The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external logs from the cost-tracking memory namespace and a local file, creating a surface for indirect prompt injection.
Ingestion points: Retrieves data from the cost-tracking namespace using mcp__plugin_ruflo-core_ruflo__memory_search and reads data from docs/benchmarks/runs/latest.json.
Boundary markers: No clear delimiters or guardrails are specified to prompt the model to ignore potential instructions embedded within the ingested cost logs or benchmark descriptions.
Capability inventory: The skill possesses powerful capabilities including the Bash tool to execute shell lines.
Sanitization: Missing validation or strict data sanitization on the retrieved log records before aggregation.
[COMMAND_EXECUTION]: The skill explicitly commands the execution of an inline node process inside a Bash shell block (node -e 'console.log(...)') to read files from the filesystem, which exposes the runtime environment to shell execution patterns.
[EXTERNAL_DOWNLOADS]: The CLI alternatives segment guides the user or agent to fetch and run an unpinned package execution command using npx @claude-flow/cli@latest from the package registry.

cost-report