The Agent Skills Directory

Data Exposure & Exfiltration (LOW): The file src/core/telemetry.ts is configured to send system information, including hostname, platform, RAM, and CPU details, to https://telemetry.spotit.dev. While this can be disabled via the OM_TELEMETRY environment variable, it is enabled by default.
Indirect Prompt Injection (MEDIUM): The skill is designed to ingest data from untrusted external sources (Category 8).
Ingestion points: Data is fetched from external URLs via src/sources/web_crawler.ts and from GitHub issue/PR comments via src/sources/github.ts.
Boundary markers: Analysis of src/server/routes/vercel.ts and src/server/routes/memory.ts shows that retrieved memories are returned to the agent as formatted strings without clear delimiters or 'ignore instructions' warnings.
Capability inventory: The skill has the capability to write to a local or remote database and perform network fetches through various connectors.
Sanitization: There is no evidence of content sanitization to strip potential prompt injection attacks from ingested text.
Obfuscation (LOW): The repository includes dozens of macOS metadata files (e.g., src/server/routes/._auth.ts, src/utils/._text.ts). These are binary AppleDouble files that contain resource fork data. While usually a result of accidental inclusion during development on macOS, they represent non-human-readable 'junk' code that litters the skill surface.
Dynamic Execution (LOW): The skill uses dynamic import() statements in src/core/memory.ts and src/server/routes/sources.ts to load connector logic based on user-provided strings. While the paths are restricted to local source files, this pattern should be monitored.

ahok-memory