release: prepare Socket v10.0.0

CONTRIBUTING.md

@@ -151,33 +151,6 @@ The root [`Socket.xcworkspace`](./Socket.xcworkspace) is a browse-only workspace
 
 Do not add a generated `.xcodeproj`, root `Package.swift`, or workspace scheme only to improve documentation editing. If Socket later gains a real root build product, document that build surface separately and update the workspace guidance in [`docs/maintainers/socket-xcode-workspace.md`](./docs/maintainers/socket-xcode-workspace.md).
 
-### Repo-Local Steward
-
-The repo-local Socket Steward prototype lives at [`.agents/socket-steward/`](./.agents/socket-steward/). It is a Python `uv` project built around the OpenAI Agents SDK, with deterministic read-only audits available without API credentials and an optional agent-backed `ask` command when `OPENAI_API_KEY` is available.
-
-Use it as maintainer assistance, not as a release gate or background service:
-
-```bash
-uv run --directory .agents/socket-steward socket-steward audit docs
-uv run --directory .agents/socket-steward socket-steward audit guidance
-uv run --directory .agents/socket-steward socket-steward audit marketplace
-uv run --directory .agents/socket-steward socket-steward plan docs-sync
-uv run --directory .agents/socket-steward socket-steward propose docs-sync
-uv run --directory .agents/socket-steward socket-steward propose docs-sync --output
-uv run --directory .agents/socket-steward socket-steward prepare docs-sync --output
-uv run --directory .agents/socket-steward socket-steward apply docs-sync --confirm
-```
-
-Socket Steward proposal report writes are limited to [`docs/agents/`](./docs/agents/). Use that directory for reviewable agent-generated Markdown reports; move durable conclusions into the owning root docs, maintainer docs, child guidance, roadmap, scripts, or marketplace metadata instead of treating reports as policy. The first guarded `apply docs-sync --confirm` mode refreshes the proposal report only; it does not mutate durable docs for TODO plans yet.
-
-Validate the steward package from its own directory:
-
-```bash
-uv run --directory .agents/socket-steward pytest
-uv run --directory .agents/socket-steward ruff check .
-uv run --directory .agents/socket-steward mypy .
-```
-
 ### Marketplace Shape
 
 The repo-root marketplace lives at [`.agents/plugins/marketplace.json`](./.agents/plugins/marketplace.json). It is a catalog, not a root aggregate plugin.

README.md

@@ -190,8 +190,7 @@ For Xcode 27 beta Markdown editing and repository browsing, open [`Socket.xcwork
 ```text
 .
 ├── .agents/
-│   ├── plugins/marketplace.json
-│   └── socket-steward/
+│   └── plugins/marketplace.json
 ├── docs/
 │   ├── agents/
 │   ├── media/

ROADMAP.md

@@ -542,7 +542,6 @@ In progress
 - [x] Treat Agent Skills as the first portability layer while keeping Codex plugins, hooks, MCP registration, custom agents, and host package formats as target-specific adapters.
 - [ ] Keep Socket's root Codex marketplace model intact until a concrete non-Codex package or export target proves it needs a broader distribution abstraction.
 - [x] Rename `agent-plugin-skills` to `agent-portability-skills` so the child plugin name matches the cross-host compatibility role.
-- [ ] Keep Socket Steward tied into this milestone as the repo-local audit, plan, and proposal engine, while `agent-portability-skills` owns reusable agent-facing portability workflows.
 - [ ] Route complex local orchestration through AgentUtils once that app exposes supported discovery, dry-run, backup, and apply contracts instead of expanding Socket plugin payloads into broad machine-management code.
 - [x] Add a checked-in Hermes Agent skill tap at root `skills/`, generated from the canonical `agent-portability-skills` source and grouped with `skills.sh.json`.
 - [x] Add `agent-portability-skills:hermes-agent-compatibility` with explicit skill, Codex bundle, MCP, and native Python plugin boundaries.
@@ -1190,22 +1189,14 @@ and test/production deployments for GitHub Actions.
 - [x] Add `game-dev-skills:metal-game-rendering-workflow` for native Metal 3 and Metal 4 renderer architecture; keep D3D/Vulkan porting in GPTK routing, MetalFX in its own workflow, and profiling evidence in `xcode-game-profiling-workflow`.
 - [x] Add `game-dev-skills:game-porting-toolkit-workflow` to choose stable GPTK 3 evaluation/shader-conversion work or beta-sensitive GPTK 4 source-port work, then load Apple’s upstream GPTK 4 skills instead of duplicating their corpus.
 - [x] Add `game-dev-skills:metalfx-game-rendering-workflow`, `metal-asset-streaming-workflow`, and `metal-neural-rendering-workflow` for native MetalFX integration, GPU asset residency/streaming, and carefully gated experimental neural rendering.
-- [x] Add the first repo-local Socket Steward prototype as a Python `uv` project under `.agents/socket-steward`, using deterministic read-only audits plus an optional OpenAI Agents SDK `ask` path before any write, LaunchAgent, or app behavior.
-- [x] Expand Socket Steward with a docs-sync planning command that emits structured recommended edits for README, CONTRIBUTING, AGENTS, ROADMAP, marketplace metadata, and child plugin guidance without applying them.
-- [x] Add `docs/agents/` as the repo-local report surface and let Socket Steward write reviewable docs-sync proposal reports there without applying the proposed documentation edits.
-- [x] Add a serialized Socket Steward `prepare docs-sync` workflow and first guarded `apply docs-sync --confirm` mode that refreshes proposal reports without mutating durable docs.
-- [ ] Add a read-only Socket Steward fan-out experiment for broad docs and guidance scans. Start with deterministic sharding by file count or total line count, keep workers read-only, merge findings into one bounded report, and compare the result against the single-process audit before deciding whether subagent fan-out belongs in the durable steward workflow.
-- [ ] Add a guarded Socket Steward write mode only after the read-only audit and planning contracts are stable, with explicit approval boundaries for file edits, validation, git operations, release workflow, and future background execution.
 - [x] Overhaul `agent-portability-skills` so its docs, tests, generated bootstrap content, and sync audit logic target Codex/OpenAI plus the open `.agents/skills` discovery mirror only. Remove stale expectations for retired child maintainer docs such as reality-audit and install-surface docs, and keep the wording away from unsupported non-Codex or generic multi-agent surfaces.
 - [x] Consolidate README, CONTRIBUTING, AGENTS, API, coordinated docs, and ROADMAP maintenance workflows into `repository-skills` so repository operations and repository documentation have one owner.
 - [x] Add a first `agent-engineering-skills:design-agent-automation-workflow` planning skill for agent and automation design. It chooses between Codex app automations, `codex exec`, Codex subagents, OpenAI Agents SDK services, LangGraph graphs, Hermes-specific workflows, or no automation yet while delegating stack-specific implementation to the owning plugin.
 - [x] Added `agent-engineering-skills:design-agent-eval-workflow` for agent, skill, prompt, and automation eval planning, and skewed automation guidance toward safe full automation with exact escalation gates instead of broad human review.
 - [x] Replaced the legacy Apple app guidance and sync routes with the unified `bootstrap-xcode-workspace` workflow and its managed `just align` contract.
 - [x] Added the former direct-main patch-refresh helper, then retired it when every semantic-version level and catalog refresh moved into the single protected-main release lifecycle.
 - [x] Reduce hand-carried release work by capturing commit-bound temporary `CODEX_HOME` marketplace smoke evidence and the final Dependabot alert query during `advance`, then append only pre-publication evidence to the checked-in release notes.
-- [ ] Explore steward-assisted release and worktree orchestration. Start with Socket Steward release preflights and cache-refresh checks, then evaluate whether `swift-steward` or sibling roles should handle read-only release readiness, PR merge sequencing, branch accounting, and parallel worktree status reports while the main thread keeps write, merge, tag, and publish ownership.
-- [ ] Design a worker-thread orchestration workflow for Codex GUI and Socket Steward use. Capture the decisions before implementation: whether the durable surface is a new Productivity Skill, a Socket Steward command, or both; which fields belong in the worker launch envelope; how model and reasoning budgets are selected; how workers report branch, worktree, validation, and cleanup state back to the coordinator; which actions remain main-thread only; and when a finished worker thread or worktree should be archived, removed, or kept for follow-up.
-- [ ] Keep Socket Steward tied into cross-Socket docs, marketplace, audit, plan, and proposal workflows as the repo-local coordination surface.
+- [ ] Design a worker-thread orchestration workflow for Codex GUI use. Capture which fields belong in the worker launch envelope, how model and reasoning budgets are selected, how workers report branch, worktree, validation, and cleanup state, which actions remain main-thread only, and when a finished worker thread or worktree should be archived, removed, or kept for follow-up.
 - [ ] Add `agent-portability-skills:audit-agent-surface-portability` for inventorying `SKILL.md`, `.codex-plugin`, `.mcp.json`, hooks, app config, custom agents, and host compatibility notes across Socket child plugins.
 - [ ] Add `agent-portability-skills:design-agent-host-adapter` for deciding whether a host needs docs-only guidance, `.agents/skills` export, native MCP config, a plugin or package adapter, or no Socket-specific support.
 - [ ] Add `agent-portability-skills:maintain-codex-plugin-surface` for Codex-specific marketplace, plugin manifest, hooks, MCP, app config, and enablement wording.
@@ -1257,13 +1248,9 @@ and test/production deployments for GitHub Actions.
 - Re-contained SwiftData persistence guidance in a dedicated Apple Dev skill and SwiftUI composition in its architecture skill, while introducing the explicit three-letter Swift prefix and Xcode-friendly concatenated filename grammar.
 - Made Socket worktree-first for implementation work while keeping the base `main` checkout as the clean coordination and release-verification surface.
 - Aligned Socket documentation-source routing away from generic documentation aggregators by making Xcode MCP `DocumentationSearch` the Apple SDK default, Dash MCP/HTTP the preferred local-docs path for installed docsets across supported stacks, and canonical upstream docs/source the fallback when Dash/local coverage is missing or stale.
-- Added the first repo-local Socket Steward prototype under `.agents/socket-steward`, giving the superproject a Python and OpenAI Agents SDK maintainer-agent scaffold with offline docs, guidance, and marketplace audits before any write-capable or background-service behavior.
-- Added Socket Steward's first docs-sync planner so the repo-local agent can produce structured read-only documentation alignment work before any guarded write mode exists.
-- Added `docs/agents/` for repo-local agent report artifacts and limited Socket Steward proposal writes to that directory.
 - Planned an `agentdeck` desktop bridge MCP and skill surface that talks to the separate `AgentDeck` macOS app over a local transport instead of bundling a signed app in the plugin cache.
 - Planned Codex GUI restart request/cancel/status tools and a narrow skill that keep restart execution in `AgentDeck` and leave automatic `when-idle` waiting blocked until a supported thread-status source exists.
 - Planned an AgentDeck agent configuration sync surface so normal Codex, Xcode Codex, and Xcode Claude can be discovered, diffed, and rendered through target-specific compatibility rules while `agentdeck` remains the Codex-facing adapter.
-- Added serialized Socket Steward prepare/apply commands so maintainers can run audit, docs-sync planning, and proposal refresh in one guarded pass.
 - Extended the roadmap maintainer skill with one-ticket add/update flags and updated automation-design guidance to prefer existing skills, plugins, scripts, and official workflow owners as the source of truth for workflow knowledge.
 - Removed the stale Apple Dev Skills release-time subtree push gate after the standalone Apple Dev Skills repository became a compatibility marketplace pointer to Socket's canonical `plugins/apple-dev-skills` payload.
 - Completed the Apple Dev Skills Phase 2 ownership cleanup: Socket now documents `plugins/apple-dev-skills` as monorepo-owned, the standalone `gaelic-ghost/apple-dev-skills` repository is pruned to a compatibility marketplace pointer, and the compatibility install smoke path is documented alongside the Socket install tests.

docs/maintainers/agent-portability-options.md

@@ -25,7 +25,7 @@ per-plugin boundary. Claude Code supports the full classified catalog; Cowork
 uses the same skills but only public remote MCP servers. Local Mac MCP servers
 remain Claude Code-only.
 
-Use `agent-portability-skills` as the reusable skill layer for these decisions. Socket Steward remains the repo-local audit, plan, and proposal engine under `.agents/socket-steward/`; Agent Portability Skills should own the agent-facing workflows that decide what is portable, what is Codex-specific, and what needs a host adapter.
+Use `agent-portability-skills` as the reusable skill layer for these decisions. It owns the agent-facing workflows that decide what is portable, what is Codex-specific, and what needs a host adapter; root validation and maintainer documentation own Socket-wide consistency checks.
 
 ## Near-Term Focus
 

docs/maintainers/agents-guidance-consolidation-plan.md

@@ -22,11 +22,11 @@ items. It mixes five kinds of guidance:
 4. cross-plugin product guidance; and
 5. current inventories, commands, and historical context.
 
-The deterministic Socket Steward checks both pass:
+The deterministic root guidance and documentation checks both pass:
 
 ```text
-socket-steward audit guidance: PASS
-socket-steward audit docs: PASS
+root guidance audit: PASS
+root documentation audit: PASS
 ```
 
 Those checks establish structural health, not semantic consistency. Manual
@@ -205,8 +205,6 @@ Apply the consolidation by current root section, not by ad hoc sentence edits:
 Run validation serially:
 
 ```bash
-uv run --directory .agents/socket-steward socket-steward audit guidance
-uv run --directory .agents/socket-steward socket-steward audit docs
 uv run scripts/validate_socket.py --profile compatibility
 ```
 
@@ -237,7 +235,7 @@ authority checks, and name the relevant validation surface.
 - Conditional reading replaces the unconditional root-doc reading list.
 - The Speak Swiftly source model and deferred-wakeup references are consistent
   across all live root docs.
-- Socket Steward guidance/docs audits and the compatibility profile pass.
+- Root guidance/documentation audits and the compatibility profile pass.
 - The final review reports word-count reduction and the behavior-preservation
   checklist, not just a clean diff.
 
@@ -281,9 +279,9 @@ now guard the word budget, live-owner routes, directly visible hard stops,
 absence of reintroduced domain/command detail, and the retired Speak Swiftly
 mirror model.
 
-Final validation passed with both Socket Steward audits, all 140 root tests,
-Mypy, Ruff, root marketplace and shared-skill metadata checks, and the Hermes
-and Claude compatibility validators.
+Final validation passed with the root guidance and documentation audits, all
+140 root tests, Mypy, Ruff, root marketplace and shared-skill metadata checks,
+and the Hermes and Claude compatibility validators.
 
 ## Suggested Change Slices
 

docs/maintainers/project-audit-skills-plugin-plan.md

@@ -139,7 +139,7 @@ score without explaining which files, commands, or observations justify it.
 - Should the long-term home be a new child plugin or a focused expansion of
   `codebase-understanding-skills`?
 - Should grading output be a Markdown report only, or should it also support a
-  small JSON shape for future Socket Steward ingestion?
+  small JSON shape for automation and downstream analysis?
 - Should "slop risk" remain internal wording, or should there be a user-facing
   "complexity risk" grade with the same practical meaning?
-- Which existing Socket Steward audit outputs should seed the first examples?
+- Which existing root audit outputs should seed the first examples?

docs/maintainers/release-workflow.md

@@ -65,11 +65,10 @@ scripts/release.sh inspect 10.0.0
 
 If GitHub is still pending, reuse one matching host-native continuation no
 sooner than five minutes later. On wakeup, run `inspect` again before any
-mutation. Failed checks, requested changes, and unreviewed comments block the
-release. A CodeRabbit quota or rate-limit diagnostic may be reviewed and marked
-addressed only when it contains no substantive finding. After that review,
-pass `--review-comments-addressed` to `advance`; the flag does not bypass
-failed or pending checks, requested changes, or commit-identity validation.
+mutation. Failed or pending checks, requested changes, and unreviewed comments
+block the release. After reviewing and resolving comments, pass
+`--review-comments-addressed` to `advance`; the flag does not bypass failed or
+pending checks, requested changes, or commit-identity validation.
 
 The required GitHub `validate` job runs the same full Socket profile used by
 `prepare`; release PRs do not rely on a weaker compatibility-only check.

docs/releases/v10.0.0.md

@@ -18,6 +18,12 @@ Socket release lifecycle.
   server-development path.
 - GitHub Actions exclusively builds Linux images and owns test and production
   cloud deployments from immutable artifacts.
+- Removed the retired repo-local maintainer agent and its reports, validation
+  hooks, documentation, roadmap entries, and generated guidance references.
+- Hermes compatibility now rejects exported skill descriptions over 240
+  characters instead of emitting advisory warnings.
+- Repository-maintenance release gates are now provider-neutral: every pending
+  check waits, every failed check blocks, and unresolved comments require review.
 - Socket releases now use one branch-backed `prepare` → `inspect` → `advance`
   workflow with full local and GitHub validation, reviewed-main evidence,
   annotated tags, structured branch accounting, and marketplace refresh last.
@@ -32,6 +38,10 @@ Socket release lifecycle.
   developer Fly deployment path.
 - Removed Socket's direct-main `patch-refresh`, version-only mutation commands,
   separate release modes, and blanket unmerged-branch override.
+- Removed the repo-local maintainer-agent product and its audit, planning,
+  proposal, and documentation-sync surfaces.
+- Removed the named third-party review-bot helper, installer payload, release
+  exemptions, tests, and guidance from repository maintenance.
 
 ## Migration
 

plugins/agent-engineering-skills/skills/design-agent-automation-workflow/references/local-agent-frameworks.md

@@ -49,8 +49,7 @@ resembles the real workflow.
 
 Choose the OpenAI Agents SDK when the application is primarily an OpenAI agent
 service: typed tools, handoffs, guardrails, approval points, traces, and a
-server-owned execution loop. Socket Steward is the deliberately small,
-read-only local example; it is not a mandate to build every agent as a service.
+server-owned execution loop.
 
 - Good first lab: a read-only repository analyst that returns a typed audit
   report and never invokes a mutating tool.