The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local binary at /home/bjorn/.codex/skill-support/bin/convex-scan via the subprocess module in scripts/scan_convex.py. This is an intended functionality for performing backend audits.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection by processing untrusted repository data.
Ingestion points: Repository files and AGENTS.md processed by the scan tool.
Boundary markers: None explicitly defined in the prompt instructions to delineate untrusted content.
Capability inventory: Subprocess execution of the convex-scan utility.
Sanitization: No explicit sanitization or filtering is visible in the skill scripts.
[SAFE]: No evidence of data exfiltration, hardcoded credentials, or obfuscated code was found. The skill operates within its defined scope of backend auditing.

convex-audit