Skip to content

crowdstrike/crowdstrike-charlotte-ai-agentworks

v1.0.0MIT

Charlotte AI AgentWorks skills for creating, managing, and invoking AI agents. Includes agent lifecycle management, invocation, knowledge base operations, and discovery of models/tools/templates.

Changelog

All notable changes to the Charlotte AI AgentWorks Skills plugin.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

1.0.0 - 2026-10-06

Added

  • Initial release of the agentworks-skills plugin
  • Knowledge Bases skill with full CRUD operations
    • Create, query, get, and update knowledge bases via native FalconPy classes
    • File upload/download with multipart support
    • Audit event querying
    • 8 Python scripts: kb_search, kb_list, kb_get, kb_upsert, kb_file_upload, kb_file_download, kb_files_list, kb_audit
  • Agent lifecycle skill
    • Create/update agents via native Agents class
    • Query and get agent details
    • Publish agents (no enable/disable per v1 scope)
    • Analyze execution patterns and suggest prompt improvements
  • Agent invocation skill
    • Invoke published agents and specific versions via native AgentInvocation/Stream classes
    • Stream invocation results (the Stream class polls and returns the full conversation; no token-by-token SSE)
    • Get messages and cancel in-flight invocations via AgentInvocation.update_agent_invocation
  • Discovery skill
    • Query models, tools, templates, agent versions, and trace spans via native FalconPy classes (Models, Tools, AgentTemplates, AgentVersions, Spans)
  • Setup skill - credential configuration guide
  • Orchestrator skill - intent routing and decision tree
  • Credential resolution from environment variables or TOML profile
    • TOML file at ~/.cache/crowdstrike-charlotte-ai-agentworks/credentials.toml
    • Environment variable support: FALCON_CLIENT_ID, FALCON_CLIENT_SECRET, FALCON_BASE_URL
  • Managed Python venv (Python 3.14+) with auto-bootstrap
    • SessionStart hook builds venv at ~/.cache/crowdstrike-charlotte-ai-agentworks/venv
    • Lazy fallback if hook doesn't fire
    • FalconPy 1.6.6+ from PyPI
  • Multi-assistant support
    • Claude Code (primary)
    • Codex, Copilot CLI, Cursor, Antigravity (via Agent Plugins format)
  • Advisory skill routing hooks (UserPromptSubmit + PreToolUse), with Claude Code and Cursor output shapes
  • Foundry redirect skill and agentworks-foundry-bridge.sh hook: agents and knowledge bases defined in a Falcon Foundry app are pointed to crowdstrike-falcon-foundry; API-managed ones stay here
  • Scripts re-exec through the managed venv when launched with an interpreter that lacks FalconPy
  • Shared display helpers in common/scripts/formatting.py
  • AGENTS.md, CLAUDE.md, GEMINI.md, and a Cursor plugin manifest and hooks file

Documentation

  • README with install instructions, architecture diagram
  • Inline skill docs (SKILL.md per skill)
  • Implementation guide for replicating KB skill pattern

FalconPy Support

  • Native typed classes for every API surface: Agents, AgentVersions (agents); AgentInvocation, Stream (invocation); KnowledgeBases, KnowledgeBaseFiles, KnowledgeBaseAuditEvents (knowledge-bases); Models, Tools, AgentTemplates, Spans (discovery). No uber-class.
  • All clients share one OAuth2 auth object (one token request per session)

Excluded from v1.0 (Per Design)

  • DELETE operations for any resource
  • Agent enable/disable actions (publish-only)
  • Soft-delete for KBs