peg/rampart

AI Summary: This GitHub issue proposes adding a policy registry to the Rampart CLI, enabling users to easily discover and install community policies without manual file handling. It will leverage a CI-generated static `index.json` (derived from policy header comments) that the CLI will read to support `search`, `install`, and `show` commands.

AI Summary: This issue proposes adding two new built-in profiles, `research-agent` (read-only) and `mcp-server` (restrictive egress), to Rampart. These profiles address different use cases than existing coding-agent-focused profiles. The task involves creating YAML policy files, adding profile names to the CLI, and ensuring they are listed by `rampart policy list`.

AI Summary: This issue proposes enhancing the `PostToolUseFailure` message to include the specific deny reason and policy name when a tool call is blocked by Rampart. Currently, the message is generic, making it difficult for the AI and user to understand the cause of the denial. The implementation challenge involves correlating the deny decision, made during `PreToolUse`, with the `PostToolUseFailure` message, likely by storing the decision in the existing session state store.

AI Summary: Rampart's current denial messages are generic, causing agents to repeatedly retry blocked commands without understanding why. The proposed solution is to enhance the `PostToolUseFailure` hook to provide specific remediation guidance, including `rampart allow` commands, directly within the denial message. This aims to prevent confused retry loops by offering clear actions to unblock commands, leveraging existing `rampart allow` functionality and the `PostToolUseFailure` hook.