The Agent Skills Directory

[METADATA_POISONING]: The reference file references/prd-analysis-template.md is deceptively described in the skill documentation as a "PRD section skeleton." Its actual content is a task prompt instructing an AI to perform a security review and output results in SARIF JSON format. This mismatch could mislead the agent into performing an unintended task or generating hijacked output during the requirements gathering phase.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, which creates a vulnerability to instructions embedded in that data.
Ingestion points: Phase 1 involves extracting requirements from external Figma design URLs and legacy Aura component source code provided by the user.
Boundary markers: The instructions do not provide explicit delimiters or warnings to ignore potentially malicious commands embedded within these external design artifacts.
Capability inventory: The skill has the capability to write documents to the workspace and execute local scripts.
Sanitization: While naming validation exists for specific fields, the broader translation of design requirements into code lacks guardrails against embedded malicious instructions.- [DYNAMIC_EXECUTION]: The skill relies on scripts that execute embedded Python code at runtime using heredocs.
Evidence: Both scripts/check-component-name.sh and scripts/detect-project-tools.sh pipe multi-line Python source code directly into the interpreter.- [COMMAND_EXECUTION]: The skill executes local scripts for environment detection and validation using parameters that could be influenced by untrusted inputs.
Evidence: SKILL.md directs the agent to run command-line tools provided with the skill. The construction of these commands by the agent, particularly when incorporating variables like component names derived from design artifacts, represents a potential injection surface if the inputs are not strictly controlled.

experience-lwc-design-generate