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 escoffier-labs/skillet --skill latent-premisesgit clone --depth 1 https://github.com/escoffier-labs/skilletWrote 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/escoffier-labs/skillet/latent-premises)<a href="https://agentmods.dev/skills/escoffier-labs/skillet/latent-premises"><img src="https://agentmods.dev/badge/skills/escoffier-labs/skillet/latent-premises/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/escoffier-labs/skillet/latent-premises"><img src="https://agentmods.dev/badge/skills/escoffier-labs/skillet/latent-premises.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 | $0.00084 | $0.01337 |
| Opus 5 | $0.00042 | $0.00668 |
| Sonnet 5 | $0.00017 | $0.00267 |
| Haiku 4.5 | $0.00008 | $0.00134 |
Grade A, and why
latent-premises 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 11d 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 — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
latent-premises
An unguarded premise is an undeclared allergen. The dish is fine for most diners, every night, until the wrong one orders it, and nothing on the menu warned anybody. Code carries the same hazard: a premise that holds on every input the system produces today, enforced by nothing, documented nowhere, that fails silently the day the premise breaks.
Core principle: a premise is only a finding when it is genuinely unenforced AND you can name what concretely breaks when it fails. Enforced premises are fine. Consequences you cannot name are hypotheses, not findings.
Read-only. This is a review lens. Fixing is a separate engagement: hand the finished report to expedite, which works the backlog in leverage order.
The taxonomy
Hunt in five categories. Each finding gets exactly one.
| Category | The code assumes | Examples |
|---|---|---|
| input | boundary or external data is well-formed, with nothing upstream guaranteeing it | a non-empty array ([0]), an existing key, a successful parse, a non-null string, an in-range number |
| contract | callee behavior the signature or types do not promise | a result is sorted, a call is idempotent, a value the type says is nullable is never null, a specific error type |
| environment | the world outside the process is arranged | an env var is set, a path exists, a service is reachable, a timezone or locale, an OS behavior |
| ordering | events happen in one sequence | init-before-use, single-threaded access to shared state, no interleaving between read and write, "this runs once" |
| cardinality | a shape or scale property holds | uniqueness assumed but not enforced, one-to-one where the data allows one-to-many, "the list is always small" |
Process
- Pin the scope. Default to the diff or the recently changed code, same as reduce.
- Enumerate the premises. For each changed function, list what it takes for granted, category by category.
- Check enforcement before flagging. Read enough surrounding code to tell whether the type system, a framework contract, or a validated boundary upstream already guarantees the premise. Enforced means NOT a finding, and re-guarding it is over-defensive noise.
- Name the break. State the concrete failure when the premise stops holding: which input or state change, which wrong result or crash. Cannot name it? Drop it.
- Classify honestly. If you can construct a failing input from code visible today, it is a present defect. Hand it to bug-hunt severity rules, not this report. If the premise holds on everything the system can currently produce, it is latent. Report it here.
- Point at one resolution. Every finding ends with exactly one arrow: guard it (add the check), document it (state the precondition where the next caller will read it), or encode it in the type (make the compiler enforce it). Pick the cheapest one that actually closes the hazard.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 11d ago First seen · 78 lines · 84 tokens per session scan A 831bed065b4e
latent-premises is a skill published in the GitHub repository escoffier-labs/skillet (4 stars, last pushed 10d ago), licensed MIT. It adds 84 tokens to every session and 1,337 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 skills, from other repositories
production-audit
Exhaustive production-readiness audit of any codebase or product. Runs diverse-angle discovery passes until two consecutive passes find nothing new, adversarially verifies every finding against the real code, and reports one flat severity-tagged list with file:line evidence. No summary, no hedging.
craft-audit
The front door to the craft skills — a whole-project production-readiness orchestrator that drives the ten per-domain craft skills (ux, frontend, backend, db, security, infra, observability, testing, lint, ai) instead of auditing one surface at a time. It discovers the project's shape, plans an audit per surface, and…
craft-backend
The Craftsman standard for building and reviewing backend code — API routes, request validation, the authentication boundary, service/business logic, error handling, route rate-limit middleware, third-party integrations, and background jobs. Use this WHENEVER the work touches the server side: adding or reviewing a…
review-work
Post-implementation gate review: run manual QA on the real surface yourself, then launch ONE gate reviewer (never a panel) to audit goal, constraints, code quality, security, missed context, and QA evidence. Use before a PR handoff or when the user explicitly asks to review completed work.
critical-code-reviewer
Rigorously review code or pull requests for correctness, security, accessibility, maintainability, tests, and edge cases. Use when users request a critical code review, want a guided walkthrough of findings, need implementer-facing feedback, or want to prepare, create, or submit a GitHub pull request review.
one-way-door
Flags irreversible decisions before commit. Use for data models, infra, auth boundaries, API contracts, event schemas, CI/CD.