Agent Evidence Mcp
面向 agent 长任务的本地 MCP，自动保存截图、短录屏、打码副本和可复盘证据。
High Impact Semantics Fixture
Inert text for high-impact semantic review.
Office Call Skill
Build and evolve a single high-fidelity counselor system for real student-counselor communication rehearsal, distillation, correction, and reality-sync practice.
Hidden Instruction Benchmark
Inert fixture for hidden instruction review.
Companion Doc Demo
Inert companion-document audit fixture.
Install Risk Demo
Demonstrates install-chain extraction and origin-integrity findings.
Report Demo
A Claude skill from xiexie-qiuligao/agent-skill-guard.
Trusted Helper
First inert skill in a local collision fixture.
Example Vendor Helper
Inert source identity mismatch fixture.
Read Only Helper
Claims a narrow capability surface for v3 manifest regression.
Api Classification
A Claude skill from xiexie-qiuligao/agent-skill-guard.
Sensitive Corpus
A Claude skill from xiexie-qiuligao/agent-skill-guard.
Threat Corpus
A Claude skill from xiexie-qiuligao/agent-skill-guard.
Normal Install Fixture
Demonstrates a regular package-manager install instruction.
Config Control Plane Demo
Inert fixture for OpenClaw config/control-plane audit.
Suppression Audit Demo
Demonstrates visible suppression and audit output.
Suspicious Sources
A Claude skill from xiexie-qiuligao/agent-skill-guard.
Trusted Helper
Second inert skill in a local collision fixture.
False Positive Normal Docs Fixture
Benign documentation that says what not to do.
Claims Vs Does Benchmark
Read-only analysis helper that does not execute commands.
Prompt Risk Demo
Demonstrates prompt-level coercion and attack-path construction.
Runtime Refinement Demo
Demonstrates how runtime facts can narrow a risky path.
Benign V3 Docs Fixture
Benign documentation-only v3 false-positive fixture.
Integrity Estate Benchmark
Inert fixture for digest and local config reference reporting.
Workspace Summary Helper
Summarize local project status for the current workspace.
Direct Exec Demo
Inert demo of direct download-to-shell risk.