The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several standard shell utilities to manage session data, including find for file discovery, stat for metadata filtering, split for batching large files, and jq for JSON processing. These are invoked in SKILL.md and the provided scripts/extract-session.sh script to parse session logs.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because it processes raw conversation history from past sessions.
Ingestion points: Conversation history is read from ~/.claude/projects/**/*.jsonl in SKILL.md (Step 2 and 3) and via scripts/extract-session.sh.
Boundary markers: The skill lacks explicit boundary markers or instructions to the subagent (Task(opus)) to treat the extracted history strictly as data. This increases the risk that instructions embedded in past conversations could be followed by the subagent during the summarization phase.
Capability inventory: The skill environment allows for subprocess execution (shell commands), file reading, and invocation of subagents.
Sanitization: While the skill filters for specific message types (text) using jq, it does not perform sanitization or filtering for malicious prompt patterns within the text content before analysis.

history-insight