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 mehdiforoozandeh/crux --skill crux-wikigit clone --depth 1 https://github.com/mehdiforoozandeh/cruxWrote 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/mehdiforoozandeh/crux/crux-wiki)<a href="https://agentmods.dev/skills/mehdiforoozandeh/crux/crux-wiki"><img src="https://agentmods.dev/badge/skills/mehdiforoozandeh/crux/crux-wiki/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/mehdiforoozandeh/crux/crux-wiki"><img src="https://agentmods.dev/badge/skills/mehdiforoozandeh/crux/crux-wiki.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 | $0.00219 | $0.03121 |
| Opus 5 | $0.00110 | $0.01561 |
| Sonnet 5 | $0.00044 | $0.00624 |
| Haiku 4.5 | $0.00022 | $0.00312 |
Grade A, and why
crux-wiki 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 — 202 lines — stays where its author put it; the contents beside it link to each section on GitHub.
crux-wiki — a literature wiki for a crux vault
This skill adds a literature layer to a crux research vault: a living, interlinked knowledge base of prior knowledge — background, methods, SOTA, baselines, datasets, definitions — that you maintain and draw on. It is [Andrej Karpathy's LLM-wiki pattern] (compile curated sources once into a persistent wiki; keep it current; don't re-derive on every query) instantiated on crux, so crux's deterministic engine does the bookkeeping the pattern needs.
It rides on the crux skill and its engine (<crux skill>/scaffold/ — the crux skill
is this skill's sibling directory in your skills dir; in a repo clone, skills/crux).
If the engine is missing, install the crux skill first
(npx skills add mehdiforoozandeh/crux --all, or the repo's ./install.sh). crux
tracks the new knowledge you're producing (the Question→Hypothesis tree); crux-wiki
tracks the existing knowledge you're building on. They are cross-linked but separate.
The one-way flow rule (hard, load-bearing)
literature (raw/) → wiki/ → informs ask · hypothesize · interpreting findings
Never the reverse. The wiki is a literature source-of-truth. A crux project's own
results — verdicts, findings, closed hypotheses — never flow into the wiki. Wiki pages
may link out to nothing but other wiki pages and their raw/ sources; they must never
cite a q*/h* tree node. This keeps provenance clean (every wiki claim traces to a
curated source, never to your own unpublished result) and structurally prevents the
self-ingestion / model-collapse failure mode that degrades agent-maintained wikis.
The reverse direction is allowed and encouraged: a tree node (a question or hypothesis)
may cite [[wiki/page]] to ground itself in the literature.
The three layers
raw/— the PI's curated sources: papers, articles, notes, data files. Immutable. You read them; you never modify them. This is the source of truth. Only the PI decides what entersraw/— that curation is the human's job in this pattern.wiki/— the compiled pages, owned by you. Concept-slug markdown files (film-conditioning.md), plus the append-onlywiki/log.mdand the co-evolvedwiki/SCHEMA.md. You create pages, update them as sources arrive, and maintain the links.wiki/SCHEMA.md— this vault's conventions (which categories it uses, page style, domain quirks). Seeded on first ingest; you and the PI co-evolve it. Read it at the start of a wiki session so you stay a disciplined maintainer, not a generic chatbot.
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 · 202 lines · 219 tokens per session scan A fc2e17a050fb
crux-wiki is a skill published in the GitHub repository mehdiforoozandeh/crux (6 stars, last pushed 7d ago), licensed MIT. It adds 219 tokens to every session and 3,121 once invoked, about $0.0011 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
ab-testing
Use when designing or analyzing a controlled experiment — falsifiable hypothesis, sample size from an MDE, reading significance/CI/power, CUPED, or rescuing tests that won't go significant. NOT recurring metric tracking (that is analytics), NOT north-star/KPI trees (that is kpi-framework), NOT projecting metrics…
product-principles
Defines 4 Risks confidence thresholds, OST hierarchy levels, Knowledge Pyramid tiers, and state design requirements. Use when evaluating user stories, setting confidence scores, referencing OST levels, scoping MVP, or determining validation sufficiency.
recipe-define
Orchestrate PRD creation from validated hypotheses — standard PRD output with 4 Risks confidence and hypothesis traceability.
recipe-validate
Orchestrate hypothesis validation through type-appropriate methods — prototypes, code analysis, market research, and expert review.
hypothesis-discipline
Manages hypothesis lifecycle, enforces validation criteria, time budgets, and confidence scoring rules. Use when creating hypotheses, updating confidence scores, setting validation criteria, handling timeouts, or recording validation results.
blueprint-standards
Defines structural design artifact formats — information architecture, user flows, content model, brand direction, and AI interaction model. Use when creating or reviewing structural design documents that precede prototype generation.