A feed research brief agent for Claude Code, Codex and Cursor
Research Brief writes a short brief of the newest stories in RSS and Atom feeds, each point backed by a quote checked word for word against its article. It works from the context ContextStream picks for the job.
Purpose-built for this one job and tested against a quality bar, it gets only the access the job needs, and a workspace owner or admin turns it on.
Or run it in the ContextStream web app, signed in with your ContextStream account. Version 1.0.0 · Updated
What does Research Brief do?
Reads RSS and Atom feeds and their articles, then writes a short brief of the main stories in which every point carries a quote checked against the article.
What it declines
It does one job. A task like one of these is declined with a reason:
Facts from outside the fetched articles, opinions, or predictions.
Researching a question across the web: it reads only the feeds given and their articles.
Web pages that are not feeds, feeds behind a login, or private networks.
Posting, emailing or publishing the brief; a schedule delivers it.
What can Research Brief access, and is it safe to turn on?
Research Brief gets a short-lived token for one workspace that lets it do only this: call models through ContextStream's model proxy, within a cap set for the run; and read the lessons, preferences and decisions ContextStream picks for its job. It works in one project, only for that run, and declines any task outside its job. Research Brief is read-only: it changes nothing in your workspace. It runs code only in its own isolated sandbox, which holds no credentials and reaches only the hosts named in your input. A workspace owner or admin turns it on.
Model agent with a sandbox. A program on ContextStream's model proxy that also drives its own sandbox. The sandbox holds no credentials. Its sandbox reaches only the hosts named in your input. It is stopped after 6 min. You may schedule it to run on its own.
Scope
What it lets a run do
model.invoke
Call models through ContextStream's model proxy, within a cap set for the run
context.read
Read the lessons, preferences and decisions ContextStream picks for its job
How is Research Brief tested?
Research Brief is checked on 7 cases, each with a known right answer: fixture feeds and articles, and requests outside its job it must decline. A case passes only if the run ends the way the case expects, declining what is not its job, and every check holds: every quote is found in its story's article or feed, every point is backed by a quote, every story comes from a feed, and nothing outside the allowed hosts is asked for. The feeds, the articles and the model's replies are fixed test data, run in a sandbox with no network, so the cases check how the brief is held to its sources, not the live model's own writing. Version 1.0.0 passes all 7.
It is a program on general-purpose models, reached through ContextStream's model proxy, not a trained model. Its prompts are versioned and tuned against these tests, and a change that misses the quality bar fails them.
What does a real run of Research Brief look like?
The input, the result (shortened where it is long), how long it took and where it ran. Live runs used public or demo material only; a run marked as from its test suite is not a live run.
Live run on ContextStream
A brief of the Rust blog and This Week in Rust
Version
1.0.0
Outcome
succeeded
Ran
2026-10-27 21:52 UTC
Took
41.4 s
ContextStream's Demos workspace, on two public feeds. The agent ran in an E2B sandbox, started for this run and removed after it.
5 stories from 2 sources, 4 summarised in 12 points, each backed by a quote from its story; 1 point dropped for quotes the sources do not hold; 1 told in their own words.
Announcing Rust 1.99.0 (Rust blog)
The Rust team released Rust 1.99.0, which stabilizes defining C-ABI variadic functions for the "C" and "C-unwind" ABIs. Quote: “Rust 1.99.0 stabilizes defining C-ABI variadic functions with "C" and "C-unwind" ABIs.”
The release also stabilizes naked variadic functions with non-"C" ABIs, which must be written using inline assembly. Quote: “This release also stabilizes support for defining naked variadic functions with non-"C" ABIs, which must be written via inline assembly.”
Documentation for Box::leak now advises against later deallocating that memory, though the release has no language semantic changes. Quote: “we have updated the documentation on Box::leak to recommend against patterns that later deallocate that memory.”
Announcing Rust 1.99.1 (Rust blog)
Rust 1.99.1 fixes regressions from prior releases, including missing wcslen link errors by adding a fallback in compiler-builtins. Quote: “Rust 1.99.1 fixes the following regressions from prior releases:”
It fixes an x86_64 FFI bool return miscompilation where bits 1-7 of the return register could be unspecified. Quote: “Rust functions returning bool through FFI could leave bits 1-7 of the return register unspecified”
It also fixes rust-lld failing to load libLLVM on macOS when the compiler was invoked without the rustup proxy. Quote: “rust-lld failed to load libLLVM on macOS”
Demoting i686 Windows targets to std-only (Rust blog)
With Rust 1.100.0, 32-bit Windows targets will still receive standard library builds but will no longer have host tools such as the compiler. Quote: “Builds of the standard library will continue to be distributed, but host tools such as the compiler will be no longer available.”
Building 32-bit Windows binaries will require cross-compiling from a supported host, and after Rust 1.100 toolchains cannot be installed on 32-bit Windows hosts. Quote: “To build 32-bit Windows binaries, cross-compiling from a still-supported host toolchain (such as a 64-bit Windows ones) will be required from now on.”
The i686-pc-windows-msvc target remains Tier 1 and CI-tested, while other 32-bit platforms are unaffected. Quote: “i686-pc-windows-msvc as a Tier 1 target still undergoes CI testing.”
brief.md
# Research brief
## [Announcing Rust 1.99.0](https://blog.rust-lang.org/2026/10/01/Rust-1.99.0/)
Rust blog, 2026-10-01
- The Rust team released Rust 1.99.0, which stabilizes defining C-ABI variadic functions for the "C" and "C-unwind" ABIs.
> "Rust 1.99.0 stabilizes defining C-ABI variadic functions with "C" and "C-unwind" ABIs."
- The release also stabilizes naked variadic functions with non-"C" ABIs, which must be written using inline assembly.
> "This release also stabilizes support for defining naked variadic functions with non-"C" ABIs, which must be written via inline assembly."
- Documentation for Box::leak now advises against later deallocating that memory, though the release has no language semantic changes.
> "we have updated the documentation on Box::leak to recommend against patterns that later deallocate that memory."
> "there are no changes to the language semantics in Rust 1.99"
## [Announcing Rust 1.99.1](https://blog.rust-lang.org/2026/10/15/Rust-1.99.1/)
Rust blog, 2026-10-15
- Rust 1.99.1 fixes regressions from prior releases, including missing wcslen link errors by adding a fallback in compiler-builtins.
> "Rust 1.99.1 fixes the following regressions from prior releases:"
> "that function is now present in Rust's compiler-builtins library as a fallback."
- It fixes an x86_64 FFI bool return miscompilation where bits 1-7 of the return register could be unspecified.
> "Rust functions returning bool through FFI could leave bits 1-7 of the
…
Checks: 4 of 5 passed
fetch blog.rust-lang.org: 10 items in the feed, 5 kept.
fetch this-week-in-rust.org: 4 items in the feed, 4 kept.
articles: 9 read, 0 failed, 0 on hosts not allowed, 0 not tried.
quotes: 1 point dropped: no quote of it is in the story's text, or it carried a link or an address.
workspace_context: 0 context items from ContextStream for this job, about 120 tokens.
2 more stories are left out here.
Task
795658b5-24a0-4d09-a9a2-9d45fbf01f2b
From its test suite, not a live run
A brief from two fixture feeds (test case)
Version
1.0.0
Outcome
succeeded
Ran
2026-10-27 17:17 UTC
Took
Not timed (test data)
The agent's offline test suite: its real program in an offline sandbox, with fixture feeds and a scripted model. Not a live run.
4 stories from 2 sources, 3 summarised in 4 points, each backed by a quote from its story; 3 points dropped for quotes the sources do not hold; 1 told in their own words.
City council approves the new harbor tram line (Harbor City News)
The council approved a tram line between the harbor and the central station by nine votes to two. Quote: “voted 9 to 2 on Tuesday night to approve a tram line linking the harbor with the central station”
Building is due to start next spring and take about three years. Quote: “Construction of the harbor tram line is expected to start next spring and take about three years”
Why the old pier is sinking (Field Notes)
The old pier has sunk by almost 20 cm since a storm two winters ago. Quote: “the old pier has sunk by almost twenty centimetres since the storm two winters ago”
Measuring tides with a cheap pressure sensor (Field Notes)
A cheap pressure sensor matched the official tide tables to within a few centimetres. Quote: “the readings matched the official tide tables to within a few centimetres”
brief.md
# Research brief
## [City council approves the new harbor tram line](https://news.example.com/articles/harbor-tram?utm_source=rss)
Harbor City News, 2099-10-14
- The council approved a tram line between the harbor and the central station by nine votes to two.
> "voted 9 to 2 on Tuesday night to approve a tram line linking the harbor with the central station"
- Building is due to start next spring and take about three years.
> "Construction of the harbor tram line is expected to start next spring and take about three years"
## [Why the old pier is sinking](https://www.othersite.net/pier-sinking)
Field Notes, 2099-10-12
- The old pier has sunk by almost 20 cm since a storm two winters ago.
> "the old pier has sunk by almost twenty centimetres since the storm two winters ago"
## [Measuring tides with a cheap pressure sensor](https://blog.example.org/posts/tide-sensor)
Field Notes, 2099-10-13
- A cheap pressure sensor matched the official tide tables to within a few centimetres.
> "the readings matched the official tide tables to within a few centimetres"
## [Ferry timetable changes for winter](https://news.example.com/articles/ferry-winter)
Harbor City News, 2099-10-10
> "Winter sailings between the harbor and Gull Island will begin on the first Monday of December, with two fewer crossings each day."
> "The ferry operator said the change follows a fall in winter passengers over the past three years, especially on the early morning crossing."
Checks: 3 of 5 passed
fetch news.example.com: 6 items in the feed, 5 kept.
fetch blog.example.org: 5 items in the feed, 4 kept.
articles: 5 read, 2 failed, 2 on hosts not allowed, 0 not tried.
quotes: 3 points dropped: no quote of it is in the story's text, or it carried a link or an address.
workspace_context: 0 context items from ContextStream for this job, about 0 tokens.
1 more stories are left out here.
Test case
brief-two-feeds, in its test suite
Depot run
0zb9r381xr
Which tools does Research Brief work with?
Claude Code, Codex, Cursor and other MCP clients: ContextStream's MCP gives them an agents tool that finds, runs and follows Research Brief. It also runs from a terminal or an agent's shell with csa run research-brief, and from the ContextStream web app.
Claude Code, Codex, Cursor and other MCP clients: connect ContextStream's MCP once (setup for each editor), then ask your agent to hand the job to Research Brief.
A terminal or an agent's shell:csa run research-brief, installed with ContextStream's command line.
The web app: the workspace's agent library.
How do I turn Research Brief on?
An owner or admin of your ContextStream workspace turns Research Brief on in the web app's agent library. Agents are opening workspace by workspace; if yours is not open yet, the library says so.
Sign in to ContextStream: the button below opens the agent library on Research Brief.
An owner or admin switches Research Brief on for the workspace.
Pick the project it works in, and run it from the library, from your coding agent or with csa run.
How do I call Research Brief from Claude Code or Codex?
Ask your coding agent to hand the job to Research Brief, or call it yourself: csa run research-brief in a terminal, or agents(action="run", agent="research-brief", input={...}) over MCP.
Connect ContextStream's MCP (it also writes each editor's connection):
In a terminal (curl -fsSL https://contextstream.io/cs/install.sh | sh installs csa):
csa run research-brief sources:='[{"url":"https://blog.rust-lang.org/feed.xml"},{"url":"https://this-week-in-rust.org/rss.xml"}]' max_stories:=5
Input
Field
Type
What it is
sources
object[], required
per_source
integer, required
Items to return per source, newest first.
articles
boolean, required
Read each item's article to quote it. Off, the quotes come from the feed's own text.
quotes_per_item
integer, required
Sentences quoted from each item, as they read in the article.
also_allow
string[], required
Hosts other than the feeds' own whose articles may be read, for feeds that link to other sites.
max_stories
integer, required
The most stories in the brief.
max_age_days
integer
Drop dated items older than this.
focus
string
What the brief is for, for example "harbor transport": stories about it come first.
What does Research Brief cost?
Research Brief costs 1,500 ContextStream credits a run, charged only when the run succeeds; a declined, failed, cancelled or timed-out run costs nothing. Credits come with every plan: Starter is free with 10,000 a month (6 runs of Research Brief, if spent on nothing else), and Pro is $19 a month with 25,000.
A run starts only once its price is accepted. Over MCP, a call without accepted_price_credits returns the price first; csa asks on a terminal, and scripts pass --yes.
Mail reads the person's recent inbox, lists what needs a reply, the commitments made by them or to them and the deadlines in it, and saves reply drafts in their Drafts for them to send.
Checked on 13 test cases with known answers: all 13 passv0.1.0