The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external bug trackers and code repositories to generate test steps, creating a surface for indirect prompt injection.\n
Ingestion points: Fetches ticket titles, bodies, and comments from external trackers (Jira, Linear, Monday, etc.) and reads commit diffs from the file system.\n
Boundary markers: Requires HTML entity escaping for all source-derived values and includes instructions to treat submission results strictly as data rather than instructions.\n
Capability inventory: Has the ability to write HTML files to the disk, call external tracker tools, and initialize a local server receiver.\n
Sanitization: Implements a proactive redaction step to filter credential-like patterns and enforces the use of textContent over innerHTML to prevent cross-site scripting (XSS).\n- [DATA_EXFILTRATION]: The skill establishes a communication channel between the interactive artifact and the agent to transmit test results.\n
The artifact uses a local, single-session loopback URL with a nonce handshake to return pass/fail states and tester notes back to the agent.\n
A mandatory redaction process is described to identify and mask API keys, tokens, and PEM blocks before they are embedded in the generated HTML file.\n- [COMMAND_EXECUTION]: The skill coordinates the use of internal platform tools and external integration tools.\n
It utilizes html-skills-listen and html-skills-stop to manage a local data receiver and session monitoring.\n
It leverages tracker-specific MCP tools to read ticket data and write back verification verdicts.\n- [DYNAMIC_EXECUTION]: The skill generates self-contained HTML artifacts with embedded logic.\n
The instructions explicitly forbid the use of runtime code highlighters or untrusted CDN scripts, requiring that syntax highlighting and DOM construction be performed safely at generation time.

html-testing-checklist