The Agent Skills Directory

[SAFE]: The skill serves as a library of context optimization techniques. Technical analysis of the utility scripts (compaction.py, context_manager.py, degradation_detector.py) confirms they are exclusively focused on string processing, token estimation, and simulation of attention mechanics using the Python standard library.\n- [INDIRECT_PROMPT_INJECTION]: The skill creates an ingestion surface for untrusted data during context assembly. Ingestion points: ContextBuilder.add_section and ProgressiveDisclosureManager.load_detail. Boundary markers: The fundamentals.md reference documentation recommends the use of XML tags and structural markers for section delimitation. Capability inventory: The skill scripts lack dangerous capabilities; there are no file system writes, network operations, or shell command executions. Sanitization: The skill employs truncation and heuristic summarization to manage token budgets.\n- [REMOTE_CODE_EXECUTION]: No remote code execution vectors were found. The code does not use eval(), exec(), or any subprocess-related functions.\n- [OBFUSCATION]: Analysis of all 10 files, including documentation and scripts, found no evidence of hidden or encoded content such as Base64-obfuscated commands or zero-width character embedding.

context-optimization