The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from GitHub (issues, PRs) which constitutes an untrusted external source. This creates a potential surface for indirect prompt injection if malicious content is embedded in issue titles or labels processed by the agent.
Ingestion points: SKILL.md and templates/triage.html (ingesting output from gh issue list and gh pr list).
Boundary markers: The skill explicitly instructs the agent to store external data as Base64 blobs within hidden textareas to prevent markup breakout and unintentional instruction execution.
Capability inventory: Read, Grep, Glob, AskUserQuestion, and the use of file writing tools to generate interactive pages.
Sanitization: Robust HTML escaping is implemented via the escapeHtml function in templates. The skill documentation includes specific instructions to mitigate common injection vectors like </textarea> or </script> tag breakout.
[COMMAND_EXECUTION]: The skill instructs the agent to use the gh (GitHub CLI) tool to retrieve repository metadata for its triage and status visualization features.
[DYNAMIC_EXECUTION]: The skill generates interactive HTML explainer and triage pages containing embedded JavaScript for features like canvas-based visualization (JS Art), decision tracking, and clipboard interaction. The JS implementation follows strict purity rules (documented in references/js-art.md) to maintain predictability and security by avoiding external state or network calls within the rendered pages.

glyph