The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes git, jj, and the openspec CLI using Python's subprocess module to retrieve repository state and project information. These calls are restricted to specific arguments. The skill includes a cli() helper that ensures binaries are sourced from the system PATH rather than the local directory to prevent binary planting attacks.- [DATA_EXPOSURE]: The skill reads project metadata and Markdown files within the user's repository. It implements path traversal protections in its walk_files and discover_specs functions by verifying that symlinked files and directories resolve within the repository root.- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted Markdown content from the repository for display in the dashboard. It implements multi-layer sanitization by escaping HTML characters in both the backend JSON API responses and the frontend rendering logic (using marked with custom renderers) to mitigate injection and cross-site scripting risks.- [NETWORK_OPERATIONS]: The skill starts a local web server (defaulting to 127.0.0.1). It includes a _host_allowed validation check that refuses requests with unexpected Host headers, protecting the local server from DNS rebinding and cross-site request forgery attacks.- [OBFUSCATION]: Vendored JavaScript libraries (preact, htm, marked) are provided as minified files for performance. The skill includes a SOURCES.md file that provides the original source URLs and SHA-256 hashes for verification, ensuring transparency of the third-party code.

openspeccer