The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads verified build metadata from verify.osec.io (OtterSec) and source code archives from codeload.github.com (GitHub). These sources are well-known and reputable services in the blockchain and developer communities.
[COMMAND_EXECUTION]: Employs standard system tools including curl for fetching data, tar for extracting archives, and find for navigating source directories. It also invokes the codex CLI tool to run AI-based analysis modules. These commands are necessary for the skill's intended audit functionality.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted source code from external programs.
Ingestion points: Data enters via resolve-program-address.mjs (GitHub downloads) and review-program.mjs (local file paths).
Boundary markers: The run-specialists.mjs script lacks explicit delimiters or instructions to ignore embedded commands when passing code bundles to specialists.
Capability inventory: The skill uses execFileSync and spawnSync for system operations across all scripts including scripts/resolve-program-address.mjs and scripts/review-program.mjs.
Sanitization: No sanitization or validation is performed on the content of the ingested source files before they are bundled and processed by the AI.

sealevel-guard-review