HadiFrt20/vouch

vouch-promote

Tournament-gate ANY change to a Claude Code artifact. Candidates can be prompts, SKILL.md files, subagent definitions, MCP/settings configs, hooks, agent flows, full pipeline.md files — anything you'd alias as @production. Spawns the vouch-judge subagent per…

Langue du texte source : anglais

vouch-init

Scaffold a new vouch pipeline project — interview-style. Produces pipeline.md describing stages, optional budget.json wallet, optional checklist.md witness template, and an initial state file. Use this BEFORE /vouch-run.

Langue du texte source : anglais

vouch-resume

Resume a failed or interrupted vouch run. Re-runs phases that are in failed/skipped state without disturbing completed phases. Useful after a crash, after topping up the budget, or after fixing whatever caused a stage to fail.

Langue du texte source : anglais

vouch-run

Execute a vouch pipeline by running stages sequentially, debiting the budget per stage, spawning the vouch-stage subagent for the work, and optionally calling vouch-adversary on any Claim objects in the result. Reads pipeline.md from the project directory;…

Langue du texte source : anglais

vouch-status

Print a compact status dashboard for a vouch project — current run progress, phase breakdown, last-N runs, budget balance, latest adversary survival rate. Useful at any point during or after a run.

Langue du texte source : anglais

vouch-verify

Run the vouch adversary against a list of Claims and write verified verdicts. Use this standalone (without /vouch-run) when you already have a JSON file of claims and want survival scores attached. Invokes the vouch-adversary subagent under the hood.

Langue du texte source : anglais

vouch-witness

Evaluate a target document against a markdown checklist. Produces a witness report with per-section pass/fail, severity-weighted score, and follow-up questions sourced from the markdown — not from code. Useful for intake quality, document audits, agent-output…

Langue du texte source : anglais