The Agent Skills Directory

[SAFE]: The skill provides comprehensive documentation and management tools for the SYNAPSE context engine. No malicious patterns or safety violations were identified across the 11 threat categories.
[COMMAND_EXECUTION]: The skill implements a set of CRUD commands (e.g., *synapse create, *synapse add, *synapse toggle) and star-commands (e.g., *brief, *dev) to manage domain files and session state. These operations are restricted to the .synapse/ and .claude/ directories within the project.
[INDIRECT_PROMPT_INJECTION]: The engine creates an injection surface by appending a <synapse-rules> XML block to prompts. This is the core intended functionality for context management. The skill mitigates risks by using explicit boundary markers and prioritizing 'NON_NEGOTIABLE' L0 Constitution rules that cannot be overridden by other layers.
Ingestion points: Reads domain rules from .synapse/* and manifest settings from .synapse/manifest (referenced in references/domains.md).
Boundary markers: Explicitly uses a <synapse-rules> XML block to wrap injected instructions (documented in references/layers.md).
Capability inventory: Modifies local files in .synapse/ and .claude/commands/synapse/ via the management commands (detailed in references/commands.md).
Sanitization: Uses a dedicated parser (domain-loader.js) to validate KEY=VALUE formats before injection (mentioned in references/manifest.md).
[PERSISTENCE]: The skill utilizes a documented Claude Code hook mechanism (.claude/hooks/synapse-engine.js) to process prompt events. This is a standard integration pattern for context-aware agents on this platform.

synapse