read-references
Skill1 starsChanged 8 months agoMITView on GitHub
Reads and summarizes the project reference (context) files before any development work begins. Prevents AI hallucination by grounding the agent in real project structure. Always invoke this before planning. Step 2 of the dev-workflow.
In this repository12 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- create-prSkill
- delete-branchSkill
- dev-workflowSkill
- fix-planSkill
- implement-fixSkill
- implement-planSkill
- init-referencesSkill
- merge-mainSkill
- plan-featureSkill
- test-lintSkill
- update-referencesSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
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 31
Read these files when they exist:
- Distinctive: line 58
Read the file. Produce a structured summary:
- Distinctive: line 109
After reading the applicable maps, verify that 2–3 key files mentioned in the maps actually exist on disk to catch stale references:
- Distinctive: line 116
If a mapped file does NOT exist:
- Distinctive: line 152
Do NOT start planning or writing code in this step
- Distinctive: line 153
Do NOT assume a file exists if it's not in the maps — verify first
- Distinctive: line 154
Do NOT summarize maps from memory — read the actual files every time
- Distinctive: line 155
Do NOT skip this step even for "small" fixes — maps catch surprises
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill: /read-references
Usage
/read-references {technology}technology: auto | laravel | react | react-native | frontend | all
Purpose
Reference files are the AI's "map" of the project. Reading them before planning ensures the agent works with real files and real function names — not hallucinated ones. This is the single most important step for preventing ghost-file errors.
Steps
1. Load the Applicable Shared References
Read these files when they exist:
references/methodology.md → Project coding standards and conventions
references/schema.md → Database tables, columns, types, relationshipsreferences/methodology.mdshould be loaded whenever presentreferences/schema.mdshould be loaded when present, especially for Laravel or backend-coupled web/mobile tasks
For each file, output a one-paragraph confirmation:
📄 references/methodology.md — loaded. Key conventions noted: [bullet list of 3–5 key points]
📄 references/schema.md — loaded. Tables noted: [comma-separated table names]2. Load Technology-Specific References
If technology = laravel, auto, or all
references/backend_map.mdRead the file. Produce a structured summary:
📄 references/backend_map.md — loaded.
Routes found : [count] endpoints
Key controllers : [list]
Key services : [list]
Key models : [list]If technology = react, auto, or all
references/frontend_map.mdRead the file. Produce a structured summary:
📄 references/frontend_map.md — loaded.
Zustand stores : [list]
React Query hooks: [list]
Key components : [list]
API base URL : [value]If technology = react-native, auto, or all
references/mobile_map.mdRead the file. Produce a structured summary:
📄 references/mobile_map.md — loaded.
Navigators : [list]
Expo plugins : [list]
Screens : [list]
Key constraints : [bullet list]If technology = frontend, auto, or all
references/frontend_map.mdSame as react above.
3. Spot-check Actual Files
After reading the applicable maps, verify that 2–3 key files mentioned in the maps actually exist on disk to catch stale references:
ls -la {path_from_map}If a mapped file does NOT exist:
⚠️ WARNING: references/backend_map.md references {file} but it does not exist.
The reference file may be stale. Flag to lead before planning.4. Produce a Context Summary
Output a final "Ready to Plan" summary:
──────────────────────────────────────
CONTEXT LOADED — READY FOR PLANNING
──────────────────────────────────────
Technology : {technology or inferred stack mix}
DB Tables : {count or n/a}
Backend map : {status}
Frontend map: {status}
Mobile map : {status}
Methodology : loaded
Key constraints for this task:
• {constraint 1 from methodology.md relevant to the task}
• {constraint 2}
• {constraint 3}
Proceed with /plan-feature to design the implementation.
──────────────────────────────────────What NOT to Do
- Do NOT start planning or writing code in this step
- Do NOT assume a file exists if it's not in the maps — verify first
- Do NOT summarize maps from memory — read the actual files every time
- Do NOT skip this step even for "small" fixes — maps catch surprises
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.

