diff --git a/context/project-skills/openspace/README.md b/context/project-skills/openspace/README.md
new file mode 100644
index 0000000..247a81a
--- /dev/null
+++ b/context/project-skills/openspace/README.md
@@ -0,0 +1,8 @@
+## OpenSpace Project Skills
+
+This directory mirrors project-scoped skills that are intentionally installed
+outside the repository under `~/.codex/projects/openspace/skills/`.
+
+Use it to keep reusable project-routing and workflow skills versioned in Git so
+they can be reviewed, migrated to other machines, and reinstalled into the
+project bucket when needed.
diff --git a/context/project-skills/openspace/validate-skill-routing/.skill_id b/context/project-skills/openspace/validate-skill-routing/.skill_id
new file mode 100644
index 0000000..df71493
--- /dev/null
+++ b/context/project-skills/openspace/validate-skill-routing/.skill_id
@@ -0,0 +1 @@
+validate-skill-routing__imp_6944a359
diff --git a/context/project-skills/openspace/validate-skill-routing/.upload_meta.json b/context/project-skills/openspace/validate-skill-routing/.upload_meta.json
new file mode 100644
index 0000000..ca8769f
--- /dev/null
+++ b/context/project-skills/openspace/validate-skill-routing/.upload_meta.json
@@ -0,0 +1,7 @@
+{
+  "origin": "captured",
+  "parent_skill_ids": [],
+  "change_summary": "Captured a workflow for validating OpenSpace skill routing by checking project/default/global bucket behavior, inspecting installed launcher/config drift, and applying minimal repair steps when router sanity checks degrade.",
+  "created_by": "gpt-5.4",
+  "tags": []
+}
diff --git a/context/project-skills/openspace/validate-skill-routing/SKILL.md b/context/project-skills/openspace/validate-skill-routing/SKILL.md
new file mode 100644
index 0000000..6a4c60d
--- /dev/null
+++ b/context/project-skills/openspace/validate-skill-routing/SKILL.md
@@ -0,0 +1,130 @@
+---
+name: validate-skill-routing
+description: Diagnose and repair OpenSpace project skill routing in this repo, including skill routing checks, router sanity check, bucket fallback validation, 自动路由, 项目 skill 路由, 路由自检, bucket 回退, 检查自动路由是否生效, and wrapper/index/config drift affecting routing.
+---
+
+# Validate Skill Routing
+
+Use this workflow when the task is about verifying or fixing OpenSpace routing behavior for this repository.
+
+Keep the scope narrow:
+
+- validate repo root to project-bucket mapping
+- validate project/default/global candidate bucket order
+- interpret router output
+- distinguish task-wording mismatch from a real routing bug
+- inspect wrapper-derived `OPENSPACE_WORKSPACE`
+- inspect `OPENSPACE_HOST_SKILL_DIRS` coverage
+- detect project skill index drift
+- decide whether stale process state requires restart
+
+Do not broaden into generic MCP health or leak cleanup unless the routing checks point there.
+
+## Workflow
+
+### 1. Confirm repo root and expected bucket
+
+Derive the canonical repo root with git, not only the current working directory.
+
+Use the repo basename as the expected bucket name.
+
+For this repository, healthy routing means the bucket resolves to `openspace`.
+
+### 2. Inspect the active routing path
+
+Inspect the active installed launcher and config path, not only repo templates.
+
+Check:
+
+- `~/.codex/bin/openspace-global-mcp`
+- `~/.codex/AGENTS.md`
+- `~/.codex/config.toml`
+
+Validate:
+
+- launcher canonicalizes workspace to git root when possible
+- launcher exports `OPENSPACE_WORKSPACE` from that canonical root
+- launcher includes the project skill dir before generic global skills
+- the configured MCP entry still points at the launcher you expect
+
+Do not conclude repo-tracked scripts are broken until the installed launchers under `~/.codex` have been checked.
+
+### 3. Run the router with three probes
+
+Use `/Users/admin/.codex/tools/route_codex_skills_via_openspace.py --cwd <cwd> --task "<task>" --json`.
+
+Run these exact probes:
+
+1. Project-bucket probe
+   - `修复 provider rollout 验证流程，确认本地 wrapper 和 repo tracked 配置一致`
+   - expected: an `openspace` project skill hit
+
+2. Default-bucket probe
+   - `重启 MCP 之后做一次健康检查，确认进程祖先和 live tool probe 都恢复正常`
+   - expected: a `default` bucket hit
+
+3. Routing-self-check probe
+   - `检查 openspace 项目 skill 路由、router sanity check、验证 project/default/global bucket fallback`
+   - expected: `validate-skill-routing` hit after this skill exists
+
+If useful, rerun from a nested subdirectory inside the repo to confirm the bucket still resolves to `openspace`.
+
+### 4. Compare router outputs
+
+Compare these fields across probes:
+
+- `bucket`
+- `candidate_dirs`
+- `prefilter_ranked`
+- `selected`
+- `selection_record`
+
+Treat the route as healthy only when:
+
+- the project probe hits the intended `openspace` project bucket
+- the default probe still hits the intended `default` bucket
+- the routing-self-check probe no longer falls back generically
+
+### 5. Localize the failure in this order
+
+If routing is wrong, diagnose in this order:
+
+1. task wording or frontmatter mismatch
+2. missing or weak project skill description
+3. stale `SKILL_INDEX.md`
+4. wrapper or config drift
+5. stale long-lived process state
+
+A global fallback on a vague meta-task is not automatically a bug.
+
+Prefer diagnosing wording and metadata mismatch before blaming the router.
+
+### 6. Apply the minimal repair sequence
+
+Apply the smallest fix that explains the failure:
+
+1. strengthen or add project-skill trigger terms in frontmatter description
+2. regenerate the openspace project skill index
+3. reinstall global wrappers only if launcher behavior changed
+4. restart Codex or daemons only if env or wrapper changes require a fresh process state
+
+Do not restart long-lived processes just because a probe was vague and matched a global skill.
+
+### 7. Re-run and record before/after
+
+Re-run the same probes after any repair.
+
+Record:
+
+- the probe strings used
+- selected bucket before and after
+- whether the installed launcher/config changed
+- whether index regeneration was required
+- whether restart was required
+
+## Guardrails
+
+- Validate actual installed launchers before diagnosing repo-tracked launcher bugs.
+- Treat vague routing-self-check prompts as ambiguous until probe evidence shows a real regression.
+- Prefer precise trigger wording when validating a project skill.
+- Keep default-bucket behavior intact while improving project-skill coverage.