The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands such as test, ls, and git to inspect the filesystem and identify the project's programming language and installed tools. These commands are used solely for environment discovery and do not execute external or untrusted code.
[EXTERNAL_DOWNLOADS]: The skill provides recommendations to install further components using the npx package runner. These references point to the author's own package repository (jwilger/agent-skills) and are presented as optional user-initiated actions.
[PROMPT_INJECTION]: The skill manages an indirect instruction surface where configuration is aggregated from multiple skills into the root instruction file. Evidence chain: 1. Ingestion point: SKILL.md Step 7 reads agents-md-setup.md from all installed skills. 2. Boundary markers: Uses managed comment blocks (e.g., ) in AGENTS.md to delimit generated content. 3. Capability inventory: File-write access to configuration files and Git commit capability. 4. Sanitization: Enforces a 32 KiB size limit on the generated AGENTS.md file and suggests content summarization if exceeded.
[DATA_EXFILTRATION]: The skill reads project metadata and local skill configurations to generate centralized instruction files. All processed data remains within the local project environment and is not transmitted to external domains.

bootstrap