The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill suggests fetching and executing the lighthouse tool and the chrome-devtools-mcp package using npx. While these are standard performance auditing tools, they involve downloading code from the npm registry.
[COMMAND_EXECUTION]: The skill executes shell-based commands including npx lighthouse for page auditing and grep for identifying hot-paths in the codebase. These actions are performed based on external parameters like URLs and function names.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could potentially contain malicious instructions.
Ingestion points: SKILL.md (Phases 1-3) defines workflows to ingest data from external URLs, network request logs, and performance trace files (such as .cpuprofile and .json).
Boundary markers: Absent; the instructions do not include specific delimiters or guardrails to prevent the agent from following instructions that might be embedded in the audited web content or trace metadata.
Capability inventory: SKILL.md (Phases 0, 1, and 5) grants the agent capabilities to navigate to arbitrary URLs, capture traces, execute shell commands via npx, and search local workspace files using grep.
Sanitization: Absent; the skill does not describe any validation or sanitization of data retrieved from external sources before it is analyzed and used to suggest code modifications.

flame-audit