Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

:bug: eventing: unauthenticated PUT /transcript rewrites what an agent resumes from, and correlation ids are listable

未关闭
  • 哪里不对？
#887 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 6 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
42/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
python
领域
api, authentication, backend, security

调研方向

Start in eventing/handlers.py and trace authentication for the transcript and group routes, then read eventrunner/transcript.py, eventrunner/runner.py, and store.py. Review the related paths in ghauth.py, skills/eventbridge/eventbridge-cli.py, main.py, and kafka_out.py, along with tests/test_ghauth.py; run the relevant tests and verify the exposed transcript, correlation-id, login, and signing behaviors are covered.

由索引模型根据 Issue 内容生成。

描述

kind/bug

Summary

Sign-in guards only the two routes that start work. POST /v0/agents and POST /v0/groups call
auth.resolve; every other route is open. Two consequences are worth separating from the general
"only starting work is authenticated" limit, because together they let an unauthenticated caller
rewrite what an agent resumes from.

Found while documenting the controls for rossoctl/rossoctl#2609. References are to eventing/ at
d9677ddd.

1. PUT /v0/agents/{id}/transcript needs no credential, and feeds claude --resume

The transcript is the session checkpoint EventRunner resumes from. On a cold pod a mode=continue
turn downloads it from EventBridge and passes it to claude --resume
(eventrunner/transcript.py:181-196, eventrunner/runner.py:238,256,301).

PUT is unauthenticated, so anyone who can name a conversation can replace the history the agent
continues with. Request signing does not help: it covers the prompt, not the resume state.

Observed against Handlers with sign-in fully configured (client id and approved list):

PUT  without credential -> 200 OK
POST /v0/agents without credential -> 401 Unauthorized
GET  without credential -> 200 OK  b'{"type":"user","message":{"role":"user","content":"forged history"}}\n'

2. Correlation ids are listable without signing in

GET /v0/groups returns the 100 most recent groups (store.py:273, no query parameter raises it —
handlers.py:206), and GET /v0/groups/{id}/status lists every member's correlation id. Both are
open. So "an attacker would have to guess the id" understates the exposure for any conversation that
belongs to a group.

Confirmed: all_groups() returns 100 when 105 groups exist.

Related

  • ghauth._post_form raises HTTPError on any GitHub 4xx (ghauth.py:57), so the friendly message
    at :74-80 is unreachable and device_flow_disabled surfaces as a traceback ending in
    HTTP Error 400: Bad Request. cmd_login catches only RuntimeError
    (skills/eventbridge/eventbridge-cli.py:514-518). tests/test_ghauth.py:265-267 patches
    _post_form, so the test never sees the status. Reproduced by patching urlopen to raise a 400.
  • Startup does not check that EB_SIGNING_KID is present in EB_VERIFY_KEYSET_PATH
    (__main__.py:68 checks only that it is set), so EventBridge can start and then flag its own group
    events. Combined with the fact that group events are signed only when EB_SIGNING_KEY_PATH gives a
    seed (kafka_out.py:102), a half-configured deployment fails closed on genuine traffic.

Impact

All of the above are documented as limits in rossoctl/rossoctl#2609, so the docs are accurate. The
transcript write is the one I would prioritise: it is an unauthenticated write that changes agent
behaviour, rather than only a read or a notification artefact.

主要语言
Python
星标
11
派生
42
平均合并
4 天 11 小时
30 天内合并 PR
40

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。

rossoctl/examples 的其他 Issue

查看 rossoctl/examples 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。