The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file references/external-sources.md identifies https://skills.sh/flpbalada/my-opencode-config/progressive-disclosure as a source to be fetched to guide retrieval decisions. This URL is from an unverified third-party domain and user (flpbalada) not associated with the skill author or well-known organizations, posing a risk of loading untrusted instructions at runtime.
[PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issues and processes it, creating a surface for indirect prompt injection.
Ingestion points: The subagents/issue-retriever.md retrieves issue titles, bodies, and comments from external GitHub repositories.
Boundary markers: The skill attempts to implement boundary protection in references/retrieval-playbook.md by instructing the agent to rewrite Markdown headings in the ingested content to prevent collision with the snapshot's structure.
Capability inventory: The skill possesses file-write capabilities (writing to docs/ in subagents/issue-retriever.md) and network read capabilities via the gh CLI.
Sanitization: While heading rewriting is used to preserve document structure, there is no evidence of comprehensive escaping or sanitization of the untrusted Markdown body content to prevent downstream injection attacks.
[COMMAND_EXECUTION]: The skill relies on the gh (GitHub) CLI tool to perform its primary function. The references/retrieval-playbook.md provides specific instructions for executing gh issue view and gh api commands to fetch remote data.

fetching-github-issue