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 brcampidelli/chimera-agent --skill chimera-when-two-results-contradict-suspect-the-apparatusgit clone --depth 1 https://github.com/brcampidelli/chimera-agentWrote 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/brcampidelli/chimera-agent/chimera-when-two-results-contradict-suspect-the-apparatus)<a href="https://agentmods.dev/skills/brcampidelli/chimera-agent/chimera-when-two-results-contradict-suspect-the-apparatus"><img src="https://agentmods.dev/badge/skills/brcampidelli/chimera-agent/chimera-when-two-results-contradict-suspect-the-apparatus/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/brcampidelli/chimera-agent/chimera-when-two-results-contradict-suspect-the-apparatus"><img src="https://agentmods.dev/badge/skills/brcampidelli/chimera-agent/chimera-when-two-results-contradict-suspect-the-apparatus.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00051 | $0.01042 |
| Opus 5 | $0.00026 | $0.00521 |
| Sonnet 5 | $0.00010 | $0.00208 |
| Haiku 4.5 | $0.00005 | $0.00104 |
Grade A, and why
chimera-when-two-results-contradict-suspect-the-apparatus 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 — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Trigger
Two measurements you produced yourself disagree by an amount you cannot account for: the run with a fraction of the data scores better than the big one; a component reads 0% on a task an adjacent measurement shows it doing routinely; a change moves a number in the direction it makes impossible.
The trigger is the size of the gap, not its existence. Two runs differing within their noise band is a sampling question, and this card has nothing to say about it. It also does not apply to your number versus somebody else's published number — different data, prompts, seeds and versions explain those all day, and treating that as an apparatus alarm will have you auditing a harness that is fine. The case here is: both sides of the contradiction are yours, and both cannot be true.
Do
- Write the contradiction down first, as two lines: the two numbers, and the exact commands and commits that produced them. Stop theorising until that is on the page — an unwritten contradiction gets softened into a puzzle within about a paragraph.
- Diff the two runs before reasoning about them: config, commit, code path, input file hash, the version of the scorer. Prefer a diff you can read over an explanation you can construct.
- Push known answers through the measurement itself. Run the reference/gold solutions through your grader before trusting any model score. If a known-correct answer is graded as a failure, the defect is in the grader and every number it has produced is void, including the one you liked.
- Only once the apparatus survives step 3 do you spend effort explaining the phenomenon.
- If the apparatus was at fault, retract the numbers rather than reinterpreting them. A score from a broken scorer is not a noisy estimate of the truth; it is unrelated to it.
Avoid
Building several careful hypotheses to explain both numbers and then testing them rigorously. That is the expensive version of this mistake: the rigour is real, the effort is real, and all of it is measured on an artefact. Method quality downstream of a broken instrument only gets you to the wrong answer with better error bars.
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 · 90 lines · 51 tokens per session scan A c33d19c446e5
chimera-when-two-results-contradict-suspect-the-apparatus is a skill published in the GitHub repository brcampidelli/chimera-agent (25 stars, last pushed yesterday), licensed Apache-2.0. It adds 51 tokens to every session and 1,042 once invoked, about $0.0003 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 2026-08-30.
Other skills, from other repositories
diagnose-hard-problem
Disciplined diagnosis loop for hard problems, diagnosing bugs and regressions. Reproduce → minimise → hypothesise → instrument → fix → regression-test.
systematic-debugging
4-phase root cause debugging: understand bugs before fixing.
reproduce-bug
Reproduce a reported bug in googleapis/mcp-toolbox and decide whether it is real, delivering an evidence-backed verdict: confirmed, already fixed, misconfiguration, client-side, works as intended, not reproducible, or blocked. Use whenever a maintainer asks you to reproduce, verify, confirm, or investigate a bug…
fix-failing-tests
Diagnose a failing test in the googleapis/mcp-toolbox repo and land a fix by reasoning from the actual error: read the failure, reproduce it, shrink it until the cause is forced into the open, then fix the cause. Use this whenever a test or CI job is red, a build breaks after a change, many packages fail at once, or a…
credit-note-fixer
Fix the tiny credit-note formatting bug and rerun the exact targeted test command.
code-review
Reviews code for bugs, security issues, and best practices.