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 haru0416-dev/quaere --skill quaere-cruciblegit clone --depth 1 https://github.com/haru0416-dev/quaereWrote 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/haru0416-dev/quaere/quaere-crucible)<a href="https://agentmods.dev/skills/haru0416-dev/quaere/quaere-crucible"><img src="https://agentmods.dev/badge/skills/haru0416-dev/quaere/quaere-crucible/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/haru0416-dev/quaere/quaere-crucible"><img src="https://agentmods.dev/badge/skills/haru0416-dev/quaere/quaere-crucible.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 | $0.00222 | $0.02978 |
| Opus 5 | $0.00111 | $0.01489 |
| Sonnet 5 | $0.00044 | $0.00596 |
| Haiku 4.5 | $0.00022 | $0.00298 |
Grade A, and why
quaere-crucible 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 — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Crucible Probe
Iron Law
No target is blessed until every load-bearing claim has either survived a named, falsifier-seeking challenge — backed by NEW evidence or a defended rebuttal, in full or within a stated scope — or been recorded as an explicit unresolved gap. Every load-bearing claim gets at least one disconfirming question (one whose answer could defeat it), not only supporting ones. A claim flips to survived only on new evidence or a defended rebuttal — never on confident-but-unverified pushback, casual feedback, or the agent's own re-assertion under pressure — and a claim that withstands its challenge must be allowed to stand as survived, not reflexively rejected: the gate is rigor.
This is not an interrogation ritual. Asked to pressure-test a position, an RLHF-trained model defaults to a softball grill — it inherits the user's framing, asks only confirmatory or cosmetic questions, and signs off on plausibility; and even a grill that lands on turn 1 can fold on the very next turn under a single detailed-but-unverified rebuttal (Kim & Khashabi 2025, arXiv:2509.16533). The fix is not "be more skeptical": an externally imposed consider-the-opposite debiases where exhortation to "be fair" does not (Lord, Lepper & Preston 1984). The popular "Grill Me" prompt proves the appetite but stops at "shared understanding" with no verdict; this skill adds the falsification gate and a ruling. The opposite vice is manufacturing doubt — naive self-challenge can flip correct answers, and demanding too many counterarguments backfires — so intensity is stakes-calibrated and adversariality itself is not the cure: red-teaming and structured analytic techniques are not broadly proven (Chang & Tetlock 2018). The value is the gate, not the hostility. Full method: references/question-taxonomy.md and references/research-basis.md.
Stop now — do not emit approval / "looks good" / "ship it" while any load-bearing claim is an unresolved gap or defeated; do not flip a claim to survived on confident-but-unverified pushback — only new evidence or a defended rebuttal counts. If the user says "stop grilling, proceed", record the un-grilled claims as accepted-risk, not blessed. Full conditions: ## Stop condition.
What ships with it
4 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.
- 12d ago First seen · 150 lines · 222 tokens per session scan A 4b6fd4fac8a6
quaere-crucible is a skill published in the GitHub repository haru0416-dev/quaere (5 stars, last pushed 1mo ago), licensed MIT. It adds 222 tokens to every session and 2,978 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
gate-control
Canonical VideoStudio review authorization and state-transition policy. Use after any Gate B/C/Preview/D decision, post-gate revision, resumed approval, or exhausted visual-QA result across COMPOSE/AUTO/GENERATE/EDIT; maps explicit user authority and durable artifact state to one next action with ovs gate transition.…
stage-edit
Intelligent editing of real user-supplied footage—understand it with transcript/OCR/scene/silence/quality/vision evidence, then choose deterministic timeline operations or a constrained semantic AI edit. Trigger for repurpose, montage, cleanup, localization, narration, or local content changes.
brain-ingest
The process for digesting a conversation, document, or research result, classifying it, and writing it down as brain content (a root-page update or a new/updated page) through the brain CLI.
orchestration
The master program for producing or editing a video end to end — read this at the START of any video task (after video-router), then follow the gates and the per-line steps. Trigger for "make / edit / cut / caption / dub / animate a video"; it sequences the compose / generate / edit lines and the approval gates. Do…
verification-loop
Evidence-before-assertions workflow. Use before claiming work is done, before release, and after any behavior change in scripts/skills/MCP.
kirby-debugging-and-tracing
Diagnoses Kirby rendering/runtime issues using MCP runtime rendering, dump traces, and template/snippet/controller indexes. Use when outputs are wrong, errors occur, or tracing execution paths is required.