t0k0sh1-skills

Adversarial second-opinion review of a finished plan, design doc, or proposal, run before it reaches the human reviewer. Ships as a subagent plus a /devils-advocate command to invoke it explicitly.

1 skill · 1 agent

Always-on action-first output, shipped as an output style and applied automatically on enable: lead with the answer or next action, number multi-step work, show what now works, suppress tangents, no preamble, recap or closers.

1 skill

One iteration of PR babysitting, built to run under /loop: checks PR state, addresses reviewer feedback and failing CI as the PR author, pushes, and ends the loop when every thread is resolved and every check has passed. User-invoked only.

1 skill

Interview mode started with /interview-me <requirement or design>: one open question per message, each preceded by a brief (facts, term definitions, trade-offs at equal depth), no recommendations, no assumed defaults, no multiple-choice menus; contradictions between answers (and with existing docs/adr records) are resolved; ends when the user approves the clarified requirement or design, then offers architecture decision records and files only the ones the user picks via the adrs CLI.

1 skill