verifying skill
Enforces a strict verification protocol requiring fresh command execution and output evidence before claiming work is complete or fixed.
- GitHub stars
- 7
- Files
- 1
- Last commit
- May 12, 2026
- License
- MIT
- Allowed tools
- —
What does the verifying skill do?
verifying enforces a strict verification protocol requiring fresh command execution and output evidence before claiming work is complete or fixed.
Summary generated from the skill's SKILL.md.
About the verifying skill
Developers who need to prevent premature claims of success in their AI workflows use this skill to enforce a strict verification protocol. It acts as a gatekeeper that requires an agent to execute fresh commands and provide actual output evidence before it is allowed to state that a task is finished, a bug is fixed, or a build is successful. Instead of relying on assumptions or previous results, the agent must follow a specific sequence of identifying a proof command, running it in full, and reading the exit code or output.
An agent should load this skill whenever it is about to commit code, create a pull request, or report that a feature is complete. It transforms vague statements like "it should work" into evidence-based reports by requiring specific tools such as npm test, pytest, or go test to be run immediately before any claim is made. This ensures that the agent provides concrete proof, such as a zero exit code or a specific number of passing tests, rather than relying on confidence or partial checks.
What an agent can do with it
- 1Confirming test resultsAsk the agent to verify that all unit tests are passing before it claims the task is done.
Written from the skill's SKILL.md. Every use case is checked against a quote from it.
How to install the verifying skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the verifying skill in Plugin
/plugin marketplace add AvinashP/AgentsAtlas /plugin install agents-atlas@agents-atlas
Install the verifying skill in Claude Code
~/.claude/skills/verifying/SKILL.md
$ npx degit AvinashP/AgentsAtlas/skills/verifying ~/.claude/skills/verifyingInstall the verifying skill in Claude Code (project)
.claude/skills/verifying/SKILL.md
$ npx degit AvinashP/AgentsAtlas/skills/verifying .claude/skills/verifyingInstall the verifying skill in Codex
~/.agents/skills/verifying/SKILL.md
$ npx degit AvinashP/AgentsAtlas/skills/verifying ~/.agents/skills/verifyingInstall the verifying skill in Gemini CLI
~/.gemini/skills/verifying/SKILL.md
$ npx degit AvinashP/AgentsAtlas/skills/verifying ~/.gemini/skills/verifyingInstall the verifying skill in Claude.ai
npx degit AvinashP/AgentsAtlas/skills/verifying verifying && zip -r verifying.zip verifying # claude.ai → Customize → Skills → + → Create skill → Upload a skill → verifying.zip
SKILL.md
View on GitHubVerification Before Completion
Evidence before claims, always.
The Iron Law
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
If you haven't run the verification command in this message, you cannot claim it passes.
The Gate Function
BEFORE claiming any status:
1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code
4. VERIFY: Does output confirm the claim?
- If NO: State actual status with evidence
- If YES: State claim WITH evidence
5. ONLY THEN: Make the claim
Skip any step = lying, not verifying
What Requires Verification
| Claim | Requires | Not Sufficient |
|---|---|---|
| "Tests pass" | Test command output: 0 failures | Previous run, "should pass" |
| "Linter clean" | Linter output: 0 errors | Partial check |
| "Build succeeds" | Build command: exit 0 | Linter passing |
| "Bug fixed" | Test original symptom: passes | Code changed |
| "Feature complete" | Requirements checklist verified | Tests passing |
Red Flags - STOP
If you catch yourself:
- Using "should", "probably", "seems to"
- About to say "Done!" before verification
- About to commit/push/PR without running tests
- Relying on partial verification
- Thinking "just this once"
STOP. Run the verification command.
Common Rationalizations
| Excuse | Reality |
|---|---|
| "Should work now" | RUN the verification |
| "I'm confident" | Confidence ≠ evidence |
| "Just this once" | No exceptions |
| "Linter passed" | Linter ≠ tests |
| "I'm tired" | Exhaustion ≠ excuse |
| "Partial check is enough" | Partial proves nothing |
Correct Patterns
Tests:
Run test command → See: 34/34 pass → "All tests pass"
NOT: "Should pass now" / "Looks correct"
Build:
Run build → See: exit 0 → "Build passes"
NOT: "Linter passed" (linter doesn't check compilation)
Bug fix:
Reproduce bug → Apply fix → Test → Bug no longer occurs → "Fixed"
NOT: "Changed the code, should be fixed"
Requirements:
Re-read requirements → Check each → Report gaps or completion
NOT: "Tests pass, must be done"
Verification Commands
# JavaScript/TypeScript
npm test # Run all tests
npm run build # Verify build
npm run lint # Check linting
# Python
pytest # Run all tests
python -m mypy . # Type checking
python -m build # Verify build
# Go
go test ./... # Run all tests
go build ./... # Verify build
# General
echo $? # Check last exit code (0 = success)
When to Apply
ALWAYS before:
- Saying work is complete
- Committing code
- Creating pull requests
- Moving to next task
- Claiming a bug is fixed
- Reporting feature is done
The Bottom Line
No shortcuts for verification.
Run the command. Read the output. THEN claim the result.
This is non-negotiable.
Files in this skill
1verifying skill questions
How do I install the verifying skill in Claude Code?
Run npx degit AvinashP/AgentsAtlas/skills/verifying ~/.claude/skills/verifying. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the verifying skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.
What is the required process for claiming a status?
The process involves identifying the proof command, executing the full command, reading the complete output and exit code, and verifying if the output confirms the claim before making it.