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 skills add Jartan-LLC/grimoire --skill review-severitygit clone --depth 1 https://github.com/Jartan-LLC/grimoireWrote 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/skills/jartan-llc/grimoire/review-severity)<a href="https://agentmods.dev/skills/jartan-llc/grimoire/review-severity"><img src="https://agentmods.dev/badge/skills/jartan-llc/grimoire/review-severity/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/jartan-llc/grimoire/review-severity"><img src="https://agentmods.dev/badge/skills/jartan-llc/grimoire/review-severity.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00035 | $0.02351 |
| Opus 5 | $0.00017 | $0.01175 |
| Sonnet 5 | $0.00007 | $0.00470 |
| Haiku 4.5 | $0.00003 | $0.00235 |
Grade A, and why
review-severity 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 12d 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 — 159 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Review Severity
Every reviewer tiers findings on one model, so a finding lands the same tier whoever raises it. A tier is earned by a named consequence plus a reachable path -- not by unease, vividness, or how easy it is to fix.
The spine
Severity = consequence-if-shipped = impact x reachability. Confidence is not part of it (separate gate, below).
- Impact = KIND (corruption / breach / crash / wrong-result / disclosure) x MAGNITUDE (damage per occurrence) x TIME-HORIZON (irreversibility is just long-horizon impact).
- Reachability = P(path reached) x P(trigger), classed live / edge / dead. Judge it at the trust boundary -- a public/exported API or persisted state is a live entry point even with no in-repo caller; a missing or weakened control on a boundary (authz, rate-limit, crypto strength, secure default) defaults live, its impact being the breach it fails to prevent. And at intended-completion for a diff -- a named wiring/flag/call path is pending-live; only provably unwired code is dead. "I couldn't find the caller" is not dead.
Tiers (directives)
- Critical -- never ship / regenerate. On a reachable path the code misbehaves or is unsafe above the magnitude floor: data loss/corruption (including silent), security breach, crash/hang/resource-exhaustion, won't build-or-run, broken reference, or a wrong result users/callers/downstream act on ("behaves right" means correct output, not merely "runs"). Also Critical when severe damage is confined to an edge but irreversible -- bounded to three kinds only: (i) silent corruption/loss, (ii) external-contract lock-in (published API/schema/event consumers depend on), (iii) irreversible disclosure (secret/ PII/key that can't be un-leaked). Cost-to-fix never demotes.
- Important -- avoid; fix by default. Deferral is a conscious, logged, recoverable choice
(owner + reason), never silent. Two doors, tagged with a
{landmine | debt}sub-label: maintainability-debt -- correct on reachable paths today but a compounding cost a future change must pay (missing tests on load-bearing logic, fragile coupling, drift-prone duplication, a misleading contract); recoverable-latent-landmine -- severe impact whose only reachable path is a genuine recoverable edge (rare input, flag-gated with a named-but-unwired path, an error branch that degrades). Irreversible impact rounds up to Critical, never sits here. - Minor -- optional polish, safe to leave. Expected damage ~0 even when reached: cosmetic/preference, comments restating code, micro-optimizations with no measured effect, debug noise, a non-exploitable info leak. A severe defect on demonstrably-dead code (proven no caller, not exported, no named flag path) is Minor -- latent-flagged so it re-tiers the instant a path is wired.
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.
- 12d ago First seen · 159 lines · 35 tokens per session scan A 22f96c0bdd34
review-severity is a skill published in the GitHub repository Jartan-LLC/grimoire (2 stars, last pushed 4d ago), licensed MIT. It adds 35 tokens to every session and 2,351 once invoked, about $0.0002 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 skills, from other repositories
review-arch
Conduct a principal-level architecture review of a system, specification, diagram, or set of design decisions. Use when asked to review an architecture, evaluate a design, assess coupling/cohesion, check for anti-patterns, audit system boundaries, or produce an Architecture Review Board (ARB) style verdict. Also use…
review-impl
Review implementation changes for a given task against architectural standards. Use when reviewing a PR, evaluating recently committed code, assessing whether implementation changes are correct and architecturally sound, or when asked to 'review my changes', 'check this implementation', 'review what I built', 'is this…
survey
After building a feature (or evaluating a plan), verify it matches what was planned, respects the project's declared architecture and standards, and is production-ready. Reports issues clearly and routes broken behavior to dredge or contested judgment calls to watershed.
watershed
Get a wider, multi-angle critique of code, a plan, or an idea through five distinct critical perspectives, followed by a sixth consolidation pass. Use when a decision is contested, high-stakes, or one careful survey is not enough.
design-parity-review
Prove a Compose UI pull request matches its intended design by diffing the rendered code (candidate) against a Figma / Stitch / Claude Design reference and posting a parity verdict. Use when asked to check a PR against a design, wire up a design-parity bot, set up the committed reference cache so runs make zero Figma…
compose-preview-review
Review pull requests that change Compose UI by rendering @Preview composables on base and head and diffing them. Use when reviewing a UI PR locally or from a CI agent session (@claude mention), authoring an agent-opened PR that touches UI, or triaging flaky or unstable previews (time/random/animation). Pairs with the…