The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external third-party package claude-conversation-extractor via pip. This tool is not from a well-known or trusted organization and serves as the primary mechanism for log access.
[COMMAND_EXECUTION]: The workflow relies on executing shell commands, specifically the claude-extract CLI tool for listing, searching, and exporting session data, as well as grep for parsing the resulting markdown files.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted data from historical chat logs. If a session log contains adversarial instructions, the agent might follow them during the summarization process.
Ingestion points: Reads exported conversation logs (e.g., exported-log.md) and command-line output from claude-extract.
Boundary markers: The skill uses markdown headers like ## 👤 User to identify boundaries but does not include explicit instructions to disregard embedded commands.
Capability inventory: Includes shell command execution (bash), file reading, and file writing.
Sanitization: None detected; instructions emphasize quoting user input "exactly," which preserves any injection payloads present in the logs.

chat-history-summarizer