The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses npx to download and execute the @agentwhy/cli package from the public npm registry. Running unverified code from an external source is a high-risk activity as the code could be modified or malicious.
[DATA_EXFILTRATION]: The skill is designed to search for and read sensitive file paths, including .env, .npmrc, secrets/, and SSH keys (.ssh/id_rsa*). While the stated intent is auditing, this involves exposing highly sensitive project credentials and private keys to the tool's execution environment.
[COMMAND_EXECUTION]: The skill executes multiple shell commands using npx and manages a background server process via the --detach flag.
[EXTERNAL_DOWNLOADS]: The skill fetches the @agentwhy/cli package from the npm registry (npmjs.com). While the registry is well-known, the specific package and its publisher are not recognized as trusted vendors in this context.
[INDIRECT_PROMPT_INJECTION]: The skill processes session records (referenced in Section 4, SKILL.md) which contain transcript data from AI interactions. This ingestion of untrusted data creates an attack surface for indirect prompt injection. The skill includes instructions to treat this output strictly as data (Section 4, SKILL.md), but it lacks formal sanitization of the external content before processing. Capabilities available include subprocess execution via npx and file system reads (Sections 1 and 2, SKILL.md).
[DYNAMIC_EXECUTION]: The skill dynamically loads and executes code from the downloaded npm package at runtime using the npx utility.

agentwhy-session-audit