feat(editor): work-item descriptions store sanitised HTML only; pages store the validated triple; hardened sanitiser

未关闭
#48 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
32/100
Issue 类型
功能
描述清晰度
描述清楚
活跃度
活跃
技术栈
html, rust, sql, wasm

调研方向

Start by reading the sanitizer in crates/thrust-editor-doc/src/sanitize.rs, its tests in crates/thrust-editor-doc/tests/sanitize.rs, and the migration and contract references named in the issue. Then trace the work item write path and page fallback write path, keeping the blocked-by #44 dependency in view. Done means the acceptance tests pass, the page binary validation rejects invalid Yjs data, the migration stores only the specified work-item fields, and the listed deviations and questions are recorded.

由索引模型根据 Issue 内容生成。

描述

area:app prio:p0 review:fable-2026-10-19

Context

The slice stores a work item's description as unsanitised text (crates/thrust-adapters-postgres/migrations/0001_init.sql:33; crates/thrust-contracts/src/lib.rs:97, :214; crates/thrust-core/src/lib.rs:260). The sanitiser exists only in the spike crate thrust-editor-doc, behind its sanitize feature, and only thrust-editor-worker depends on it. It provides plane_builder(), Plane's nh3 allow-list (crates/thrust-editor-doc/src/sanitize.rs:100-115); hardened_builder(), which adds value filters (:167-177); validate_binary_data (:208-233); and strip_tags (:235-237). Sanitizer::default() and validate_html_content both use the Plane builder (:192, :197-200). ADR-0012 (accepted in round 27) stores client html, json and binary verbatim at P0 (adrs/ui/0012-editor-and-i18n-reuse.md:33), as decided under Q-F5.1 in round 11 (QUESTIONS.md:153).

Problem

  • FR-14, from spike 0006: work-item descriptions carry no binary (v1 returns it null in 64 of 64; research/0008 §2.1, line 42). The v1 serializer excludes description_json (apps/api/plane/api/serializers/issue.py:73), and the web editor initialises from description_html (apps/web/core/components/editor/rich-text/description-input/root.tsx:239). Only the page fallback path builds all three forms (apps/web/core/hooks/use-page-fallback.ts).
  • Storing client JSON next to sanitised HTML therefore creates two sources of truth, one of them unsanitised.
  • FR-23: the hardened config neutralises 125 of 125 executable payloads, where Plane's allow-list neutralises 116, and it leaves all 195 editor documents unchanged. Plane's allow-list changes 66 of the 195 for exactly two reasons: it strips issue-embed-component (5 documents) and rewrites rel (61) (research/0008 §7, lines 217-229).
  • At P1 the Rust converter derives the binary from HTML (§8).

Proposal

  • Work items store description_html (sanitised) and description_stripped (strip_tags of the sanitised HTML). There is no description_json or description_binary column at P0. Compat and v1 writes accept description_json and discard it. At P1, thrust-editor-doc derives JSON and the binary on first collab open.
  • Pages store the triple the fallback path sends. The HTML is sanitised. The binary is base64-decoded and checked with validate_binary_data, as Plane's PageBinaryUpdateSerializer does (apps/api/plane/app/serializers/page.py:180-198), and also checked to decode with yrs. Plane does not do that last check, so record it with the deviations below. Pages are an M3 screen, so this half lands with the pages work.
  • Sanitiser changes:
    • Make the hardened builder the default.
    • Allow issue-embed-component and its five attributes.
    • Preserve rel (turn off ammonia's link_rel and allow rel on a).
    • Accept that ammonia 4.x output is not byte-identical to nh3's 4.0.0.
    • Sanitise in the use-case layer, so compat and v1 share one path, and record the schema version (ADR-0012).
  • Record deviations D-016 to D-020 (handoff A4.4). Build no browser-flavour renderer (Q-0012.4).

Acceptance criteria

  • The migration gives work items description_html and description_stripped, and neither description_json nor description_binary.
  • A contract test sends one payload of each of the nine classes that survive Plane's allow-list (research/0008 §7, line 225) through create and update, and reads every one back neutralised. Control: with plane_builder() as the default, the test fails.
  • allow_list_preservation_of_editor_html (crates/thrust-editor-doc/tests/sanitize.rs:15-64) asserts 195 of 195 parse-equal under the new allow-list instead of only printing counts. Control: under plane_builder() it fails (129 of 195).
  • A page write with a truncated or non-Yjs binary gets a 400, and a valid fallback triple is stored.
  • DEVIATIONS.md has D-016 to D-020, and Q-0012.2, Q-0012.4, Q-F5.1 and Q-0061.4 are answered.

References

  • Review: .agents/plans/contrail/reviews/2026-10-19-fable-architecture-review.md (FR-14, FR-23)
  • research/0008-editor-conversions.md §2.1, §5 rows 7 and 8, §5.3, §7, §8, §11
  • ADR-0012 (:33; round-29 note), ADR-0061 (round-29 note); DEVIATIONS D-012, C-5, D-016 to D-020
  • Q-F5.1, Q-0012.2, Q-0012.4, Q-0061.4
  • crates/thrust-editor-doc/src/sanitize.rs:14, :100-115, :167-237; tests/sanitize.rs:15-64, :225-243
  • Plane oracle: apps/api/plane/api/serializers/issue.py:73; apps/api/plane/app/serializers/page.py:173-198; apps/web/core/components/editor/rich-text/description-input/root.tsx:239
  • Beads: plane-byd.6

Blocked by: #44

Finding: FR-14, FR-23

主要语言
Rust
星标
0
派生
0
平均合并
1 小时 35 分钟
30 天内合并 PR
67

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

aRustyDev/thrust 的其他 Issue

查看 aRustyDev/thrust 的全部 Issue

相似的 Issue

更多 Rust Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。