The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's preamble contains logic to automatically perform a git pull origin main from a third-party repository (github.com/udayanwalvekar/clearshot) if the user has opted into automatic updates. This allows the repository owner to deploy and execute arbitrary code on the user's system without further review.
[COMMAND_EXECUTION]: The skill uses the ! syntax to instruct the agent to run local shell scripts included with the skill, such as onboarding.sh and telemetry-setup.sh. This execution occurs in the user's local shell environment.
[DATA_EXFILTRATION]: In the epilogue, the skill generates a unique installation ID by hashing the system's hostname and current username. This ID, along with OS, architecture, and session duration data, is transmitted to a remote telemetry server (provided via _CS_CONVEX_URL) using curl.
[EXTERNAL_DOWNLOADS]: The skill fetches version metadata from a remote GitHub URL (raw.githubusercontent.com/udayanwalvekar/clearshot/main/VERSION) using curl.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted user images. 1. Ingestion points: UI screenshots, mockups, and wireframes provided by the user in SKILL.md. 2. Boundary markers: None identified to prevent the agent from obeying instructions embedded within the analyzed images. 3. Capability inventory: Full shell access including file writes (~/.clearshot), network operations (curl), and version control (git pull). 4. Sanitization: No input validation or sanitization is performed on the data extracted from the visual interface before it is processed by the agent.

clearshot