The Agent Skills Directory

[DATA_EXFILTRATION]: Step 1.1 in Phase 1 of the SKILL.md instructions captures the entire process.env object into a memory variable. System environment variables frequently contain high-value secrets such as API keys, authentication tokens, and database credentials, posing a high risk of sensitive data exposure.
[DATA_EXFILTRATION]: The skill performs broad environment reconnaissance, including capturing the full working directory path via process.cwd() and conducting filesystem analysis and edit history tracking (analyzeFileSystem, getRecentEdits) without clear scoping or user-defined limits.
[COMMAND_EXECUTION]: The skill uses npx claude-flow@alpha agent-spawn to dynamically create sub-agents (researcher, analyst, planner) whose tasks are generated directly from user-provided input. This architecture allows potentially malicious user instructions to influence the execution and logic of secondary agent processes.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes unvalidated user requests through an autonomous decomposition and planning chain.
Ingestion points: Untrusted user input is captured in Phase 1 of SKILL.md and README.md.
Boundary markers: There are no explicit delimiters or boundary markers used when the user request is passed to the researcher agent for decomposition.
Capability inventory: The skill can spawn new agent processes, write files to the /tmp directory, and store/retrieve data from a persistent memory system.
Sanitization: No sanitization, escaping, or validation of the user request is implemented before the content is used to define sub-agent tasks in Step 2.1.

when-analyzing-user-intent-use-intent-analyzer