The Agent Skills Directory

Dynamic Module Loading: The script scripts/event-schema.mjs utilizes dynamic import() to load a local module to extract event type metadata. While it defaults to a path within the repository, the tool allows an operator to specify an arbitrary file path via the --indexer argument, which executes the code of the imported module in the current process.
Evidence: const mod = await import(pathToFileURL(path.resolve(indexerPath)).href); in scripts/event-schema.mjs.
Context: This capability is required for the tool to analyze event handling logic across different versions of the codebase, but it requires the operator to ensure the provided path is trusted.
Untrusted Data Ingestion (Indirect Prompt Injection Surface): The skill ingests JSON schema files that may originate from external sources or user-specified paths.
Ingestion points: scripts/event-schema.mjs reads and parses JSON schema files.
Boundary markers: Delimiters are not used; the script processes the schema structure directly.
Capability inventory: The script can write to the repository's assets and perform dynamic imports.
Sanitization: The tool implements robust sanitization and defensive bounds (recursion depth, signature length, and work unit budgets) to prevent resource exhaustion or denial-of-service from crafted/malformed schemas.
File System Access: The tool performs file system operations to read Copilot configuration and write metadata updates to the repository.
Evidence: await fs.writeFile(target, text); in scripts/event-schema.mjs.
Context: These actions are scoped to the skill's assets and the Copilot environment. The implementation uses security-conscious flags like O_NOFOLLOW to prevent symbolic link attacks during file reads.

session-log-canvas-update