The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is designed to download and install content from external registries, specifically skills.sh and agentskills.io.
Evidence: Uses npx to fetch and run CLI tools that interact with these remote marketplaces.
Evidence: Downloads skill definitions (SKILL.md files and associated scripts) into the project's .claude/skills/ directory.
[REMOTE_CODE_EXECUTION]: The skill executes remote code by downloading packages from the npm registry at runtime.
Evidence: Commands like npx --yes skills find and npx --yes @agentskill.sh/cli search download and execute code from the npm registry without explicit version pinning or integrity checks.
Evidence: The 'install' workflow downloads external skill content that the AI agent will subsequently load and execute as part of its core capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection via the external data it processes.
Ingestion points: Search results (names, descriptions) and the full content of installed SKILL.md files are sourced from untrusted community registries (skills.sh, agentskills.io) where attackers could publish malicious instructions.
Boundary markers: There are no mentioned delimiters or sanitization steps to prevent instructions inside a downloaded skill from overriding the agent's behavior during the search/review phase.
Capability inventory: The skill uses npx for package management and has the ability to write files to the local file system (.claude/skills/).
Sanitization: No sanitization or validation of the downloaded skill content is performed before installation.
[DYNAMIC_EXECUTION]: The skill dynamically modifies the agent's runtime environment by adding new executable instructions.
Evidence: The installation process adds new SKILL.md files to the active skills directory, which are then integrated into the agent's ontology via omcustom ontology build.
[COMMAND_EXECUTION]: The skill executes shell commands to check for conflicts and manage the local environment.
Evidence: ls .claude/skills/ | grep -w "<skill-name>" is used to check for existing skills.

skills-sh-search