v1.1.0

Full Changelog: v1.0.0...v1.1.0

v1.0.0

Full Changelog: v0.6.0...v1.0.0

v0.6.0

Full Changelog: v0.5.0...v0.6.0

v0.5.0

Full Changelog: v0.4.0...v0.5.0

v0.4.0

Full Changelog: v0.3.1...v0.4.0

v0.3.1 — LLM Judge, File Path Scan, Safety Gates

Full Changelog: v0.3.0...v0.3.1

v0.3.0

New

• Behavioral baseline detection -- scan compares probe responses against a benign baseline to catch agents that silently comply with attacks instead of refusing
• Scan history -- results saved to .checkagent/history/, shows score delta between runs
• HTML reports -- checkagent scan --report output.html for compliance documentation
• Smarter wrap -- auto-detects PydanticAI, LangChain, CrewAI, OpenAI Agents SDK agent types
• Better prompt analysis -- expanded patterns to catch real-world phrasings
• Trace import safety screening -- import-trace now flags vulnerabilities in production traces

Fixed

• Async stack traces on every --llm-judge scan
• "No safety issues detected" shown when all probes errored
• wrap() crash on framework agent classes
• False positive on scope-limiting refusals ("I can only help with X")
• Windows CI path handling
• Missing ToolBoundary export

Stats

2,285 tests. CI green on Linux, macOS, Windows.

pip install checkagent==0.3.0

v0.2.0 — Security Audit Edition

Major release: CheckAgent is now a security audit tool, not just a testing framework.

New capabilities:

• checkagent scan --sarif — SARIF 2.1.0 output, upload to GitHub Security tab
• checkagent scan --llm-judge — LLM-based evaluation, eliminates false positives ($0.01/scan)
• checkagent scan --url — scan any HTTP endpoint, any language
• checkagent wrap — auto-generate wrapper for any agent class
• checkagent analyze-prompt — instant system prompt security analysis (zero cost)
• checkagent ci-init — generate GitHub Actions workflow
• Execution tracing — see every LLM call, tool call, guardrail check during scans
• Auto-detect agent patterns (.run(), .invoke(), .kickoff())
• 101 attack probes across 6 categories (injection, jailbreak, PII, scope, data enumeration, groundedness)
• Branded safety badges for READMEs
• --repeat N for deterministic CI gates
• --json with fully populated fields
• Remediation guidance in scan output

Docs:

• 18 documentation pages at checkagent.xydac.com
• Comparison guides, framework tutorials, OWASP mapping
• Honest limitations page

Stats: 2,107 tests | CI green on Linux, macOS, Windows

Install:

pip install checkagent
checkagent demo
checkagent scan my_agent:fn --llm-judge

v0.1.2

Version reporting fix, demo animation, --json and --badge scan flags.

v0.1.1 — Asyncio auto-config fix

Fixes async test examples failing without explicit asyncio_mode configuration. All README examples now work out of the box.