Search Results for "evidence"

...Instead of reviewing only the final code change, it examines how an agent understood the task, executed work, validated results, delivered safely, and captured reusable lessons. It gathers project and session evidence while marking missing or unobserved behavior explicitly. Reports organize findings across five Agent Work Loop dimensions and connect each conclusion to visible supporting evidence. Supported gaps become prioritized recommendations with expected outcomes, repair boundaries, and acceptance checks. Repeated reports can be compared through a history view to observe workflow trends without claiming unsupported causation. ...

...It requires agents to define acceptance gates before claiming that work is finished. Each gate can specify a command, expected result, working directory, and recorded evidence. A bundled checker validates the ledger, runs approved checks, and can reverify previously completed work. The workflow emphasizes reviewing returned work and reporting only conclusions supported by verifiable evidence. It can be installed for Claude Code, Codex CLI, and other skill-compatible agents, with additional templates and orchestration guidance for more complex tasks.

...The repository contains both a public loop library website and the Loopy skill that agents can use to discover, audit, adapt, run, and prepare loops. It helps turn open-ended prompts into measurable workflows with evidence, stopping criteria, and human approval points. The skill can work with agent environments such as Codex, Cursor, and Claude Code through the Skills CLI. Overall, it is useful for teams that want safer, reusable, and more reliable agent workflows instead of one-off prompting.

...It can capture console/network logs, DOM snapshots, and screenshots, and expose them as typed resources the agent can query or act on. The design aims to make IDE agents (e.g., Cursor, Claude Desktop) more “web-aware,” enabling workflows like reproducing a bug, collecting evidence, and proposing fixes without copy-pasting. Documentation and community guides outline a quick setup, including the extension, the MCP server process, and common troubleshooting steps. The project is actively maintained, with public commit activity and discussions around connectivity and reliability in different IDEs. Overall, it bridges the gap between local coding agents and the real browser they need to observe and manipulate.

...The skill produces human-readable reports, detailed finding traces, structured JSON output, and independent verification against the actual source code. Overall, it is designed to make AI-assisted security review more systematic, adversarial, and evidence-based.

...It offers MCP server support and SSE APIs, so IDEs and agent clients can drive the same workflow programmatically. The result is a repeatable process for scoping questions, collecting evidence, and producing a concise report with citations and reasoning steps.