anti-hallucination
Skill0 starsChanged 4 months agoMITView on GitHub
Prevent fabricated facts, invented APIs, and citation confabulation at the point of generation. Use when generating factual claims, code examples, API references, library names, configuration values, error messages, or citations — anything where 'sounds plausible' is not the same as 'is real'.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- eliabdevweb-svg/opencode-skills-MCPtheir own version, first seen Oct 9, 2026
- udsy19/.claudetheir own version, first seen Oct 5, 2026
- Quad4-Software/skillstheir own version, first seen Oct 3, 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 versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- dariia-m/my_claude_skills★ 35same name
- Quad4-Software/skills★ 1same name
- eliabdevweb-svg/opencode-skills-MCP★ 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 repository38 other files it gives its agents.
- .github/copilot-instructions.mdCopilot instructions
- agent-creatorSkill
- agent-reviewSkill
- ai-memory-setupSkill
- alex-banner-generationSkill
- brain-auditSkill
- browser-toolsSkill
- code-reviewSkill
- creative-writingSkill
- critical-thinkingSkill
- deep-reviewSkill
- doc-hygieneSkill
- docx-to-mdSkill
- git-workflowSkill
- greeting-checkinSkill
- html-to-mdSkill
- humanizerSkill
- instruction-creatorSkill
- instruction-reviewSkill
- lint-clean-markdownSkill
- markdown-mermaidSkill
- markdown-sanitization-chainSkill
- md-to-emlSkill
- md-to-htmlSkill
- md-to-txtSkill
- md-to-wordSkill
- meditationSkill
- planSkill
- problem-framing-auditSkill
- prompt-creatorSkill
- prompt-reviewSkill
- security-and-hardeningSkill
- skill-creatorSkill
- skill-reviewSkill
- spikeSkill
- status-reportingSkill
- systematic-debuggingSkill
- test-driven-developmentSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
1
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 24
If the answer to *"have I actually seen this work, or am I generating something that sounds plausible?"* is "the second one," stop.
“Common” is measured: kept by at least 6 in 10 of the 4 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
Anti-Hallucination
Anti-hallucination prevents fabrication. Awareness detects errors. Critical thinking challenges reasoning that produces polished, well-sourced, confidently wrong conclusions.
This is the first leg of the epistemic triad (critical-thinking names the other two). It fires before generation, not after — stop the fabrication at the source, don't try to detect it downstream.
When to Use
Activate this discipline whenever about to generate any of:
- Factual claims about external systems, libraries, APIs, services
- Code examples invoking specific function names, parameters, return types
- Citations, references, URLs, paper titles, author names
- Configuration values, default settings, version numbers
- Error messages, log lines, output formats
- Capability claims about tools, frameworks, or platforms
If the answer to "have I actually seen this work, or am I generating something that sounds plausible?" is "the second one," stop. Verify or acknowledge uncertainty.
The Core Discipline
| Stage | Question | If unsure |
| Input-discipline (what I'm about to generate) | Am I about to write something I can't verify is real? | Stop. Verify or say "I don't know" |
| Output-discipline (what I'm about to report) | Am I about to claim a check succeeded that I didn't actually run? | Cite what I checked, or hedge the claim |
The line in the middle is between thinking and typing. Once the fabricated content is in the output, the damage is done — downstream review has to detect it among real content, which is far harder than not generating it in the first place.
Signals That Fire This Skill
Input-discipline signals
| Signal | Response |
"I think there might be a parseFile() method..." | Stop. Read the docs or grep the source. |
"One approach could be calling foo.bar(opts)..." | Verify the API exists with those params. |
"Try this workaround: set FOO_DEBUG=1..." | Confirm the env var is real. |
| About to invent step-by-step instructions for an unfamiliar tool | Stop. Read the actual docs first. |
| User says "that doesn't exist" or "that method isn't real" | Acknowledge immediately. No defending the fabrication. |
Output-discipline signals
| Signal | Response |
| "No matches found" / "Verified clean" / "Nothing returned" | Before reporting absence, confirm the search actually executed against the intended scope. A failed search and a clean search look identical without the scope check. |
| "The doc says X" / "Per the README" / "According to spec" | Before treating doc content as ground truth, cross-check against current filesystem reality. Docs drift from code. Cite both. |
| "I checked and..." / "Verified that..." / "Confirmed..." | Name what was actually checked: file path, command, output snippet. Unattributed "verified" is theatre. |
Verification Patterns
| Need to claim | Verify by |
| An API method exists | Read the function definition or import; don't trust intellisense alone for unfamiliar libraries |
| A config key works | Find it in source / docs, not in another LLM's example |
| A version number is current | Check the package registry or release notes, not training-data memory |
| A file path exists | Test-Path / ls / fs.existsSync in the current workspace |
| A command succeeded | Cite exit code or output, not "ran cleanly" |
| Absence (no matches found) | Cite the search scope; an absent result and a misconfigured search look identical |
Anti-Patterns
| Anti-pattern | Why it fails | Correction |
| "I'll generate the answer and check it after" | Once written confidently, downstream review struggles to flag it | Verify before generating |
| "Probably works like X" | "Probably" is the fabrication signal | Either verify, or say "I don't know" |
| Defending a fabrication when challenged | Doubles down on the original lie | Acknowledge, correct, move on |
| Generating plausible-looking citations | Citation confabulation is a well-documented LLM failure mode | Cite only what's been seen; mark inferred refs as inferred |
| "The documentation must say..." | Inferring doc content is fabrication | Read the doc or hedge the claim |
| Adding "(citation needed)" to fabricated content | The hedge doesn't make the fabrication safer; it just labels it | Remove the fabrication; don't decorate it |
Composition With Other Skills
| Sibling | What it adds |
| epistemic-calibration | The always-on signal tables that fire this discipline at runtime; the calibration vocabulary ("I don't know" > confident lie) |
| critical-thinking | The third leg of the epistemic triad — challenges reasoning after generation; this skill prevents the inputs that reasoning would otherwise compound |
| system-prompt-skepticism | Applies the same don't-invent discipline to instruction interpretation: don't fabricate preconditions, don't confabulate rationale |
The Three Legs of Epistemic Integrity
| Skill | Question | Catches |
| anti-hallucination (this) | Am I making something up? | Fabricated facts, invented APIs, citation confabulation |
| awareness (via epistemic-calibration) | Am I wrong about something? | Retry loops, overconfidence, version errors, manipulation |
| critical-thinking (via critical-thinking) | Am I right for the right reasons? | Bad reasoning, missed alternatives, unexamined assumptions |
Each catches what the others miss. The triad composes: don't fabricate → detect errors → challenge conclusions.
Falsifiability
This skill needs revision if any of the following occur by 2026-08-29 (90 days):
- A user reports fabricated content shipped in 2+ sessions where this discipline should have fired
- The input-discipline signal table fails to catch a new fabrication failure mode (e.g., new LLM-specific confabulation pattern)
- The "verify before generating" pattern is bypassed via post-hoc citation-needed decoration ≥3 times in observed work
- Verification patterns become stale (e.g., new ecosystem ships where the listed verification method doesn't apply)
Track in repo curation logs when available.
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.