The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing a CLI tool using uv tool install from a local file system path (~/Repos/batterie/deglacer), which constitutes an external dependency installation outside of the skill's own file set.
[PROMPT_INJECTION]: The skill is designed to ingest and parse Claude Code session logs (~/.claude/projects/), creating a surface for indirect prompt injection. Malicious content stored in past conversations (from human messages or tool outputs) could influence the agent's behavior when this data is processed.
Ingestion points: Session history files in .jsonl format located within ~/.claude/projects/ subdirectories.
Boundary markers: No specific boundary markers or instructions to ignore embedded instructions are included in the extraction recipes.
Capability inventory: The skill utilizes Bash (permitting file system access and execution), Read, Grep, and Glob tools.
Sanitization: There is no evidence of sanitization or filtering of the historical conversation data before it is presented to the agent for interpretation.
[COMMAND_EXECUTION]: The skill provides numerous complex Bash commands and jq filters to be executed by the agent for the purpose of exploring and searching session history on the host machine.

deglacer