The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill's primary function is to process conversation history, which is untrusted data and can contain malicious instructions.
Ingestion points: Systematically scans the full conversation history from the agent's context (SKILL.md).
Boundary markers: Uses structured markdown headers (e.g., ## Decisions) but lacks explicit delimiters or instructions to ignore commands embedded in the source conversation history.
Capability inventory: The skill relies on the agent's default tools for file and shell operations to continue work from summaries.
Sanitization: Employs aggressive summarization and article stripping which provides limited implicit filtering, but no rigorous sanitization is performed on input text.
[PROMPT_INJECTION]: Context Bloat and Obfuscation. Several reference files (e.g., references/architecture-patterns.md, references/security-best-practices.md, references/performance-optimization.md) contain 150 identical technical sections consisting of the same paragraphs repeated verbatim.
This results in a documentation footprint of over 10,000 redundant lines which serves no functional purpose for conversation compression and acts as obfuscation by noise, potentially hiding malicious content or confusing automated analysis tools.
[NO_CODE]: Functional footprint analysis. The skill consists entirely of markdown instructions and documentation. There are no executable scripts (Python, JavaScript, or Shell) included in the skill package, reducing the immediate risk of direct command execution or remote code execution from the skill's own files.

core-context-compressor