The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data by reading and importing historical Claude session transcripts from the local filesystem. This creates a surface for indirect prompt injection, as a prior session might contain malicious instructions (e.g., content Claude read from the web or external files) that are then re-introduced into the active agent's context during the import process.
Ingestion points: Reads .jsonl files from ~/.claude/projects/ and ~/.claude/transcripts/.
Boundary markers: The skill instructions in SKILL.md do not mandate the use of explicit delimiters or 'ignore' instructions when interpolating transcript content into the prompt.
Capability inventory: The agent can execute local Python scripts and the claude CLI, and read/write files as part of the session import workflow.
Sanitization: The Python helper script parses the raw JSON lines but does not perform content sanitization or instruction-filtering on the extracted text before it is returned to the agent.
[COMMAND_EXECUTION]: The skill's workflow relies on executing a local Python helper script (scripts/claude_session_tool.py) and the claude CLI to discover, rank, and export session transcripts.
[SAFE]: The provided Python script (scripts/claude_session_tool.py) uses only Python standard library modules and performs legitimate file system reads within the user's home directory specifically targeted at Claude's official configuration and history folders.

resume-claude-here