run-llama / llama-agents / 40905277487
85%

DEFAULT BRANCH: main
Ran
Jobs 12
Files 235
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 85.173% (+0.02%) from 85.155%
40905277487

push

github

web-flow 
Require starlette>=1.0.1 in the appserver and server (CVE-2026-48710) (#840)

## What

Raise the starlette floor to `>=1.0.1` in `llama-agents-server` (was
`>=0.39.0`) and add the same direct floor to `llama-agents-appserver`.
Includes a patch changeset for both packages. `uv.lock` changes only in
its requirement metadata; the resolved starlette stays at 1.4.0.

## Why

| Vulnerability | Package | Fixed in | On CISA KEV |
| --- | --- | --- | --- |
| CVE-2026-48710 / GHSA-86qp-5c8j-p5mr: a crafted `Host` header changes
`request.url.path` and can bypass path-based checks | starlette <= 1.0.0
| 1.0.1 | Yes |

The appserver doesn't run from its image's own venv. Bootstrap runs `uv
sync` on the deployment's project, then `uv pip install
llama-agents-appserver==<pinned>` into that project venv. `uv pip
install` keeps any starlette already present if it satisfies the
declared bounds. Today the effective floor comes from
`prometheus-fastapi-instrumentator` 8.x, which requires
`starlette>=1.0.0`. So a project whose lock resolved starlette 1.0.0
keeps running a vulnerable version. Appserver 0.11.x has no starlette
floor at all.

Released images 0.12.0 to 0.12.2 already ship starlette 1.4.0, and
0.11.5 ships 0.46.2. This change makes the floor hold for project venvs
too, not just for the image.

## Exposure

The appserver itself doesn't use `request.url.path` for auth or routing.
It's only used to skip access-log lines for health and UI-proxy requests
(`configure_logging.py`). So on an affected version, the worst
in-process effect is a missing access-log line. User workflow code or
middleware that branches on `request.url.path` would be exposed, and
that's the case this floor guards against.

## Verification

- `uv lock`, then `uv run dev -p llama-agents-server -p
llama-agents-appserver`: 611 passed (123 appserver, 488 server).
- `uv run pre-commit run --files <changed files>`: passed (toml checks,
ty, codespell, toml-sort).
- Replayed the bootstrap install step... (continued)

18962 of 22263 relevant lines covered (85.17%)

1.15 hits per line

Jobs
ID Job ID Ran Files Coverage
1 llama-index-workflows - 40905277487.1 53
94.18
GitHub Action Run
2 llama-agents-core - 40905277487.2 15
81.34
GitHub Action Run
3 llama-agents-dev - 40905277487.3 12
76.5
GitHub Action Run
4 llama-agents-integration-tests - 40905277487.4 102
64.93
GitHub Action Run
5 llama-agents-appserver - 40905277487.5 17
81.96
GitHub Action Run
6 llama-agents-server - 40905277487.6 31
86.15
GitHub Action Run
7 llama-index-utils-workflow - 40905277487.7 1
70.84
GitHub Action Run
8 llama-agents-control-plane - 40905277487.8 31
74.57
GitHub Action Run
9 llamactl - 40905277487.9 48
79.79
GitHub Action Run
10 llama-agents-dbos - 40905277487.10 11
75.72
GitHub Action Run
11 llama-agents-agentcore - 40905277487.11 5
70.22
GitHub Action Run
12 llama-agents-client - 40905277487.12 4
95.32
GitHub Action Run
Source Files on build 40905277487