loki-grill
Loki-grill interrogates a specification document using Claude's devil's-advocate approach to surface 10-15 critical questions exposing ambiguities, missing acceptance criteria, unstated assumptions, and security or scalability blind spots before development begins. Use this command to harden requirements and identify gaps in specs before implementing code or feeding them into larger development loops.
mkdir -p ~/.claude/commands && curl -fsSL https://raw.githubusercontent.com/asklokesh/loki-mode/HEAD/.claude/commands/loki-grill.md -o ~/.claude/commands/loki-grill.mdloki-grill.md
Harden a spec before any code is written. Loki's grill invokes the provider once with a Devil's-Advocate prompt to surface the 10-15 hardest questions that expose ambiguities, missing acceptance criteria, unstated assumptions, and security/scale blind spots. A grilled spec is a better Reason input to the RARV-C loop. Steps: 1. Run the interrogation on the spec ($ARGUMENTS, or the default resolution prd.md / .loki/generated-prd.md if empty): ``` loki grill $ARGUMENTS ``` It writes `.loki/grill/report.md`. It requires a provider CLI and fails cleanly (exit 3) when none is available: it never fabricates questions. 2. Read `.loki/grill/report.md` and present the findings to the user grouped by category: - Ambiguities and missing acceptance criteria - Unstated assumptions - Security blind spots - Scale and reliability blind spots 3. For each hard question, suggest a concrete way to resolve it in the spec (a precise acceptance criterion, an explicit assumption made explicit, a security control, a stated limit). Do not silently answer them yourself; the point is to harden the human's intent. 4. If the user wants the questions embedded in the spec for the record, offer: ``` loki grill --apply $ARGUMENTS ``` which appends a "Grill findings" section to the spec file. Ask before modifying the spec. Report only what the grill produced. If the provider was unavailable, say so plainly and do not invent questions.
Check whether the spec has drifted from its lock using Loki's living-spec drift detection, and summarize the report.
Run Loki's deterministic PR verification on the current change and summarize the evidence verdict.
Launch Loki Mode autonomous SDLC agent. Handles PRD-to-deployment with minimal human intervention. Invoke for multi-phase development tasks, bug fixing campaigns, or full product builds.
Pause for review every N tasks - selective autonomy pattern
Applies prompt repetition to improve accuracy for non-reasoning LLMs