO_SCAN
 
-    violation_summary = build_violation_summary([local_scan_result]) if local_scan_result.issue_detected else None
+    violation_summary, shas = (
+        build_violation_summary([local_scan_result]) if local_scan_result.issue_detected else (None, ())
+    )
     return ScanOutcome(
         violation_summary=violation_summary,
         scan_id=local_scan_result.scan_id,
         verdict=_parse_verdict(local_scan_result.verdict),
+        shas=shas,
     )
 
 
diff --git a/cycode/cli/apps/ai_guardrails/scan/utils.py b/cycode/cli/apps/ai_guardrails/scan/utils.py
index 645d7636..988c37b5 100644
--- a/cycode/cli/apps/ai_guardrails/scan/utils.py
+++ b/cycode/cli/apps/ai_guardrails/scan/utils.py
@@ -8,14 +8,17 @@
 import os
 import sys
 from collections import defaultdict
+from gettext import ngettext
 from pathlib import Path
 from typing import TYPE_CHECKING
 
 from cycode.cli.apps.ai_guardrails.scan.policy import get_policy_value
 from cycode.cli.cli_types import SeverityOption
+from cycode.cli.utils.string_utils import obfuscate_text
 
 if TYPE_CHECKING:
     from cycode.cli.models import LocalScanResult
+    from cycode.cyclient.models import Detection
 
 # Keeps the hook message readable when a single file trips dozens of detections
 MAX_VIOLATION_DETAIL_LINES = 5
@@ -98,35 +101,50 @@ def output_json(obj: dict) -> None:
     print(json.dumps(obj), end='')  # noqa: T201
 
 
-def _build_detection_lines(
+def _describe_detection(detection: 'Detection', content: str) -> str:
+    details = detection.detection_details
+    description = detection.type or detection.message
+
+    line = details.get('line')
+    if isinstance(line, int) and line >= 0:
+        description += f' (line {line + 1})'  # the scanner reports 0-based lines
+
+    start, length = details.get('start_position'), details.get('length')
+    if isinstance(start, int) and isinstance(length, int) and start >= 0 and length > 0:
+        value = content[start : start + length]
+        if value:
+            description += f': {obfuscate_text(value)}'
+
+    return description
+
+
+def build_violation_summary(
     local_scan_results: list['LocalScanResult'], max_lines: int = MAX_VIOLATION_DETAIL_LINES
-) -> str:
-    """One line per distinct finding: what it is, and the value hash identifying it.
+) -> tuple[str, tuple[str, ...]]:
+    """One line per distinct finding (what it is, where, the masked value), and the value hashes to ignore them by.
 
-    The value hash is safe to display; the value itself is not. Detections excluded by an existing
+    The masked value and hash are safe to display; the value itself is not. Detections excluded by an existing
     ignore rule are already gone from `document_detections`, so only what actually blocked is listed.
     """
-    type_by_sha = {}
+    # Keyed by value hash so a repeated value is listed once; detections without one key by their description
+    finding_by_key = {}
     for local_scan_result in local_scan_results:
         for document_detections in local_scan_result.document_detections:
             for detection in document_detections.detections:
+                description = _describe_detection(detection, document_detections.document.content)
                 sha = detection.detection_details.get('sha512')
-                if sha and sha not in type_by_sha:
-                    type_by_sha[sha] = detection.type or detection.message
-
-    if not type_by_sha:
-        return ''
+                finding_by_key.setdefault(sha or description, (sha, description))
 
-    lines = [f'  - {detection_type}: {sha}' for sha, detection_type in list(type_by_sha.items())[:max_lines]]
-    remaining = len(type_by_sha) - len(lines)
+    listed = list(finding_by_key.values())[:max_lines]
+    lines = [_build_severity_line(local_scan_results)] + [f'  - {description}' for _, description in listed]
+    remaining = len(finding_by_key) - len(listed)
     if remaining:
         lines.append(f'  - ...and {remaining} more')
 
-    return '\n' + '\n'.join(lines)
+    return '\n'.join(lines), tuple(sha for sha, _ in listed if sha)
 
 
-def build_violation_summary(local_scan_results: list['LocalScanResult']) -> str:
-    """Build violation summary string with severity breakdown and emojis."""
+def _build_severity_line(local_scan_results: list['LocalScanResult']) -> str:
     detections_count = 0
     severity_counts = defaultdict(int)
 
@@ -143,5 +161,5 @@ def build_violation_summary(local_scan_results: list['LocalScanResult']) -> str:
         count = severity_counts[severity]
         severity_parts.append(f'{emoji} {severity.upper()} - {count}')
 
-    summary = f'Cycode found {detections_count} violations: {" | ".join(severity_parts)}'
-    return summary + _build_detection_lines(local_scan_results)
+    violations = ngettext('violation', 'violations', detections_count)
+    return f'Cycode found {detections_count} {violations}: {" | ".join(severity_parts)}'
diff --git a/tests/cli/commands/ai_guardrails/ides/test_claude_code.py b/tests/cli/commands/ai_guardrails/ides/test_claude_code.py
index 12652160..d7d37e3e 100644
--- a/tests/cli/commands/ai_guardrails/ides/test_claude_code.py
+++ b/tests/cli/commands/ai_guardrails/ides/test_claude_code.py
@@ -194,9 +194,11 @@ def test_parse_empty_payload_defaults() -> None:
 def test_build_prompt_responses() -> None:
     claude = ClaudeCode()
     assert claude.build_hook_response(HookDecision.allow(AiHookEventType.PROMPT)) == {}
+    # The blocked prompt carries the secret, so it must not be echoed back under the reason
     assert claude.build_hook_response(HookDecision.deny(AiHookEventType.PROMPT, 'no!')) == {
         'decision': 'block',
         'reason': 'no!',
+        'hookSpecificOutput': {'hookEventName': 'UserPromptSubmit', 'suppressOriginalPrompt': True},
     }
 
 
@@ -210,7 +212,8 @@ def test_build_permission_responses() -> None:
         'hookSpecificOutput': {
             'hookEventName': 'PreToolUse',
             'permissionDecision': 'deny',
-            'permissionDecisionReason': 'user!',
+            # A deny reason reaches the model, which needs both the findings and its instructions
+            'permissionDecisionReason': 'user!\nagent!',
         }
     }
 
diff --git a/tests/cli/commands/ai_guardrails/ides/test_codex.py b/tests/cli/commands/ai_guardrails/ides/test_codex.py
index b867de39..e69ef0b7 100644
--- a/tests/cli/commands/ai_guardrails/ides/test_codex.py
+++ b/tests/cli/commands/ai_guardrails/ides/test_codex.py
@@ -149,12 +149,12 @@ def test_build_mcp_execution_allow_and_deny() -> None:
     # Explicit permissionDecision "allow" is rejected by older Codex runtimes.
     assert allow == {}
 
-    deny = codex.build_hook_response(HookDecision.deny(AiHookEventType.MCP_EXECUTION, 'secret in args!'))
+    deny = codex.build_hook_response(HookDecision.deny(AiHookEventType.MCP_EXECUTION, 'secret in args!', 'agent!'))
     assert deny == {
         'hookSpecificOutput': {
             'hookEventName': 'PreToolUse',
             'permissionDecision': 'deny',
-            'permissionDecisionReason': 'secret in args!',
+            'permissionDecisionReason': 'secret in args!\nagent!',
         }
     }
 
diff --git a/tests/cli/commands/ai_guardrails/ides/test_copilot.py b/tests/cli/commands/ai_guardrails/ides/test_copilot.py
index eed1b8a6..7914e76a 100644
--- a/tests/cli/commands/ai_guardrails/ides/test_copilot.py
+++ b/tests/cli/commands/ai_guardrails/ides/test_copilot.py
@@ -282,12 +282,12 @@ def test_deny_prompt_response_shape() -> None:
 
 
 def test_deny_tool_response_shape() -> None:
-    response = Copilot().build_hook_response(HookDecision.deny(AiHookEventType.FILE_READ, 'Sensitive file'))
+    response = Copilot().build_hook_response(HookDecision.deny(AiHookEventType.FILE_READ, 'Sensitive file', 'agent!'))
     assert response == {
         'hookSpecificOutput': {
             'hookEventName': 'PreToolUse',
             'permissionDecision': 'deny',
-            'permissionDecisionReason': 'Sensitive file',
+            'permissionDecisionReason': 'Sensitive file\nagent!',
         }
     }
 
diff --git a/tests/cli/commands/ai_guardrails/ides/test_cursor.py b/tests/cli/commands/ai_guardrails/ides/test_cursor.py
index 60834020..f54e6752 100644
--- a/tests/cli/commands/ai_guardrails/ides/test_cursor.py
+++ b/tests/cli/commands/ai_guardrails/ides/test_cursor.py
@@ -175,15 +175,22 @@ def test_build_prompt_responses() -> None:
 def test_build_permission_responses() -> None:
     cursor = Cursor()
     assert cursor.build_hook_response(HookDecision.allow(AiHookEventType.FILE_READ)) == {'permission': 'allow'}
+    # beforeReadFile drops agent_message and feeds user_message to the model, so it carries both
     assert cursor.build_hook_response(HookDecision.deny(AiHookEventType.FILE_READ, 'user!', 'agent!')) == {
+        'permission': 'deny',
+        'user_message': 'user!\nagent!',
+        'agent_message': 'user!\nagent!',
+    }
+    assert cursor.build_hook_response(HookDecision.deny(AiHookEventType.MCP_EXECUTION, 'user!', 'agent!')) == {
         'permission': 'deny',
         'user_message': 'user!',
-        'agent_message': 'agent!',
+        # The agent never sees user_message, so it gets the findings too
+        'agent_message': 'user!\nagent!',
     }
     assert cursor.build_hook_response(HookDecision.ask(AiHookEventType.MCP_EXECUTION, 'u', 'a')) == {
         'permission': 'ask',
         'user_message': 'u',
-        'agent_message': 'a',
+        'agent_message': 'u\na',
     }
 
 
diff --git a/tests/cli/commands/ai_guardrails/scan/test_handlers.py b/tests/cli/commands/ai_guardrails/scan/test_handlers.py
index 88c8f30a..41ab6c33 100644
--- a/tests/cli/commands/ai_guardrails/scan/test_handlers.py
+++ b/tests/cli/commands/ai_guardrails/scan/test_handlers.py
@@ -1,6 +1,7 @@
 """Tests for AI guardrails handlers."""
 
 import os
+import sys
 import threading
 import time
 from multiprocessing import synchronize
@@ -96,18 +97,33 @@ def test_handle_before_submit_prompt_with_secrets_blocked(
     mock_scan: MagicMock, mock_ctx: MagicMock, mock_payload: AIHookPayload, default_policy: dict[str, Any]
 ) -> None:
     """Test that prompt with secrets is blocked."""
-    mock_scan.return_value = ScanOutcome('Found 1 secret: API key', 'scan-id-456', GuardrailsMode.BLOCK)
+    mock_scan.return_value = ScanOutcome('Found 1 secret: API key', 'scan-id-456', GuardrailsMode.BLOCK, ('sha-aaa',))
 
-    result = handle_before_submit_prompt(mock_ctx, mock_payload, default_policy)
+    with patch.object(sys, 'argv', ['/opt/cycode/bin/cycode']):
+        result = handle_before_submit_prompt(mock_ctx, mock_payload, default_policy)
 
     assert result.action == DecisionAction.DENY
     assert result.event_type == AiHookEventType.PROMPT
     assert 'Found 1 secret: API key' in result.user_message
+    # The agent never saw the prompt, so the developer gets a complete command (running binary, real sha) to pass on
+    assert 'Ask your code agent to run `/opt/cycode/bin/cycode ignore --by-sha sha-aaa`' in result.user_message
     mock_ctx.obj['ai_security_client'].create_event.assert_called_once()
     call_args = mock_ctx.obj['ai_security_client'].create_event.call_args
     assert call_args.args[2] == AIHookOutcome.BLOCKED
     assert call_args.kwargs['block_reason'] == BlockReason.SECRETS_IN_PROMPT
 
+    # Several secrets: one command per finding, in the listed order
+    mock_scan.return_value = ScanOutcome('Found 2 secrets', 'scan-id-457', GuardrailsMode.BLOCK, ('sha-aaa', 'sha-bbb'))
+    with patch.object(sys, 'argv', ['/opt/cycode/bin/cycode']):
+        result = handle_before_submit_prompt(mock_ctx, mock_payload, default_policy)
+
+    assert 'Remove the secrets and resend.' in result.user_message
+    assert result.user_message.endswith(
+        'then resend:\n'
+        '  `/opt/cycode/bin/cycode ignore --by-sha sha-aaa`\n'
+        '  `/opt/cycode/bin/cycode ignore --by-sha sha-bbb`'
+    )
+
 
 @patch('cycode.cli.apps.ai_guardrails.scan.handlers._scan_text_for_secrets')
 def test_handle_before_submit_prompt_with_secrets_warned(
@@ -226,18 +242,25 @@ def test_handle_before_read_file_with_secrets(
 ) -> None:
     """Test that file with secrets is blocked, and merely reported when the verdict says so."""
     mock_is_denied.return_value = False
-    mock_scan.return_value = ScanOutcome('Found 1 secret: password', 'scan-id-456', GuardrailsMode.BLOCK)
+    mock_scan.return_value = ScanOutcome(
+        'Found 1 secret: password', 'scan-id-456', GuardrailsMode.BLOCK, ('sha-aaa', 'sha-bbb')
+    )
     payload = AIHookPayload(
         event_name='FileRead',
         ide_provider='cursor',
         file_path='/path/to/file.txt',
     )
 
-    result = handle_before_read_file(mock_ctx, payload, default_policy)
+    with patch.object(sys, 'argv', ['/opt/cycode/bin/cycode']):
+        result = handle_before_read_file(mock_ctx, payload, default_policy)
 
     assert result.action == DecisionAction.DENY
     assert result.event_type == AiHookEventType.FILE_READ
     assert 'Found 1 secret: password' in result.user_message
+    # The agent may ignore a finding, but only on the developer's confirmation
+    assert 'Only for a finding the user confirms is a false positive' in result.agent_message
+    assert '\n  `/opt/cycode/bin/cycode ignore --by-sha sha-aaa`' in result.agent_message
+    assert '\n  `/opt/cycode/bin/cycode ignore --by-sha sha-bbb`' in result.agent_message
     call_args = mock_ctx.obj['ai_security_client'].create_event.call_args
     assert call_args.args[2] == AIHookOutcome.BLOCKED
     assert call_args.kwargs['block_reason'] == BlockReason.SECRETS_IN_FILE
@@ -485,34 +508,46 @@ def _local_scan_result_with_detections(*shas: str) -> LocalScanResult:
 
 
 def test_violation_summary_lists_one_line_per_distinct_sha() -> None:
-    """A blocked developer needs the value hash to act on the finding (e.g. `cycode ignore --by-sha`)."""
-    summary = build_violation_summary([_local_scan_result_with_detections('sha-aaa', 'sha-bbb', 'sha-aaa')])
+    """A blocked developer needs to find the value (line, masked value) and act on it (`cycode ignore --by-sha`)."""
+    local_scan_result = _local_scan_result_with_detections('sha-aaa', 'sha-bbb', 'sha-aaa')
+    local_scan_result.document_detections[0].document.content = 'first line\ntoken=ghp_abcdefghijklmnop\n'
+    local_scan_result.document_detections[0].detections[0].detection_details.update(
+        {'line': 1, 'start_position': 17, 'length': 20}
+    )
+
+    summary, shas = build_violation_summary([local_scan_result])
 
     assert 'Cycode found 3 violations' in summary
-    assert 'GitHub Token: sha-aaa' in summary
-    assert 'GitHub Token: sha-bbb' in summary
-    # Repeated values collapse to one line; the hash identifies the value, not the occurrence
-    assert summary.count('sha-aaa') == 1
+    assert '  - GitHub Token (line 2): ghp**************nop' in summary
+    assert 'ghp_abcdefghijklmnop' not in summary
+    # Without position details there is nothing to locate or mask, but the finding is still listed
+    assert '  - GitHub Token\n' in summary + '\n'
+    # Repeated values collapse to one finding; the hash identifies the value, not the occurrence
+    assert shas == ('sha-aaa', 'sha-bbb')
 
 
 def test_violation_summary_caps_the_detection_lines() -> None:
     """A file full of detections must not turn the hook message into a wall of text."""
     shas = [f'sha-{index}' for index in range(MAX_VIOLATION_DETAIL_LINES + 3)]
 
-    summary = build_violation_summary([_local_scan_result_with_detections(*shas)])
+    summary, listed_shas = build_violation_summary([_local_scan_result_with_detections(*shas)])
 
-    assert summary.count('GitHub Token: ') == MAX_VIOLATION_DETAIL_LINES
+    assert summary.count('  - GitHub Token') == MAX_VIOLATION_DETAIL_LINES
     assert '...and 3 more' in summary
+    # Only listed findings get an ignore command
+    assert listed_shas == tuple(shas[:MAX_VIOLATION_DETAIL_LINES])
 
 
-def test_violation_summary_omits_lines_without_a_sha() -> None:
-    """Non-secret scan types carry no value hash, so there is nothing to list."""
+def test_violation_summary_lists_findings_without_a_sha() -> None:
+    """Non-secret scan types carry no value hash: the finding is still listed, there is just nothing to ignore by."""
     local_scan_result = _local_scan_result_with_detections('sha-aaa')
     local_scan_result.document_detections[0].detections[0].detection_details = {}
 
-    summary = build_violation_summary([local_scan_result])
+    summary, shas = build_violation_summary([local_scan_result])
 
-    assert 'GitHub Token' not in summary
+    assert 'Cycode found 1 violation:' in summary
+    assert '  - GitHub Token' in summary
+    assert shas == ()
 
 
 # Tests for handle_before_mcp_execution