The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes an install-hook script that modifies the user's ~/.claude/hooks/sessionEnd file to automatically execute the indexing process after every Claude Code session. This persistence mechanism is documented and required for the skill's auto-indexing feature.\n- [DATA_EXFILTRATION]: To provide search functionality, the skill reads sensitive local conversation logs from ~/.claude/projects/ and copies them to a local archive in ~/.config/superpowers/. Additionally, the summarizer.ts component sends conversation excerpts to the Anthropic API using the official Claude SDK to generate searchable summaries. This is consistent with the skill's purpose but involves processing sensitive data externally.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes historical conversation data that could contain malicious instructions designed to influence the agent's behavior during summarization or retrieval.\n
Ingestion points: parser.ts reads .jsonl conversation files from project directories.\n
Boundary markers: The search-agent.md template uses Markdown sections to separate data, but there are no explicit delimiters used during raw text processing for embeddings or summaries to differentiate between data and instructions.\n
Capability inventory: The skill uses npx tsx for code execution and the Anthropic API for text processing.\n
Sanitization: No explicit sanitization or filtering of historical chat content is performed before it is processed by the summarizer or search agent.

Remembering Conversations