Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/tjboudreaux/rosettanpx agentmods add commands/tjboudreaux/rosetta/rosetta-grillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/tjboudreaux/rosetta/rosetta-grill)<a href="https://agentmods.dev/commands/tjboudreaux/rosetta/rosetta-grill"><img src="https://agentmods.dev/badge/commands/tjboudreaux/rosetta/rosetta-grill.svg" alt="Measured on agentmods" height="20"></a>What 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 5.1 | $0.00082 | $0.01023 |
| Opus 5 | $0.00041 | $0.00511 |
| Sonnet 5 | $0.00016 | $0.00205 |
| Haiku 4.5 | $0.00008 | $0.00102 |
Grade A, and why
rosetta-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 8d ago.
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.
How it starts
The opening of the file, as written. The whole thing — 69 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/rosetta-grill — grounded interrogation that produces a decision record
grill-me interrogates a plan from scratch. rosetta-grill does it grounded in what's already
decided and what the code actually does, and closes the loop by writing the resolved decision into
the library — so the conversation produces a durable, code-anchored ADR/PDR instead of evaporating.
Workflow
1. Ground in the resolved decision state (don't re-litigate settled decisions)
Before asking anything, resolve the current state for the topic so questions are informed, not naive:
- Locate the decision library (
./decisionsor runrosettafirst if none exists). - For the topic terms, run the resolver — this returns the current decision(s), follows
supersession, and flags conflicts:
python3 scripts/decisions.py resolve --root decisions --text "<topic terms>" python3 scripts/decisions.py search --root decisions --text "<topic terms>" --status Accepted - Read any existing
.agents/ground-truth.mdfor the verified current state. git log --oneline -20+ read the code the topic touches — code/git is the truth hierarchy's top authority (SKILL.md step 6). A plan that contradicts current code must justify the change.
Open by stating what's already settled: "ADR 0042 (Accepted) makes X the current approach, and
foo.py confirms it. Are you changing that? If so this will supersede 0042." If resolve returns a
conflict (≥2 current records), surface it first — that's the highest-value thing to resolve.
2. Grill — one question at a time, recommendation-first, grounded
Walk the decision tree, resolving dependencies one-by-one (the grill-me discipline). For each question:
- Give your recommended answer and the reasoning.
- Anchor it to the resolved state: cite the ADR/code it agrees-with or would-supersede.
- If a question is answerable from the codebase or the decision library, answer it yourself (Read/
Grep/
decisions.py get) instead of asking. - Push on: unstated assumptions, what it supersedes, scope (subsystem/region/tier — Rosetta evals show decisions are often conditional), reversibility, and how you'd verify it in code/git later.
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.
- 8d ago First seen · 69 lines · 82 tokens per session scan A 830399141045
rosetta-grill is a command published in the GitHub repository tjboudreaux/rosetta (2 stars, last pushed 1mo ago), licensed MIT. It adds 82 tokens to every session and 1,023 once invoked, about $0.0004 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
research-verify
Verify existing research findings against independent primary sources. Upgrades confidence from 'sources agree' to 'independently verified.'.
discuss
5-phase requirements-elicitation interview that produces an EARS-validated context/ .md.
sdd
Generate a Kiro-style spec triple — specs/ /{requirements,design,tasks}.md.
adr
Create an Architecture Decision Record in memory-bank BACKLOG.
traceability-gen
Regenerate traceability.md from specs + plans + tests.
spec
Interview and write a spec for a change to this codebase, then list the policies it touches.