reflect

Pass

Audited by Gen Agent Trust Hub on May 5, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes conversation history which may contain untrusted data or adversarial instructions intended to subvert the learning extraction process.\n
  • Ingestion points: Reads conversation history from ~/.claude/ccrecall.db (as described in references/analysis-patterns.md) and the current session context window.\n
  • Boundary markers: No explicit boundary markers or delimiters are used to separate user data from instructions during analysis.\n
  • Capability inventory: The skill has the capability to write proposed learnings into local skill files in .claude/skills/ or ~/.claude/skills/.\n
  • Sanitization: There is no automated sanitization of the extracted content; however, the skill explicitly requires manual user approval before any file writes are performed.\n- [COMMAND_EXECUTION]: The skill utilizes mcp-sqlite-tools to execute SQL queries against a local database file to retrieve message history for analysis.\n- [DATA_EXPOSURE]: The skill identifies and accesses ~/.claude/ccrecall.db, a file containing sensitive conversation history. This data is used locally within the agent's context to facilitate the skill's primary function of reflecting on previous interactions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 5, 2026, 11:56 PM