The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests codebase content to generate instructions and documentation, creating a vulnerability surface where malicious code or comments in the target project could influence the generated rules.\n
Ingestion points: The skill reads package.json, lockfiles, configuration files, and directory structures using Read, Grep, Glob, find, ls, and tree commands (found in SKILL.md).\n
Boundary markers: Generated content is wrapped in  and  managed blocks in AGENTS.md and related rule files (found in SKILL.md).\n
Capability inventory: The skill possesses Write, Edit, Bash, and WebFetch capabilities to create documentation and perform repository discovery (found in SKILL.md).\n
Sanitization: The skill includes a hard rule to never read secrets, specifically excluding .env, .env.local, and other credential-bearing files (found in SKILL.md).\n- [COMMAND_EXECUTION]: The skill utilizes restricted shell commands (ls, git, tree, find) via the Bash tool for project discovery and environment detection. While these capabilities are powerful, they are scoped to discovery tasks in SKILL.md. The skill also explicitly documents a policy of not auto-executing project scripts without separate user authorization and review.\n- [EXTERNAL_DOWNLOADS]: The skill utilizes WebFetch and references external templates and source code hosted on GitHub. References to github.com and raw.githubusercontent.com target a well-known service and are used for documentation templates and repository metadata (found in SKILL.md and assets/platform.md).

agents-generator