The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests data from external sources and the local codebase without explicit sanitization or instructions to ignore embedded prompts.\n
Ingestion points: The specify phase (in references/specify.md) fetches data via WebSearch and WebFetch to research prior art and standards. The execute phase reads existing project files.\n
Boundary markers: The skill uses structured Markdown headers (e.g., ## Part A, ## Part B) to organize context, but does not provide specific boundary markers or 'ignore instructions' delimiters for untrusted content.\n
Capability inventory: The skill can execute shell commands through project test runners, linters, and compilers (in references/verification.md) and can spawn subagents to perform delegated work.\n
Sanitization: No explicit filtering, escaping, or validation of externally fetched content is mentioned in the instructions.\n- [COMMAND_EXECUTION]: The skill is designed to execute shell commands to verify code correctness, including running tests, type checkers, and linters.\n
Evidence: references/execute.md and references/verification.md define a workflow where the agent runs configured project tools such as npm test, tsc, and eslint.\n- [EXTERNAL_DOWNLOADS]: The skill uses web search and fetch tools to gather information from external repositories and documentation sites during the research phase.\n
Evidence: references/specify.md Part A3 instructions explicitly tell the agent to use WebSearch and WebFetch to read about state-of-the-art approaches and open-source projects.

forge