The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary script scripts/summon.py uses subprocess.run to invoke the claude CLI for generating session handover briefs and the fzf utility for interactive session selection. These executions are scoped to these specific binaries and the distillation process utilizes a restricted system prompt (_DISTILL_INSTRUCTION).
[INDIRECT_PROMPT_INJECTION]: The recover and pick modes ingest untrusted data from local session transcripts (.jsonl files).
Ingestion points: Transcripts are read in extract_conversation within scripts/summon.py.
Boundary markers: The skill uses a structured prompt template (_DISTILL_INSTRUCTION) to guide the LLM during distillation.
Capability inventory: The skill can perform file-write operations (shutil.copy2, shutil.move) within Claude Desktop's local session directories and execute the claude CLI.
Sanitization: The _text_only function filters for conversational text and explicitly skips tool_use and tool_result blocks, and a character budget (EXTRACT_BUDGET_DEFAULT) is enforced to limit the volume of processed data.
[DATA_EXPOSURE]: The skill accesses and manipulates sensitive session metadata and transcripts stored in %APPDATA%/Claude/claude-code-sessions/ and ~/.claude/projects/. These operations are local-only and are required for the skill's stated purpose of session management and account transfer.

summon