- 
- 
+
+
+
@@ -84,6 +80,29 @@ hide:
-
+
+ 
+ 
+
-
- Research
-
- The leak is coming from inside the sandbox
-A simple experiment shows why network-layer redaction isn’t enough.
-
- agent-security
- runtime-and-harness
+
+ Announcements
+ Introducing OpenShell Research
+Hello, world—from the OpenShell Research team.
@@ -84,6 +80,29 @@ hide:
Recent notes
+
+
+
@@ -123,11 +137,6 @@ hide:
@@ -150,11 +159,6 @@ hide:
@@ -177,11 +181,6 @@ hide:
diff --git a/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md b/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md
index eb9f1ff0..1e003ccd 100644
--- a/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md
+++ b/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md
@@ -12,17 +12,13 @@ categories:
- Examples
tags:
- reachy-mini
- - openshell
- edge-ai
- sandbox
- robotics
- policy
authors:
- kirit93
-card_tags:
- - edge-ai
- - reachy-mini
- - policy
+
---
diff --git a/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md b/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md
index db0f4fa7..39f6a00a 100644
--- a/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md
+++ b/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md
@@ -12,18 +12,13 @@ social_image: "assets/robotics-policy-prover/robotics-policy-prover-hero.png"
categories:
- Research
tags:
- - openshell
- robotics
- physical-ai
- formal-methods
- policy
- - agents
authors:
- zredlined
-card_tags:
- - physical-ai
- - formal-methods
- - policy
+
---
diff --git a/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md b/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md
index de2f08c4..b31f292a 100644
--- a/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md
+++ b/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md
@@ -10,17 +10,12 @@ social_image: "assets/long-horizon-agent-evals/back-and-forth-by-session.png"
categories:
- Research
tags:
- - openshell
- - agents
- policy
- - security
+ - agent-security
- long-horizon
authors:
- zredlined
-card_tags:
- - agents
- - policy
- - security
+
---
diff --git a/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md b/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md
index 148af98a..d0be337a 100644
--- a/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md
+++ b/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md
@@ -11,17 +11,12 @@ social_image: "assets/agent-policy-prover/hero-concept.png"
categories:
- Research
tags:
- - openshell
- formal-methods
- z3
- - agents
- - security
+ - agent-security
authors:
- zredlined
-card_tags:
- - formal-methods
- - z3
- - agents
+
---
diff --git a/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md b/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md
index 38ccbdb9..5fd6eab3 100644
--- a/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md
+++ b/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md
@@ -13,15 +13,13 @@ hero_image_dark: "../../assets/pi-admission/hero-dark.png"
categories:
- Research
tags:
- - openshell
- agent-security
+ - runtime-and-harness
- redaction
- pi
authors:
- johnnygreco
-card_tags:
- - agent-security
- - runtime-and-harness
+
---
diff --git a/docs/dev-notes/posts/2026-09-28-introducing-openshell-research.md b/docs/dev-notes/posts/2026-09-28-introducing-openshell-research.md
new file mode 100644
index 00000000..1dbad6d2
--- /dev/null
+++ b/docs/dev-notes/posts/2026-09-28-introducing-openshell-research.md
@@ -0,0 +1,73 @@
+---
+title: "Introducing OpenShell Research"
+date: 2026-09-28
+description: "Hello, world—from the OpenShell Research team."
+subtitle: "Hello, world—from the OpenShell Research team."
+author: "Johnny Greco"
+agent_markdown: true
+social_image: "assets/openshell-research/hero.png"
+hero_image: "../../assets/openshell-research/hero.png"
+hero_image_dark: "../../assets/openshell-research/hero-dark.png"
+hero_image_alt: "The OpenShell logo and Research label surrounded by green network nodes and translucent boundaries."
+authors:
+ - johnnygreco
+categories:
+ - Announcements
+tags:
+ - agent-security
+---
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ Research
+
+ The leak is coming from inside the sandbox
+A simple experiment shows why network-layer redaction isn’t enough.
+
+
+
+ Johnny Greco
+
+ Read note
+
+
+ Johnny Greco
+
+ Read note
+
@@ -96,11 +115,6 @@ hide:
What we have learned applying formal methods to control AI agents
An intro to using formal methods to reason about permission changes in long-running AI agents.
-
- formal-methods
- z3
- agents
-
@@ -123,11 +137,6 @@ hide:
Adversarial prototype: AI policy auto-approval for long-horizon agents in OpenShell
Can an attacker agent convince an equally capable reviewer agent to grant a capability that is explicitly prohibited?
-
- agents
- policy
- security
-
@@ -150,11 +159,6 @@ hide:
Can Formal Methods Govern AI-Generated Robot Actions?
A robotics experiment asks whether an independent, SMT-backed policy boundary can efficiently govern AI-generated plans before they reach a simulated or physical robot.
-
- physical-ai
- formal-methods
- policy
-
@@ -177,11 +181,6 @@ hide:
Bringing Privacy and Security to the Edge with OpenShell
Edge agents handle sensitive data and make decisions with physical consequences. Reachy Mini shows why privacy and safety controls must be deterministic and local.
-
- edge-ai
- reachy-mini
- policy
-
diff --git a/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md b/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md
index eb9f1ff0..1e003ccd 100644
--- a/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md
+++ b/docs/dev-notes/posts/2026-07-20-policy-controlling-reachy-mini-with-openshell.md
@@ -12,17 +12,13 @@ categories:
- Examples
tags:
- reachy-mini
- - openshell
- edge-ai
- sandbox
- robotics
- policy
authors:
- kirit93
-card_tags:
- - edge-ai
- - reachy-mini
- - policy
+
---
diff --git a/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md b/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md
index db0f4fa7..39f6a00a 100644
--- a/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md
+++ b/docs/dev-notes/posts/2026-08-07-formal-methods-ai-generated-robot-actions.md
@@ -12,18 +12,13 @@ social_image: "assets/robotics-policy-prover/robotics-policy-prover-hero.png"
categories:
- Research
tags:
- - openshell
- robotics
- physical-ai
- formal-methods
- policy
- - agents
authors:
- zredlined
-card_tags:
- - physical-ai
- - formal-methods
- - policy
+
---
diff --git a/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md b/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md
index de2f08c4..b31f292a 100644
--- a/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md
+++ b/docs/dev-notes/posts/2026-08-27-adversarial-policy-review-long-horizon-agents.md
@@ -10,17 +10,12 @@ social_image: "assets/long-horizon-agent-evals/back-and-forth-by-session.png"
categories:
- Research
tags:
- - openshell
- - agents
- policy
- - security
+ - agent-security
- long-horizon
authors:
- zredlined
-card_tags:
- - agents
- - policy
- - security
+
---
diff --git a/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md b/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md
index 148af98a..d0be337a 100644
--- a/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md
+++ b/docs/dev-notes/posts/2026-09-10-learning-formal-methods-agent-policy-prover.md
@@ -11,17 +11,12 @@ social_image: "assets/agent-policy-prover/hero-concept.png"
categories:
- Research
tags:
- - openshell
- formal-methods
- z3
- - agents
- - security
+ - agent-security
authors:
- zredlined
-card_tags:
- - formal-methods
- - z3
- - agents
+
---
diff --git a/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md b/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md
index 38ccbdb9..5fd6eab3 100644
--- a/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md
+++ b/docs/dev-notes/posts/2026-09-26-network-redaction-is-not-enough.md
@@ -13,15 +13,13 @@ hero_image_dark: "../../assets/pi-admission/hero-dark.png"
categories:
- Research
tags:
- - openshell
- agent-security
+ - runtime-and-harness
- redaction
- pi
authors:
- johnnygreco
-card_tags:
- - agent-security
- - runtime-and-harness
+
---
diff --git a/docs/dev-notes/posts/2026-09-28-introducing-openshell-research.md b/docs/dev-notes/posts/2026-09-28-introducing-openshell-research.md
new file mode 100644
index 00000000..1dbad6d2
--- /dev/null
+++ b/docs/dev-notes/posts/2026-09-28-introducing-openshell-research.md
@@ -0,0 +1,73 @@
+---
+title: "Introducing OpenShell Research"
+date: 2026-09-28
+description: "Hello, world—from the OpenShell Research team."
+subtitle: "Hello, world—from the OpenShell Research team."
+author: "Johnny Greco"
+agent_markdown: true
+social_image: "assets/openshell-research/hero.png"
+hero_image: "../../assets/openshell-research/hero.png"
+hero_image_dark: "../../assets/openshell-research/hero-dark.png"
+hero_image_alt: "The OpenShell logo and Research label surrounded by green network nodes and translucent boundaries."
+authors:
+ - johnnygreco
+categories:
+ - Announcements
+tags:
+ - agent-security
+---
+
+
+
+
+
+
+
+# Introducing OpenShell Research
+
+
+ 
+ 
+
+
+
+
+
+
+Hello, world—from the OpenShell Research team.
+ +
+
+
+
+
+
+
+
+
+
+ + Dev Note + + Announcements +
+>_ Hello, world!
+ +Today we announced the [NVIDIA Open Agent Safety Platform](https://developer.nvidia.com/blog/nvidia-open-agent-safety-platform-a-reference-for-continuous-in-silicon-agent-monitoring), which runs on the open-source OpenShell platform. This feels like a great opportunity to introduce our team — **OpenShell Research**. We’re a cross-functional team of researchers, engineers, and product managers working together to shape OpenShell’s roadmap and advance the safe and secure use of autonomous AI agents in the enterprise and beyond. + +We believe AI agents have enormous potential to deliver value and help people, but realizing this potential requires understanding and limiting the harm they can cause. And as agents become more capable, the risks of deploying them evolve, which is why we started this team — to ensure OpenShell keeps pace with the demands of safe agent deployment. Our initial research efforts will span harness integrations, long-running and multi-agent systems, adversarial testing, policy verification using formal methods, and applications in robotics. We’ll build prototypes, run benchmarks and large-scale experiments, and turn what we learn into dependable OpenShell capabilities — we’re just getting started! + +We also believe that open research is essential to making AI safe. This is why we are committed to working with the open-source community and building in public. [Dev Notes](../index.md) is where we’ll share our latest research, announcements, and the implementation lessons we learn along the way. When it makes sense to share (and maintain) code, we’ll put it in our [OpenShell Research repo](https://github.com/NVIDIA/OpenShell-Research). All code used in our research is always available upon request. + +
+
+## Join us
+
+
+
+We’re hiring! We are actively looking for researchers and engineers who love to build and are passionate about pushing the boundaries of what’s possible through open research. If this sounds like you, please reach out and consider [applying to join our team](https://nvidia.wd5.myworkdayjobs.com/NVIDIAExternalCareerSite/job/US-Remote/Software-Engineer--OpenShell_JR2020825).
diff --git a/docs/development/index.md b/docs/development/index.md
index 19d4b475..b87e29f5 100644
--- a/docs/development/index.md
+++ b/docs/development/index.md
@@ -65,7 +65,10 @@ The renderer validates this list before updating any generated content. Missing,
unknown, or multiple categories fail the build with an author-facing error.
Categories appear consistently on index cards and article bylines. Categories
remain available even when they have no posts; do not create placeholder notes.
-Use tags for subject areas such as robotics, formal methods, or agent security.
+Use one `tags` list for specific topics such as `robotics`, `formal-methods`, or
+`agent-security`. Avoid generic `openshell` and `agents` tags; use
+`agent-security` consistently instead of `security`. Tags feed site search
+metadata; articles and index cards do not display tag badges.
Readers can combine category and author filters on the Dev Notes index. The
renderer derives the author menu from every post's author IDs and always includes
@@ -76,7 +79,6 @@ Back/Forward navigation. Article byline names and categories link to these
filtered views. Empty combinations show a message and a clear-filters action.
Keep the full index readable when JavaScript is unavailable.
-The renderer uses `card_tags` as the card's tags, falling back to `tags`.
An optional `card_variant` must have matching card and
artwork CSS modifiers in `docs/stylesheets/dev-notes.css`. Set `hero_image` to
an image path relative to the post, and describe it with `hero_image_alt`. The
diff --git a/docs/stylesheets/dev-notes.css b/docs/stylesheets/dev-notes.css
index c07dd0ec..69b4724d 100644
--- a/docs/stylesheets/dev-notes.css
+++ b/docs/stylesheets/dev-notes.css
@@ -760,6 +760,11 @@ body:has(.openshell-home-page) .md-path {
letter-spacing: -0.035em;
}
+.dev-notes-featured .journal-section__head {
+ padding-left: 0.6rem;
+ border-left: 3px solid var(--openshell-green);
+}
+
.dev-notes-page [hidden] {
display: none !important;
}
@@ -848,9 +853,9 @@ body:has(.openshell-home-page) .md-path {
.dev-note-card--featured .dev-note-card__link {
display: grid;
- grid-template-columns: repeat(2, minmax(0, 1fr));
+ grid-template-columns: minmax(0, 1.2fr) minmax(0, 1fr);
align-items: center;
- gap: 1.5rem;
+ gap: 2rem;
}
.dev-note-card__visual {
@@ -998,7 +1003,7 @@ body:has(.openshell-home-page) .md-path {
}
.md-typeset .dev-note-card--featured h3 {
- font-size: clamp(1.85rem, 3.5vw, 2.5rem);
+ font-size: clamp(2.1rem, 4.1vw, 2.95rem);
line-height: 1.08;
text-wrap: balance;
}
@@ -1012,25 +1017,7 @@ body:has(.openshell-home-page) .md-path {
}
.md-typeset .dev-note-card--featured .dev-note-card__summary {
- font-size: 1rem;
-}
-
-.dev-note-card__tags {
- display: flex;
- flex-wrap: wrap;
- gap: 0.28rem 0.7rem;
- margin-top: 0.6rem;
- color: var(--openshell-accent);
- font-family: var(--openshell-mono);
- font-size: 0.54rem;
- letter-spacing: 0.045em;
- line-height: 1.4;
- text-transform: uppercase;
-}
-
-.dev-note-card__tags span::before {
- margin-right: 0.12rem;
- content: "#";
+ font-size: 1.08rem;
}
.dev-note-card__footer {
@@ -1103,7 +1090,7 @@ body:has(.openshell-home-page) .md-path {
}
.dev-notes-recent {
- margin-top: 1rem;
+ margin-top: 1.75rem;
padding-top: 0.5rem;
border-top: 1px solid var(--openshell-rule);
}
@@ -1144,7 +1131,7 @@ body:has(.openshell-home-page) .md-path {
.dev-note-card--recent .dev-note-card__copy {
display: grid;
grid-template-columns: minmax(0, 1fr);
- grid-template-areas: "meta" "title" "summary" "tags" "footer";
+ grid-template-areas: "meta" "title" "summary" "footer";
gap: 0.35rem;
min-width: 0;
}
@@ -1165,12 +1152,6 @@ body:has(.openshell-home-page) .md-path {
font-size: 0.78rem;
}
-.dev-note-card--recent .dev-note-card__tags {
- grid-area: tags;
- align-self: center;
- margin-top: 0;
-}
-
.dev-note-card--recent .dev-note-card__footer {
grid-area: footer;
margin-top: 0;
@@ -1330,6 +1311,32 @@ body:has(.dev-note-header) .md-main__inner {
line-height: 1.5;
}
+.md-typeset .dev-note-section-heading--compact > h2 {
+ font-size: 1.25rem;
+}
+
+.md-typeset .dev-note-greeting {
+ font-family: var(--openshell-mono);
+ font-size: clamp(1.25rem, 3vw, 1.8rem);
+ font-variant-ligatures: none;
+}
+
+.dev-note-greeting__prompt {
+ color: var(--openshell-green);
+}
+
+.dev-note-greeting > span {
+ display: inline-block;
+ white-space: pre;
+ /* Reveal the 16 characters once, keeping the full text and layout intact. */
+ animation: dev-note-type-in 1.3s steps(16, end) 0.2s backwards;
+}
+
+@keyframes dev-note-type-in {
+ from { clip-path: inset(0 100% 0 0); }
+ to { clip-path: inset(0 0 0 0); }
+}
+
.md-typeset .dev-note-equation {
margin: 1.4rem 0 1.7rem;
padding: 1.1rem 1.35rem;
@@ -1480,7 +1487,7 @@ body:has(.dev-note-header) .md-main__inner {
}
.dev-note-card--featured .dev-note-card__visual {
- width: min(100%, 21rem);
+ width: min(100%, 26rem);
justify-self: center;
}
@@ -1490,7 +1497,6 @@ body:has(.dev-note-header) .md-main__inner {
"visual meta"
"visual title"
"summary summary"
- "tags tags"
"footer footer";
gap: 0.4rem 0.75rem;
}
@@ -1591,6 +1597,10 @@ body:has(.dev-note-header) .md-main__inner {
scroll-behavior: auto;
}
+ .dev-note-greeting > span {
+ animation: none;
+ }
+
.openshell-drawer-button,
.md-overlay,
.md-sidebar--primary,
@@ -1603,6 +1613,17 @@ body:has(.dev-note-header) .md-main__inner {
}
}
+@media print {
+ .dev-note-greeting > span {
+ animation: none;
+ }
+}
+
+/* Keep topic metadata searchable without displaying non-interactive badges. */
+body:has(.dev-note-header) .md-typeset .md-tags {
+ display: none;
+}
+
/* Dev note figures and theme-specific images */
body:not([data-md-color-scheme="slate"]) .dev-note-image--dark,
body[data-md-color-scheme="slate"] .dev-note-image--light {
diff --git a/scripts/render-dev-notes.py b/scripts/render-dev-notes.py
index ec7e31f4..f1773936 100644
--- a/scripts/render-dev-notes.py
+++ b/scripts/render-dev-notes.py
@@ -272,22 +272,14 @@ def render_card_copy(post: dict[str, Any]) -> str:
published = post["published"]
description = require_string(metadata, "description", path)
category = require_category(metadata, path)
- tags = require_list(metadata, "card_tags", path) or require_list(metadata, "tags", path)
authors = post["authors"]
- tags_html = ""
- if tags:
- tag_items = "\n".join(f" {html.escape(tag)}" for tag in tags)
- tags_html = f"""
-
-{tag_items}
-
"""
return f"""
{html.escape(category)}
{html.escape(title)}
-{html.escape(description)}
{tags_html} +{html.escape(description)}
{render_card_authors(authors)}
Read note
diff --git a/tests/test_render_dev_notes.py b/tests/test_render_dev_notes.py
index cc15dc2c..228a1a94 100644
--- a/tests/test_render_dev_notes.py
+++ b/tests/test_render_dev_notes.py
@@ -27,7 +27,6 @@ def make_post(
description: str = "What the experiment taught us.",
categories: list[str] | None = None,
tags: list[str] | None = None,
- card_tags: list[str] | None = None,
variant: str = "",
hero_image: str = "",
) -> dict[str, object]:
@@ -38,10 +37,8 @@ def make_post(
"description": description,
"authors": ["ada"],
"categories": ["Research"] if categories is None else categories,
- "tags": tags or ["agents", "runtime"],
+ "tags": tags or ["agent-security", "runtime"],
}
- if card_tags is not None:
- metadata["card_tags"] = card_tags
if variant:
metadata["card_variant"] = variant
if hero_image:
@@ -118,18 +115,19 @@ def test_visual_stamp_uses_stable_canonical_date(self) -> None:
self.assertIn("2026.06.05", visual)
self.assertNotIn("OSR—", visual)
- def test_card_tags_override_tags_and_variant_reaches_outer_card(self) -> None:
+ def test_cards_preserve_topic_metadata_without_rendering_tags(self) -> None:
post = make_post(
"2026-06-05-note.md",
- tags=["fallback"],
- card_tags=["preferred", "evaluation"],
+ tags=["agent-security", "evaluation"],
variant="launch",
)
- card = renderer.render_featured_card(post)
- self.assertIn("dev-note-card--launch", card)
- self.assertIn(">preferred<", card)
- self.assertIn(">evaluation<", card)
- self.assertNotIn(">fallback<", card)
+ for render_card in (renderer.render_featured_card, renderer.render_recent_card):
+ card = render_card(post)
+ self.assertIn("dev-note-card--launch", card)
+ self.assertNotIn("dev-note-card__tags", card)
+ self.assertNotIn("agent-security", card)
+ self.assertNotIn("evaluation", card)
+ self.assertEqual(post["metadata"]["tags"], ["agent-security", "evaluation"])
def test_hero_image_replaces_generated_visual(self) -> None:
post = make_post(
@@ -162,13 +160,12 @@ def test_metadata_is_html_escaped(self) -> None:
"2026-06-05-note.md",
title="Runtime