The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs a mandatory, read-only version check by fetching metadata from the author's official repository at https://raw.githubusercontent.com/rayfould/Syncora/main/package.json. This operation is limited to comparing version strings and does not execute or evaluate any remote content.
[COMMAND_EXECUTION]: The runtime spawns a git subprocess in scripts/lib/drift-source.mjs to provide change detection evidence and uses a node child process in scripts/lib/workspace.mjs as a safety-hardened reader on Windows systems. These executions are constrained to hardcoded commands with validated arguments.
[INDIRECT_PROMPT_INJECTION]: The skill manages a knowledge graph derived from untrusted Markdown files, presenting a potential attack surface for indirect prompt injection.
Ingestion points: Notes are identified by scripts/lib/graph-scanner.mjs and processed for agent context in scripts/lib/task-context.mjs.
Boundary markers: The skill employs explicit lane headers and content delimiters (e.g., <<<SYNCORA_PROJECT_DATA:...>>>) to isolate project data from instructions.
Capability inventory: The runtime performs file system operations (read/write) within validated roots and executes git for drift detection. It includes transactional recovery and rollback mechanisms to maintain environment integrity.
Sanitization: Untrusted data is processed with strict byte limits, fatal UTF-8 decoding, and JSON escaping before interpolation into context packs, with explicit instructions to the agent to treat note content as data only.

syncora