The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Node.js script scripts/skill-sync.mjs to create symbolic links between local skill folders and agent-specific directories. This operation is used to integrate skills into various AI platforms. Evidence: scripts/skill-sync.mjs implements atomicSymlink using node:fs's symlinkSync and renameSync. Mitigations: The script includes path safety validation and strictly requires an --approve flag from the user before performing any writes.
[EXTERNAL_DOWNLOADS]: The skill references and interacts with several external registries and marketplaces to discover new skills. It includes instructions for fetching remote content using standard tools. Evidence: references/search-playbook.md contains a curl command to a skill registry for discovery. The skill also mentions external registries such as agentskills.io, aiskillstore.io, and claude-plugins.dev in references/discovery-surfaces.md.
[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of third-party tools and skills using package managers and specialized CLI installers. Evidence: references/discovery-manifests.md lists commands like npx skills add and npx octocode skill for managing skill lifecycles. Mitigations: references/fetch-remote.md and references/install-gates.md mandate a safety scan of remote content and scripts before installation.
[DATA_EXPOSURE]: The skill identifies and accesses configuration directories for various AI agents within the user's home directory to facilitate skill installation. Evidence: references/install-destinations.md lists paths such as ~/.claude/skills/ and ~/.cursor/skills/ as valid targets for installation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process skill definitions from untrusted external sources which could potentially contain malicious instructions. Ingestion points: references/fetch-remote.md defines the process for fetching remote skills via GitHub or registry APIs. Boundary markers: references/quality-rubric.md and references/skill-review.md instruct the agent to inspect the full content of any candidate skill before execution. Capability inventory: The skill can write to the file system and execute commands via CLI installers. Sanitization: The skill includes a dedicated linter scripts/skill-review.mjs to validate the structure and content of skills against a set of safety rules.

octocode-skills