The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation includes examples of shell commands for validating compression output.
Evidence: validation.md contains examples such as diff <(echo "$original") <(echo "$reconstructed") and extract_entities commands that use redirection to /tmp/.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it is designed to ingest and reconstruct arbitrary text provided by users or external sources.
Ingestion points: Input text provided for compression as described in SKILL.md and levels.md enters the agent context at runtime.
Boundary markers: The skill does not define specific delimiters or "ignore embedded instructions" warnings to isolate the processed text from agent commands.
Capability inventory: The documentation in validation.md suggests the agent possesses capabilities for shell execution (diff), tool usage (extract_entities), and file system writes.
Sanitization: No evidence of input validation, escaping, or filtering of the source content was found to prevent malicious instructions from being executed during the compression or reconstruction phases.

Compress