diff --git a/skills/dd-apm/spec.yaml b/skills/dd-apm/spec.yaml
index e763d59..6aa567b 100644
--- a/skills/dd-apm/spec.yaml
+++ b/skills/dd-apm/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"  # main as of 2026-04-15
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"  # main as of 2026-04-15
   path: "dd-apm"
-  version: "0.2.0"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
@@ -21,3 +21,15 @@ security:
   allowed_issues:
     - rule_id: MANIFEST_MISSING_LICENSE
       reason: "datadog-labs/agent-skills is licensed MIT at the repository root; upstream does not embed an SPDX license identifier in per-skill SKILL.md frontmatter."
+    - rule_id: COMPOUND_EXTRACT_EXECUTE
+      reason: |
+        FP: cisco-ai-skill-scanner matched the documented `pup` CLI install
+        snippet in k8s-ssi/agent-install/SKILL.md:67,
+        k8s-ssi/troubleshoot-ssi/SKILL.md:48 and
+        linux-ssi/troubleshoot-ssi/SKILL.md:48. The snippet downloads the
+        official pup release tarball from the same vendor
+        (github.com/datadog-labs/pup/releases, version pinned via the GitHub
+        releases API) and pipes it through `tar xz` into /usr/local/bin — the
+        canonical, vendor-published CLI install instruction shown to the user,
+        not a hidden/malicious archive payload. No executable threat.
+        datadog-labs/agent-skills @9bcb3ceafacae78dbba76c9459a878fc7d6a0d10.
diff --git a/skills/dd-docs/spec.yaml b/skills/dd-docs/spec.yaml
index 1e1c216..4fbe512 100644
--- a/skills/dd-docs/spec.yaml
+++ b/skills/dd-docs/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
   path: "dd-docs"
-  version: "0.2.0"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
diff --git a/skills/dd-llmo-eval-bootstrap/spec.yaml b/skills/dd-llmo-eval-bootstrap/spec.yaml
index 8811bac..8886e4f 100644
--- a/skills/dd-llmo-eval-bootstrap/spec.yaml
+++ b/skills/dd-llmo-eval-bootstrap/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"  # main as of 2026-04-15
-  path: "dd-llmo/eval-bootstrap"
-  version: "0.2.0"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"  # main as of 2026-04-15
+  path: "dd-llmo/llm-obs-eval-bootstrap"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
@@ -60,3 +60,37 @@ security:
         False positive - matches `run()` on SKILL.md:537 inside example Python
         code (likely `evaluator.run()` or similar). Not multi-agent attack
         traffic. Verified at digest 98343f304cbd4439b3d7640cfe64f78070e44d68.
+    # --- ATR_2026_* rule-pack hits introduced when the eval-bootstrap SKILL.md
+    # grew (~681 -> ~1180 lines) at ref 9bcb3ce, adding the "publish online
+    # LLM-judge evaluators" workflow. All are substring / word-fragment matches
+    # on benign documentation prose and backtick-wrapped upstream tool/CLI/template
+    # names in a skill whose explicit purpose is bootstrapping LLM evaluators.
+    # No executable threat. datadog-labs/agent-skills @9bcb3ceafacae78dbba76c9459a878fc7d6a0d10.
+    - rule_id: ATR_2026_00001
+      reason: "FP: matched prose 'these become the domain-specific evaluator category' (SKILL.md:371). Documentation guidance, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00004
+      reason: "FP: matched JSON/code example fragment `{role: \"` in an eval message schema. Documentation/code, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00010
+      reason: "FP: matched backtick-wrapped upstream tool/CLI names — `get_llmobs_evaluator`, `list_llmobs_eval*`, `create_or_update_llmobs_eval*`, `pup llm-obs evals get-eval`, `/eval`. Documentation references to tools the skill instructs the agent to call against the user's own Datadog tenant, not executed payloads. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00012
+      reason: "FP: word-fragment matches — `integration_account_id`, table cells, tool-name fragments like `get_llmobs_span_details(trace_id`. Documentation strings, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00030
+      reason: "FP: matched `run()` (SKILL.md:613) inside an example Python experiment-client snippet. Not an attack vector. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00032
+      reason: "FP: matched prose 'skip this step and proceed directly' (SKILL.md:396). Workflow instruction, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00040
+      reason: "FP: matched the word 'Deploy' in prose 'Deploy to Datadog LLM Experiments' (SKILL.md:729). Documentation step, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00051
+      reason: "FP: matched the prose phrase 'For each' (loop-over-traces guidance). Documentation, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00066
+      reason: "FP: matched Datadog eval-prompt template placeholders — `{{input_data}}`, `{{output_data}}`, `{{span_input}}`, `{{meta.input.messages[*].content}}`. These are the documented templating syntax for LLM-judge prompts, not injected payloads. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00090
+      reason: "FP: matched prose 'Extract the rules implicitly followed across observed outputs' (SKILL.md:375). Analysis guidance, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00091
+      reason: "FP: matched literal `\\n` newline escapes inside JSON/code example blocks. Documentation/code, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00110
+      reason: "FP: matched `eval (` inside an example Python evaluator function signature. Not a code-eval sink. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00111
+      reason: "FP: word-fragment matches — `summarizing a policy` (example intent category), `eval_scope` (identifier). Documentation strings, no executable threat. datadog-labs/agent-skills @9bcb3ce."
+    - rule_id: ATR_2026_00213
+      reason: "FP: matched the literal words 'system prompt' / 'System Prompt' in prose describing Datadog span fields and eval dimensions. Documentation, not a system-prompt-extraction attack. datadog-labs/agent-skills @9bcb3ce."
diff --git a/skills/dd-llmo-eval-session-classify/spec.yaml b/skills/dd-llmo-eval-session-classify/spec.yaml
index 67f066e..d266073 100644
--- a/skills/dd-llmo-eval-session-classify/spec.yaml
+++ b/skills/dd-llmo-eval-session-classify/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
-  path: "dd-llmo/eval-session-classify"
-  version: "0.2.0"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
+  path: "dd-llmo/llm-obs-session-classify"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
diff --git a/skills/dd-llmo-eval-trace-rca/spec.yaml b/skills/dd-llmo-eval-trace-rca/spec.yaml
index a9ef196..33426b6 100644
--- a/skills/dd-llmo-eval-trace-rca/spec.yaml
+++ b/skills/dd-llmo-eval-trace-rca/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
-  path: "dd-llmo/eval-trace-rca"
-  version: "0.2.0"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
+  path: "dd-llmo/llm-obs-trace-rca"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
diff --git a/skills/dd-llmo-experiment-analyzer/spec.yaml b/skills/dd-llmo-experiment-analyzer/spec.yaml
index 9810b1a..86833c3 100644
--- a/skills/dd-llmo-experiment-analyzer/spec.yaml
+++ b/skills/dd-llmo-experiment-analyzer/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
-  path: "dd-llmo/experiment-analyzer"
-  version: "0.2.0"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
+  path: "dd-llmo/llm-obs-experiment-analyzer"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
diff --git a/skills/dd-logs/spec.yaml b/skills/dd-logs/spec.yaml
index c422739..8ad48e6 100644
--- a/skills/dd-logs/spec.yaml
+++ b/skills/dd-logs/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
   path: "dd-logs"
-  version: "0.2.0"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
diff --git a/skills/dd-monitors/spec.yaml b/skills/dd-monitors/spec.yaml
index 1c33e4c..33643af 100644
--- a/skills/dd-monitors/spec.yaml
+++ b/skills/dd-monitors/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
   path: "dd-monitors"
-  version: "0.2.0"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"
diff --git a/skills/dd-pup/spec.yaml b/skills/dd-pup/spec.yaml
index 4a11568..affe4c0 100644
--- a/skills/dd-pup/spec.yaml
+++ b/skills/dd-pup/spec.yaml
@@ -9,9 +9,9 @@ metadata:
 
 spec:
   repository: "https://github.com/datadog-labs/agent-skills"
-  ref: "2f664fd5042d6838dfc06b605c6e2619eb8079c5"
+  ref: "9bcb3ceafacae78dbba76c9459a878fc7d6a0d10"
   path: "dd-pup"
-  version: "0.2.0"
+  version: "0.3.0"
 
 provenance:
   repository_uri: "https://github.com/datadog-labs/agent-skills"