The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/fe-gate.sh executes code downloaded from the NPM registry via npx -y impeccable@3.6.1. This is a remote execution pattern used to run the Impeccable audit tool.
[COMMAND_EXECUTION]: The skill executes multiple local scripts and system commands to perform its workflow.
Evidence includes the execution of python3 $SKILL/scripts/design-sync.py, bash $SKILL/scripts/fe-gate.sh, and node $SKILL/scripts/viewport-check.mjs as documented in SKILL.md and scripts/fe-gate.sh.
[EXTERNAL_DOWNLOADS]: The skill triggers external downloads from well-known services during its execution.
It fetches the impeccable package from the NPM registry.
The scripts/viewport-check.mjs script requires playwright, which typically downloads browser binaries (Chromium) to perform headless testing.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection through its data ingestion features.
Ingestion points: The skill ingests external data in references/intake.md via 'Route C', which uses hallmark study <URL> to fetch and analyze arbitrary web pages. It also reads local design documents in 'Route B'.
Boundary markers: There are no explicit instructions or delimiters shown in the provided files to prevent the agent from following instructions embedded in the analyzed web content or design files.
Capability inventory: The skill has the capability to write files (scripts/design-sync.py), execute shell commands (scripts/fe-gate.sh), and access the network.
Sanitization: While scripts/design-sync.py uses regex for structured token extraction, the initial intake of external HTML or design documentation does not show evidence of sanitization for natural language instructions.

prd-grade-fe