The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses local application data including Claude Code project logs (~~/.claude/projects) and Codex session rollouts (~~/.codex/sessions). These logs contain historical conversation history and code snippets. Extracted insights are transmitted to the Sibyl network service for cross-project consolidation, which is consistent with the skill's stated purpose of knowledge maintenance.
[COMMAND_EXECUTION]: The instructions utilize standard Unix utilities such as find, grep, stat, and du to discover and analyze log files. It also uses the sibyl command-line tool for interacting with the knowledge repository.
[DYNAMIC_EXECUTION]: Inline Python scripts (python3 -c) are used for structured parsing of JSONL message formats to accurately extract specific fields like message content and session titles. This is a routine method for processing complex nested data structures in a CLI environment.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted historical chat data. While it includes logic to prioritize high-signal segments (like user corrections), instructions or malicious patterns embedded in past conversations could potentially influence the extraction and consolidation process.
Ingestion points: Reads historical session logs from ~/.claude/projects/*.jsonl and ~/.codex/sessions/*.jsonl.
Boundary markers: None explicitly defined in the extraction scripts; the process relies on the agent's ability to distinguish between raw content and instructions.
Capability inventory: Uses the sibyl CLI (write/network access), bash for file discovery, and python3 for parsing.
Sanitization: No explicit sanitization of the extracted log content is performed before passing it to the consolidation tool.

dream