The Agent Skills Directory

[SAFE]: The skill's primary function is to manage and compress conversation context to improve agent performance and efficiency. No malicious patterns or security risks were identified. No obfuscation, persistence mechanisms, or hardcoded credentials were detected.
[EXTERNAL_DOWNLOADS]: The skill mentions using the tiktoken library for token count measurement. This package is retrieved from a well-known service, the official Python Package Index (PyPI), for standard tokenization tasks.
[COMMAND_EXECUTION]: The instructions demonstrate the use of standard system utilities (such as tail, grep, jq, and awk) for log filtering and structural minimization. It also references a utility script, context_retrieve.py, which is used locally to manage and fetch archived context data from the .claude/ directory.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface by processing tool outputs (Bash, Read, Grep) through a summarization hook to manage context tokens.
Ingestion points: Tool outputs are captured and archived by the tool_output_summarizer hook described in modules/reversible-compression.md.
Boundary markers: The process involves creating structural digests and content-addressed archives; however, specific instruction-based boundary markers to prevent model obedience to embedded commands are not explicitly defined in the provided markdown instructions.
Capability inventory: The skill utilizes the context_retrieve.py script and common shell tools to process and retrieve archived content.
Sanitization: No specific content sanitization or instruction-filtering is mentioned for the archived tool outputs beyond structural minimization and truncation.

compression-strategy