The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes locally defined shell commands such as bun test, tsc --noEmit, and curl to verify code quality and audit security headers during the 'Automated Gate' phase.
[EXTERNAL_DOWNLOADS]: The 'Fresh Critic' role is designed to fetch external reference content, known as 'The Bar', from live URLs or public repositories to perform objective benchmarking against production standards.
[DYNAMIC_EXECUTION]: Acceptance criteria and test commands are recorded in a GAUNTLET_JOB_CONTRACT.md file at runtime, which are then used to drive the automated verification steps in subsequent rounds.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data from URLs during the critique phase, presenting an injection surface.
Ingestion points: External reference URLs and fetched repository content.
Boundary markers: The skill mandates the use of an isolated subagent ('Fresh Critic') with no memory of the builder's reasoning and utilizes 'Blind A/B' evaluation to minimize bias.
Capability inventory: The orchestration has access to bash, run_command, and file manipulation tools.
Sanitization: Content is evaluated by a critic agent rather than being directly parsed or executed as instructions.

gauntlet-loop