Claude-skill-registry blog-fact-checking
install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/blog-fact-checking" ~/.claude/skills/majiayu000-claude-skill-registry-blog-fact-checking && rm -rf "$T"
manifest:
skills/data/blog-fact-checking/SKILL.mdsource content
Fact Checking
What to Verify
- Claims about external tools/libraries
- Version numbers and API details
- Quotes and attributions
- Technical specifications
- Links match what's claimed in text
- Configuration examples (file paths, formats, options)
- Performance claims (optimization suggestions, benchmark numbers)
Process
-
User directs what to check
- "Check the Redis claim in paragraph 3"
- "Verify the Vagrant version requirements"
- "Is the systemd behavior I described accurate?"
-
Fetch the source
- Use
to get referenced documentationweb_fetch - Read official docs, not secondary sources when possible
- Use
-
Compare claim vs source
- Does the claim match what the source says?
- Is version information current?
- Are quotes/code examples accurate?
-
Report findings
- ✅ Verified: matches source
- ⚠️ Outdated: source has changed
- ❌ Mismatch: claim doesn't match source
- 🚨 Hallucinated: config format/option doesn't exist in docs
Configuration Examples - Special Scrutiny
CRITICAL: Configuration examples are high-risk for hallucination. Before approving any config:
-
Verify the file path exists in official docs
- does this file format exist?.tool-name/config.yml
- is this the documented path?config/settings.json
-
Verify the configuration options
- Are the option names exactly as documented?
- Are the data types correct (array vs object vs string)?
- Are nested paths correct?
-
Check for deprecated formats
- Has the config format changed in recent versions?
- Are we showing old patterns that no longer work?
-
Common hallucination patterns to watch for:
- Inventing
files.hidden-dir/config.yml - Creating YAML configs when tool uses TOML or JSON
- Mixing up config locations (LSP
vs separate config files)init_options - Assuming config file exists because directory exists (e.g.,
≠.ruby-lsp/
).ruby-lsp/config.yml
- Inventing
Red flags:
- "You can configure X via
" without a documentation linksome/path.yml - Config examples with TODO(@claude) markers still in them
- Performance claims without measurements ("this reduces time by 50%")
Not Exhaustive
This is targeted checking, not an audit of every claim. User points to specific sections they want verified.
Example Workflow
User: "Check if I got the LightDM systemd behavior right in the 'Display Manager Symlink' section"
Action:
- Fetch systemd documentation on service types
- Fetch LightDM documentation if available
- Compare claim about "static" service type
- Report: Verified/Mismatch/Unclear
Response Format
**Checked**: LightDM systemd service behavior ✅ **Verified**: LightDM is indeed a "static" unit type requiring explicit symlink to display-manager.service Source: systemd.unit(5) man page confirms static units cannot be enabled without symlinks. **Note**: Minor point - the systemd docs use slightly different terminology but your explanation is accurate.
Tools
for documentationweb_fetch- Always cite sources checked
- Focus on technical accuracy, not writing style