Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/data-wise/craft/grillgit clone --depth 1 https://github.com/Data-Wise/craftWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00024 | $0.00649 |
| Opus 5 | $0.00012 | $0.00324 |
| Sonnet 5 | $0.00005 | $0.00130 |
| Haiku 4.5 | $0.00002 | $0.00065 |
Grade A, and why
grill scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
/craft:grill — Interrogate Before Building
This command is a thin shim. The canonical behavior lives in the
grillskill. This file preserves the explicit/craft:grillslash entry point and owns the argument surface.
Convergent counterpart to /craft:brainstorm: brainstorm is divergent (generates
options); grill is convergent (interrogates a position to find gaps, contradictions, and
unresolved dependencies before you implement).
When Invoked
- Load the canonical procedure: read
${CLAUDE_PLUGIN_ROOT:-.}/skills/workflow/grill/SKILL.mdand follow it exactly — resolve target, codebase-first pre-answer sweep, select attack angles, the deliberate one-question-at-a-time grill loop (Recommended-first + per-option consequence), milestone checkpoints, durableGRILL-*.mdcapture viacommands/utils/grill_ledger.py, and the one-directional handoff into/craft:plan. - Do not reimplement here. Any change to grill behavior must be made in the skill, never duplicated into this shim.
The flags above (--bound, --no-capture, --yes/--non-interactive, --refine/--no-refine,
--no-tests/--no-docs) are parsed here and passed through to the skill.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 52 lines · 24 tokens per session scan A 8bf1582b19e2
grill is a command published in the GitHub repository Data-Wise/craft (4 stars, last pushed 16d ago), licensed MIT. It adds 24 tokens to every session and 649 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
readme-audit
Audit a README's reading flow by treating it as a conversion funnel. Every section is evaluated against a target reader and a terminal action. Produces a structured audit blueprint that feeds directly into /readme-restructure.
readme-restructure
Execute a /readme-audit blueprint. Rewrites the README to the evaluator → conversion → links structure and creates linked docs by extracting and reorganising existing content.
build-fix
Fix build and type failures with minimal diffs.
checkpoint
Record a verified checkpoint before the next phase.
doctor
Check the install surface for missing files and invalid manifests.
feature-dev
Drive a feature from plan to implementation to review.