The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of third-party software from sources not included in the trusted vendors list.
Evidence includes instructions to run npx skills@latest add mattpocock/skills and codex plugin add ponytail@ponytail.
Directs users to external installation instructions at https://github.com/DietrichGebert/ponytail#install.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of remote code by instructing the user (or the agent on behalf of the user) to fetch and install external packages.
Specifically uses npx skills@latest which downloads and executes packages from the npm registry.
[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands to interact with the environment and external services.
Uses gh api --paginate to interact with GitHub's REST API.
Uses git for branch management (git checkout, git push).
Executes npx and codex commands for plugin management.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests and processes untrusted data from external issue trackers.
Ingestion points: Fetches full issue bodies and comments from a configured tracker (e.g., GitHub Issues) using gh api.
Boundary markers: Absent. The prompts passed to subagents (e.g., the 'Implement' and 'Review' briefs) do not include delimiters or instructions to disregard malicious content embedded within the work items or specs.
Capability inventory: The skill possesses capabilities for shell command execution (git, gh, npx), subagent spawning, and modification of the repository (commits, branch pushes, PR creation).
Sanitization: Absent. There is no evidence of filtering or escaping the text retrieved from issue comments before it is interpolated into the prompts for subagents.

ship-spec