Skip to content

CLI Reference

Run sigil --help or sigil <subcommand> --help for up-to-date usage. This page is a tour of the common subcommands.

sigil eval <pr-ref> --service <svc> [options]

Deploy PR + baseline, run scenarios against both, score, and append eval.complete to the ledger.

sigil decide <pr-ref> --service <svc>

Apply the threshold policy to the most recent eval for this PR and emit the decision. Exit 0 (ALLOW), 1 (REVIEW), 2 (BLOCK).

sigil ci owner/repo#42 --service <svc> [--comment] [--auto-merge] [--dry-run]

eval + decide + GitHub integration in one. See CI Integration.

sigil scenario lint [--service <svc>]

Parse + static-check every scenario. Verifies capabilities, metadata, and sandbox-safety rules.

sigil scenario generate --from <spec> --service <svc>

Five-input pipeline (spec, diff, bugs, traces, source) → LLM test plan → scenario code → three-stage validation (lint, parse, optional execution via --verify) with repair loop.

sigil generate-types [--service <svc>]

Emit .sigil/types/sigil.lua — a LuaLS type stub that gives editor autocomplete for the full DSL.

sigil ledger list [--service <svc>] [--kind <kind>] [--since <dur>]
sigil replay <eval-id>

Re-execute an eval bit-for-bit from its reproducibility tuple and diff against the stored result.

sigil trust show --service <svc>
sigil init --service <svc>

Manual ephemeral environment management. Mostly used for debugging.

sigil override <decision-id> --to <allow|review|block> --reason <text>

Record a human override. Appends override.human to the ledger.

Internal MCP stdio server used when provider = "claude-code" in [judge]. Not usually invoked directly.