Tracking issue for PR #886: Agent Skill authoring and validation platform (OSPP task 26d8e0076)

Summary

PR #886 implements the Agent Skill creation and validation platform for OSPP 2026 task 26d8e0076 (面向 SkillHub 的 Agent Skill 创建与验证平台). It closes the gap named in the task brief: users currently author skills by hand — this adds a unified draft → validate → fix → publish loop on top of SkillHub.

What the PR delivers

Draft authoring: draft CRUD with scaffolded SKILL.md, file editor with optimistic concurrency (expectedRevision), resource files
Runtime binding: local-script and OpenAI-compatible agents, tool allowlists; MCP server declarations (http/sse/stdio) with server-side probing during validation
Three-layer validation: structure (SKILL.md rules) / config (binding, MCP reachability) / behavior (script + prompt tasks with assertions)
Fix loop: findings with locations and suggested patches, diff preview, one-click apply (revision bump), re-validation history
Publish integration: validated drafts submit into the existing pipeline (security scan → review → publish)
Security: SSRF policy on OpenAI/MCP endpoints (cloud-metadata always blocked), stdio MCP off by default with docker isolation, envRefs allowlist; docker execution-mode default with startup guard
RISC-V64: repeatable real-board CI on physical VisionFive 2 hardware (task requirement field: 支持架构 RISC-V64)

Verification evidence

Backend: 1085 tests, 0 failures; live smoke 39/39 (scripts/authoring-smoke-test.sh); publish-pipeline E2E 16/16 (scripts/authoring-publish-e2e.sh)
Real-board RISC-V64 CI (free board runners, fork-guarded workflow): authoring flow green 7 consecutive runs; full backend suite on the board BUILD SUCCESS — 1081 tests, 0 failures (green run)
DCO ✅ / CLA ✅, MERGEABLE, rebased on the latest main; design doc docs/26-skill-authoring-platform.md, example skill in examples/skill-authoring/

Scale

19 commits, 150 files changed (+17.9k / −971) — larger than the usual PR here, we split it into 15 reviewable commits by layer.

Ask

The PR has been Ready with all checks green for several days. Could a maintainer take a look and advise on the review path? Happy to split it, — the mentor is cc'd (ylhu16@iflytek.com) in case the OSPP timeline needs it landed sooner.

Issue Triage

Route: triage/core
Priority: priority/p1 (3.7/5)
Effort: effort/m (3/5)
Confidence: 3/5
Risk: high
Analysis Mode: rules-only

Why

Touches core SkillHub workflows (publish, review, auth).
Contains urgency signals (failure, security).
Text suggests a bounded change (docs).

High-Risk Signals

Mentions sensitive areas (auth).

Maintainer Brief

Summary: PR #886 implements the Agent Skill creation and validation platform for OSPP 2026 task 26d8e0076 (面向 SkillHub 的 Agent Skill 创建与验证平台). It closes the gap named in the task brief: users currently author skills by hand — this adds a unified draft → validate → fix → publish loop on top of SkillHub.
Why core:

Touches a higher-risk area where automated fixes should not be trusted without maintainer review.
The issue is important, but a maintainer still needs to tighten scope before implementation.
Mentions sensitive areas (auth).
Reproduction or operator path:
Recreate the operator flow described in the issue and confirm the current failure mode.
Suspected areas:
CLI publish command option parsing and namespace-aware publish flow
Authentication, session, and identity integration
Risks to watch:
Mentions sensitive areas (auth).
Namespace-scoped behavior can regress default publish/install flows if fallback routing is not preserved.
Validation checklist:
Validate the primary reported workflow end-to-end after the fix.
Verify both namespace-scoped behavior and the default non-namespace flow.

Next Action

Hand the issue to a core maintainer and use a local programming agent for reproduction, scoping, and validation.

问题分流结果

路由: triage/core
优先级: priority/p1 (3.7/5)
修复投入: effort/m (3/5)
信息完整度: 3/5
风险: 高
分析模式: rules-only

原因

涉及 SkillHub 核心流程（publish, review, auth）
包含紧急信号（failure, security）
文本显示改动范围较可控（docs）

高风险信号

提到敏感区域（auth）

维护者交接摘要

概要: PR #886 implements the Agent Skill creation and validation platform for OSPP 2026 task 26d8e0076 (面向 SkillHub 的 Agent Skill 创建与验证平台). It closes the gap named in the task brief: users currently author skills by hand — this adds a unified draft → validate → fix → publish loop on top of SkillHub.
为何进入 core:

触及高风险区域，未经 maintainer 审查不应直接信任自动修复
问题本身重要，但仍需要 maintainer 先收敛范围再实施
提到敏感区域（auth）
复现路径或操作路径:
按 issue 中描述的操作路径复现，并确认当前失败模式
怀疑影响区域:
CLI 发布命令参数解析与 namespace 感知发布流程
认证、会话与身份集成
重点风险:
提到敏感区域（auth）
namespace 范围行为如果没有保留 fallback routing，可能回归默认 publish/install 流程
验证清单:
修复后端到端验证主报告流程
同时验证 namespace 范围行为与默认非 namespace 流程

下一步

交给 core maintainer，并结合本地编程Agent协助完成复现、收敛范围与验证闭环

Thanks for the detailed write-up and the RISC-V evidence. Given the size and the security-sensitive surface (endpoint probing, stdio MCP, docker execution), we'd like to review this in layers rather than as one 18k-line PR. Suggested path:

Rebase on main and renumber the migration: main now has V66__promotion_update.sql and V67__promotion_revocation_request.sql, so yours should become V68__authoring_platform.sql (that's the current conflict).
Split into stacked PRs, each green on its own: (a) domain model + migration + draft CRUD + structure-layer validation; (b) runtime binding, MCP probing and the endpoint/stdio/env security policies (we'll do a dedicated security review here); (c) fix loop + publish integration; (d) web UI + e2e; (e) docs and example skill.
Keep riscv64-native.yml in your fork. It can't run upstream since it needs the board runners, and linking the green runs from the docs is enough evidence for the RISC-V64 requirement.

We'll start reviewing as soon as (a) is up. Keep #934 as the tracking issue and link each PR there.

@FenjuFu

Thanks @FenjuFu — that path works for us. Plan of record:

Rebase on main and renumber the migration to V68__authoring_platform.sql (noted the promotion migrations took V66/V67).
Split into the stacked PRs (a)–(e) exactly as suggested, each green on its own; (b) will carry the endpoint/stdio/env security policies for the dedicated security review.
riscv64-native.yml stays fork-only; the docs link the green board runs (7× authoring flow + full suite BUILD SUCCESS) as the RISC-V64 evidence.

We'll link PR (a) here as soon as it's up. Thanks again for taking a look!

Layer (a) is up: #935 — draft domain model + V68 migration + draft CRUD (scaffolded SKILL.md) + structure-layer validation with fix suggestions. 54 files, +4.0k lines; domain 609 tests green, draft-flow integration test (broken frontmatter → fixable finding → repair → SUCCEEDED) green, full suite running on the branch now.

Remaining layers (b)–(e) will stack on top of (a) and link here as they go.

Layer (b) — runtime bindings, MCP probing and the security policies — is stacked and ready for review: zjncs#1 (diff is the (b)-only slice, +5.7k lines on top of #935).

Since the stacked base branch lives on the fork until (a) merges, the PR is opened against the fork branch; it will be retargeted to upstream main as soon as #935 lands. Happy to move things around if you'd prefer a different arrangement.

Layers (c)–(e) coming up the same way.

The full stack is up. Layers and their diffs (each green on its own; the fork PRs will be retargeted to upstream main as #935 lands):

Layer	Scope	PR	Size
(a)	draft domain model + V68 migration + draft CRUD (scaffolded SKILL.md) + structure validation	#935 (upstream)	54 files, +4.0k
(b)	runtime bindings, MCP probing, endpoint/stdio/env security policies	zjncs#1 (fork, stacked)	58 files, +5.7k
(c)	fix loop (apply/dismiss suggested patches) + publish-pipeline submission	zjncs#2 (fork, stacked)	9 files, +0.7k
(d)	web workbench (editor, binding form, live run console, fix preview, submit) + Playwright e2e	zjncs#3 (fork, stacked)	26 files, +6.0k
(e)	design doc, example skill, live verification + RISC-V64 re-verify scripts	zjncs#4 (fork, stacked)	11 files, +1.7k

Verification per layer: (a) full suite 1049 tests 0 failures; (b) security/MCP/adapter unit tests + runtime-flow integration test (full suite re-run in progress, will report); (c) full suite 1105 tests 0 failures incl. the complete authoring flow; (d) typecheck + 922 web unit tests (only the two pre-existing upstream theme tests fail, identically on plain main); (e) docs/scripts only, scripts syntax-checked.

#886 (the monolithic PR) is superseded by this stack — closing it in favor of the layers above.

feat(authoring): draft domain model, CRUD and structure-layer validation