citation-guard
Verify citations against the research corpus to prevent hallucinated references and enforce GRADE compliance
Best use case
citation-guard is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
It is a strong fit for teams already working in Codex.
Verify citations against the research corpus to prevent hallucinated references and enforce GRADE compliance
Teams using citation-guard should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/citation-guard/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How citation-guard Compares
| Feature / Agent | citation-guard | Standard Approach |
|---|---|---|
| Platform Support | Codex | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Verify citations against the research corpus to prevent hallucinated references and enforce GRADE compliance
Which AI agents support this skill?
This skill is designed for Codex.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
Related Guides
AI Agents for Coding
Browse AI agent skills for coding, debugging, testing, refactoring, code review, and developer workflows across Claude, Cursor, and Codex.
Cursor vs Codex for AI Workflows
Compare Cursor and Codex for AI coding workflows, repository assistance, debugging, refactoring, and reusable developer skills.
AI Agents for Startups
Explore AI agent skills for startup validation, product research, growth experiments, documentation, and fast execution with small teams.
SKILL.md Source
# citation-guard
Automatically verify citations when agents generate content that makes factual claims or references research.
## Triggers
Alternate expressions and non-obvious activations (primary phrases are matched automatically from the skill description):
- "fabricated citations" / "hallucinated references" → citation authenticity check
- "DOI check" → citation DOI validation
- "verify sources" in research context → citation verification
## Purpose
This skill acts as a passive citation guard, activating whenever agents generate content that includes citations or factual claims requiring evidence. It prevents citation hallucination by verifying references exist before they are written.
## Behavior
When triggered, this skill:
1. **Intercept citation generation**:
- Detect when an agent is about to write content with citations
- Extract all REF-XXX, DOI, and @-mention references
2. **Verify against corpus**:
- Check each reference exists in `.aiwg/research/sources/` or `.aiwg/research/findings/`
- Validate REF-XXX identifiers match frontmatter
- Flag any references not in corpus
3. **Check GRADE compliance**:
- Load quality assessment for cited sources
- Verify hedging language matches evidence quality level
- Suggest language corrections if overclaiming detected
4. **Allow or warn**:
- All citations valid: Proceed silently
- GRADE violation: WARN with suggested language fix
- Hallucinated citation: BLOCK with specific error
5. **Track gaps**:
- Log uncitable claims to `.aiwg/research/TODO.md`
- Track frequently needed but missing sources
## Activation Conditions
```yaml
activation:
always_active_for:
- technical-writer
- documentation-synthesizer
- requirements-analyst
- architecture-designer
- domain-expert
- technical-researcher
triggers_on_content:
- pattern: "REF-\\d{3}"
- pattern: "@\\.aiwg/research/"
- pattern: "according to"
- pattern: "research (shows|demonstrates|suggests|indicates)"
- pattern: "\\(\\w+ et al\\., \\d{4}\\)"
```
## Integration
This skill uses:
- `project-awareness`: Load project's research corpus path
- Citation Verifier agent for deep verification when needed
## References
- @$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/rules/citation-policy.md - Citation policy rules
- @$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/agents/citation-verifier.md - Citation Verifier agent
- @$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/schemas/research/hallucination-detection.yaml - Detection patterns
- @.aiwg/research/docs/grade-assessment-guide.md - GRADE methodologyRelated Skills
verify-citations
Verify all citations in a document against the research corpus
citation-check
Check a file for citation quality and GRADE compliance
citation-backfill
Rebuild bidirectional citation edges across the research corpus. Scans outgoing citation tables, computes the inverse map, and rewrites all incoming tables to match. Fixes one-directional citation graphs after batch inductions.
aiwg-orchestrate
Route structured artifact work to AIWG workflows via MCP with zero parent context cost
venv-manager
Create, manage, and validate Python virtual environments. Use for project isolation and dependency management.
pytest-runner
Execute Python tests with pytest, supporting fixtures, markers, coverage, and parallel execution. Use for Python test automation.
vitest-runner
Execute JavaScript/TypeScript tests with Vitest, supporting coverage, watch mode, and parallel execution. Use for JS/TS test automation.
eslint-checker
Run ESLint for JavaScript/TypeScript code quality and style enforcement. Use for static analysis and auto-fixing.
repo-analyzer
Analyze GitHub repositories for structure, documentation, dependencies, and contribution patterns. Use for codebase understanding and health assessment.
pr-reviewer
Review GitHub pull requests for code quality, security, and best practices. Use for automated PR feedback and approval workflows.
YouTube Acquisition
yt-dlp patterns for acquiring content from YouTube and video platforms
Quality Filtering
Accept/reject logic and quality scoring heuristics for media content