anti-hallucination
Skill1 starsChanged 44 days ago0BSDView on GitHub
This skill is a discipline checklist for not fabricating facts. Use it whenever an answer depends on specific names, versions, URLs, flags, APIs, statistics, or citations: verify against a live source when tools allow, and ask the user when they do not. Covers the failure modes - recalled versions, invented packages, plausible URLs, stale APIs - and the fallback chain.
6 other projects use a version of this.
Adopted by
Most recently taken up by
- fabioc-aloha/Alex_ACT_ONEtheir own version, first seen Nov 15, 2026
- fabioc-aloha/Alex_ACT_Editiontheir own version, first seen Oct 13, 2026
- eliabdevweb-svg/opencode-skills-MCPtheir own version, first seen Oct 9, 2026
- udsy19/.claudetheir own version, first seen Oct 5, 2026
- fabioc-aloha/alex-cognitive-architecturetheir own version, first seen Apr 26, 2026
- dariia-m/my_claude_skillstheir own version, first seen Mar 12, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- dariia-m/my_claude_skills★ 35same name
- eliabdevweb-svg/opencode-skills-MCP★ 0same name
- fabioc-aloha/Alex_ACT_Edition★ 0same name
- fabioc-aloha/Alex_ACT_ONE★ 0same name
- fabioc-aloha/alex-cognitive-architecture★ 0same name
- udsy19/.claude★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository75 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agentsSkill
- android-connectivitySkill
- android-devSkill
- android-mediaSkill
- android-performanceSkill
- blue-teamingSkill
- bug-huntingSkill
- cdnSkill
- ci-securitySkill
- contextSkill
- coolifySkill
- dep-versionsSkill
- devopsSkill
- devsecopsSkill
- dockerSkill
- electronSkill
- firecrackerSkill
- gatewaySkill
- ghSkill
- goSkill
- i18nSkill
- k3sSkill
- lxmfSkill
- lxmfySkill
- mcp-toolkitSkill
- media-stackSkill
- mediawikiSkill
- memorySkill
- meshchatxSkill
- mesh-vpnSkill
- micronSkill
- microservices-haSkill
- nebulaSkill
- no-slopSkill
- no-trickerySkill
- opensubsonicSkill
- osintSkill
- owaspSkill
- pnpmSkill
- podmanSkill
- project-scaffoldSkill
- python-packagingSkill
- qaSkill
- releaseSkill
- research-methodsSkill
- reticulum-goSkill
- reticulumSkill
- rngitSkill
- rnsSkill
- rustSkill
- sastSkill
- shipped-textSkill
- stylometrySkill
- surrealdb-cliSkill
- surrealdb-docsSkill
- surrealdb-jsSkill
- surrealdb-pythonSkill
- surrealdbSkill
- surrealdb-vectorSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 25
If a fact is checkable and unchecked, it is a guess.
- Distinctive: line 65
Do not produce a filled-in guess to look productive.
- Distinctive: line 75
When a claim came from an unverified secondary source, say "reported by" not "is".
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
When to use this skill
- Any claim that has a ground truth: a flag name, a function signature, a version, a date, a URL, a CVE ID, a quote.
- The environment lacks web search, curl, or registry access and the answer depends on current external facts.
- You catch yourself writing "as of" without having checked.
The core rule
If a fact is checkable and unchecked, it is a guess. A guess stated confidently is worse than a question asked.
Failure modes to catch
- Recalled versions and package names. Model memory of "latest" is frozen at training time and often wrong even then. Invented names are a real attack surface: slopsquatting registers the plausible names LLMs hallucinate. Always look up. See the dep-versions skill for per-registry commands.
- Plausible URLs. Never construct a URL you have not seen. A wrong domain is a phishing vector. If a URL is needed and cannot be fetched or verified, ask the user for it.
- Flag and API drift. CLIs rename flags between majors. Check
--helpoutput or man pages on the installed binary, not memory of another version. - Statistics and dates. "In 2024, X%" claims need a source. If you cannot name the source, drop the number or mark it approximate.
- Fabricated citations. Never produce a paper title, CVE, or link you did not retrieve. Citing a real-looking but wrong reference is worse than no citation.
- Confident extrapolation. "This repo probably uses X because similar repos do." Check the repo.
Verification ladder
- Local ground truth first: the repo, lockfiles,
--help, the vendored source. Read it. - Then live checks:
npm view,go list -m,curlthe page,gh api, web search. - Then the user: when no tool can reach the fact, ask. Provide the specific question, not a vague "I need info".
- Last resort, state the uncertainty inline: "I believe X but could not verify. Check before relying on it." Do this rarely. A habit of hedging everything is its own failure mode.
When there is no way to check
Say so directly and ask. One sentence: what you need, why, and the
exact form (a file, a version, a paste of --version output, a URL).
Do not produce a filled-in guess to look productive. The user's
preference, stated in this repo's AGENTS.md, is a specific
clarifying question over a wrong answer.
Output hygiene
- Separate CONFIRMED facts from assumptions in research write-ups. The research-methods skill has the labeling contract.
- Mark time-sensitive claims (versions, prices, API status) with the date checked so staleness is visible later.
- When a claim came from an unverified secondary source, say "reported by" not "is".
Related
dep-versions- registry lookups and release-age rules.research-methods- source hierarchy and currency checks.no-trickery- trusting fetched web content safely.no-slop- output quality rules in generated artifacts.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

