Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion capabilities/ai-red-teaming/agents/ai-red-teaming-agent.md
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,7 @@ The AI Red Teaming capability provides these tools:
**Session Context (Iterative Refinement):**

- **save_session_context** — Save current attack context (target, goal, results) for follow-up attacks
- **get_session_context** — Retrieve previous attack context to auto-fill parameters
- **get_session_context** — Pull-only, red-teaming resume aid: call ONLY when continuing a prior attack (e.g. "try another attack", "add transforms") to auto-fill target/goal/models. Do NOT call it to start a task or for non-red-teaming work.
- **clear_session_context** — Clear session to start fresh

**Results & Analytics:**
Expand Down
2 changes: 1 addition & 1 deletion capabilities/ai-red-teaming/capability.yaml
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
schema: 1
name: ai-red-teaming
version: "1.10.0"
version: "1.10.1"
description: >
Probe the security and safety of AI applications, agents, and foundation models.
Orchestrates adversarial attack workflows to discover vulnerabilities in LLMs,
Expand Down
13 changes: 7 additions & 6 deletions capabilities/ai-red-teaming/tools/session.py
Original file line number Diff line number Diff line change
Expand Up @@ -102,16 +102,17 @@ def save_session_context(

@safe_tool
def get_session_context() -> str:
"""Retrieve the current session context for iterative refinement.
"""Retrieve saved AI red-teaming session context when *resuming* a prior attack.

Returns the last target, goal, models, transforms, and results
so you can build on previous attacks without the user re-specifying
everything. Use this at the start of follow-up commands like
"try another attack" or "add transforms".
AI-red-teaming only. Use ONLY to continue a previous red-teaming assessment —
e.g. follow-ups like "try another attack", "add transforms", or "re-run against
the same target" — so you don't have to re-specify the target/goal/models. It is
pull-only: you do NOT need to call this to begin a task, and it is not relevant to
non-red-teaming work. Returns an empty result if no prior red-teaming session exists.
"""
session = _load()
if not session or "current" not in session:
return "No session context found. Run an attack first to establish context."
return "No prior red-teaming session. Nothing to resume — proceed with the task directly."

current = session["current"]
lines = [
Expand Down
Loading