empirical-validation
Skill952 starsChanged 3 months agoMITView on GitHub
Requires proof before marking work complete — no "trust me, it works
- Sends data out
In this repository12 other files it gives its agents.
- adapters/CLAUDE.mdCLAUDE.md
- codebase-mapperSkill
- context-compressorSkill
- context-fetchSkill
- context-health-monitorSkill
- debuggerSkill
- executorSkill
- plan-checkerSkill
- plannerSkill
- subagent-delegationSkill
- token-budgetSkill
- verifierSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
Run the appropriate command or action
- Very specific: line 37
Add to .gsd/JOURNAL.md under the task
- Distinctive: line 42
If not, task is NOT complete
- Distinctive: line 76
Never use these as justification for completion:
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Empirical Validation
Core Principle
"The code looks correct" is NOT validation.
Every change must be verified with empirical evidence before being marked complete.
Validation Methods by Change Type
| Change Type | Required Validation | Tool |
| UI Changes | Screenshot showing expected visual state | browser_subagent |
| API Endpoints | Command showing correct response | run_command |
| Build/Config | Successful build or test output | run_command |
| Data Changes | Query showing expected data state | run_command |
| File Operations | File listing or content verification | run_command |
Validation Protocol
Before Marking Any Task "Done"
- Identify Verification Criteria
- What should be true after this change?
- How can that be observed?
- Execute Verification
- Run the appropriate command or action
- Capture the output/evidence
- Document Evidence
- Add to
.gsd/JOURNAL.mdunder the task - Include actual output, not just "passed"
- Add to
- Confirm Against Criteria
- Does evidence match expected outcome?
- If not, task is NOT complete
Examples
API Endpoint Verification
# Good: Actual test showing response
curl -X POST http://localhost:3000/api/login -d '{"email":"test@test.com"}'
# Output: {"success":true,"token":"..."}
# Bad: Just saying "endpoint works"UI Verification
# Good: Take screenshot with browser tool
- Navigate to /dashboard
- Capture screenshot
- Confirm: Header visible? Data loaded? Layout correct?
# Bad: "The component should render correctly"Build Verification
# Good: Show build output
npm run build
# Output: Successfully compiled...
# Bad: "Build should work now"Forbidden Phrases
Never use these as justification for completion:
- "This should work"
- "The code looks correct"
- "I've made similar changes before"
- "Based on my understanding"
- "It follows the pattern"
Integration
This skill integrates with:
/verify— Primary workflow using this skill/execute— Must validate before marking tasks complete- Rule 4 in
GEMINI.md— Empirical Validation enforcement
Failure Handling
If verification fails:
- Do NOT mark task complete
- Document the failure in
.gsd/STATE.md - Create fix task if cause is known
- Trigger Context Health Monitor if 3+ failures
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.

