The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the @a11y-skills/audit package from the npm registry and installs Chromium via the Playwright CLI (npx playwright install chromium) as specified in SKILL.md. These are well-known technology registries and services.
[REMOTE_CODE_EXECUTION]: The skill uses npx to dynamically download and execute the @a11y-skills/audit CLI tool at runtime. This execution pattern is documented in SKILL.md and is intended for the skill's primary auditing purpose.
[COMMAND_EXECUTION]: The skill utilizes shell access to run the audit CLI and install browser binaries. Command execution is restricted to specific npx command prefixes in the allowed-tools configuration in SKILL.md, which limits the risk of arbitrary shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill's functionality involves auditing third-party URLs, which introduces an attack surface for instructions embedded in external web content.
Ingestion points: Data enters the agent's context via mcp__playwright__browser_navigate and WebFetch in SKILL.md when accessing external URLs.
Boundary markers: The agent is instructed to follow structured reporting guidelines in references/output-format.md and assign standard status values, but there are no explicit instructions to ignore instructions found within the audited content.
Capability inventory: The agent possesses capabilities to write files (Write), execute scoped shell commands (Bash), and run JavaScript within the browser (mcp__playwright__browser_evaluate), all defined in SKILL.md.
Sanitization: There is no evidence of sanitization or filtering of the ingested web content before it is processed by the agent in the provided files.

auditing-wcag