feat: support model-native Deep Scan workers

sdk/typescript/_bundled_plugin/preflight/capability-profiles.toml

@@ -1,16 +1,5 @@
 version = 1
 
-[capabilities.deep_scan_phase_skills]
-kind = "plugin_skills"
-plugin = "codex-security"
-required = [
-  "attack-path-analysis",
-  "finding-discovery",
-  "security-scan",
-  "threat-model",
-  "validation",
-]
-
 [capabilities.delegated_workers]
 kind = "runtime"
 check = "delegation_available"
@@ -45,15 +34,6 @@ op = "=="
 value = true
 default = true
 
-[capabilities.native_multi_agent_v2]
-kind = "multi_agent_mode"
-owner = "native"
-version = "v2"
-
-[capabilities.sdk_v2_child_config_compatible]
-kind = "config_absent"
-path = "agents.max_threads"
-
 [profiles.security_diff_scan]
 description = "Capabilities for Git-backed Codex Security diff scans."
 
@@ -98,21 +78,6 @@ reason = "Enabling goals makes long scans easier to resume and audit."
 [profiles.deep_security_scan]
 description = "Capabilities for deep repository-wide Codex Security scans."
 
-[[profiles.deep_security_scan.requirements]]
-capability = "deep_scan_phase_skills"
-severity = "block"
-reason = "Deep scan depends on Codex Security phase skills loaded in the current runtime."
-
-[[profiles.deep_security_scan.requirements]]
-capability = "native_multi_agent_v2"
-severity = "block"
-reason = "Deep scan launches SDK discovery workers with native Subagents v2; legacy V1 and bridge-owned parent configurations are incompatible with that worker runtime."
-
-[[profiles.deep_security_scan.requirements]]
-capability = "sdk_v2_child_config_compatible"
-severity = "block"
-reason = "Deep scan child sessions inherit Codex configuration while enabling Subagents v2, so the legacy agents.max_threads setting must be removed before those sessions can start."
-
 [[profiles.deep_security_scan.requirements]]
 capability = "goal_tools"
 severity = "suggest"
@@ -144,28 +109,6 @@ summary = "Recommended Codex setup for deep repository-wide scans."
 path = "features.goals"
 value = true
 
-[[profiles.deep_security_scan.remediation.variants]]
-mode = "v1"
-
-[[profiles.deep_security_scan.remediation.variants.patches]]
-kind = "remove"
-path = "agents.max_threads"
-
-[[profiles.deep_security_scan.remediation.variants.patches]]
-path = "features.multi_agent_v2.enabled"
-value = true
-
-[[profiles.deep_security_scan.remediation.variants.patches]]
-path = "features.multi_agent_v2.max_concurrent_threads_per_session"
-value = 4
-
-[[profiles.deep_security_scan.remediation.variants]]
-mode = "v2"
-
-[[profiles.deep_security_scan.remediation.variants.patches]]
-kind = "remove"
-path = "agents.max_threads"
-
 [[routes]]
 skill = "security-diff-scan"
 profile = "security_diff_scan"

sdk/typescript/_bundled_plugin/references/config-preflight.md

@@ -60,7 +60,7 @@ When the profile includes remediation patches, present the concrete config delta
 
 Some remediation patches have `kind = "host_setting"`. Present those as host-level setup guidance, not as edits to persistent Codex config.
 
-Deep Security Scan uses MCP-owned SDK sessions rather than the parent thread's worker pool. Its preflight requires native V2 but does not derive outer worker concurrency from the parent session cap or require parent delegation depth. When migrating a legacy V1 configuration for Deep Security Scan, remediation removes `agents.max_threads`, sets `features.multi_agent_v2.enabled = true`, and sets `features.multi_agent_v2.max_concurrent_threads_per_session = 4`. The coordinator overrides each child session's cap from the per-user deep-scan configuration. Codex rejects the legacy V1 thread setting and explicit V2 mode together.
+Deep Security Scan uses MCP-owned SDK sessions rather than the parent thread's worker pool. Its preflight does not require a particular parent delegation runtime, ownership, capacity, or depth. Discovery workers inherit the scan's model and run under the verified read-only worker sandbox.
 
 Do not warn merely because a user's value differs from the profile's suggested patch. Warn or block only when the evaluated capability requirement is unmet.
 

sdk/typescript/_bundled_plugin/skills/deep-security-scan/SKILL.md

@@ -72,7 +72,7 @@ Confirm these plugin skills are available in the active runtime:
 - `$codex-security:validation`
 - `$codex-security:attack-path-analysis`
 
-The discovery tool launches Codex workers that may use Subagents v2. The active configuration must satisfy the deep profile's native-v2 requirement. The worker count is configured separately from this thread's subagent allowance.
+The discovery tool manages its own workers independently of this thread's delegation runtime and subagent allowance.
 
 Continue after a `ready` result, explaining material warn or suggest limitations. For `blocked` or `incomplete` results with actionable remediation, first classify the session using `../../references/config-preflight.md`. In an interactive session, present the exact reasons, helper-reported config file path, and config changes, then use that reference's native `request_user_input` → `request_codex_security_user_input` → plain-chat fallback sequence before editing persistent configuration. Stop for the answer without creating a goal or starting discovery. In `codex exec`, headless, automation, or another non-interactive session, do not ask or wait; apply only helper-provided ordinary config patches to the helper's `user_config_path`, rerun preflight once, and continue only if it becomes `ready`. Never guess which Codex home is active or hide a higher-precedence conflict with a lower-precedence edit. If an interactive user declines required remediation, ask whether to cancel the durable desktop scan with `cancel_codex_security_scan` or leave it running for a later retry.