- Home
- Skills
- Testing & Debugging
- Evidence-First Debugger
Works with the AI Agents you already use
Evidence-First Debugger
Debug with evidence, not guesses: ranked hypotheses, a minimal fix, and a regression test for every bug.
Free
Evidence-First Debugger
Example session with this skill installed
My test test_parse_date fails with ValueError: day is out of range for month
when the input is "2024-02-30". Debug it.
- Read your context and instructions
- Compiled the evidence-first debugger
Summary
Root cause: parse_date validates the month and day separately, so
"2024-02-30" passes both checks and only fails later inside datetime.
Evidence
| Observation | Source |
|---|---|
Error raised at utils/dates.py:42 | stack trace |
| Day checked only against range 1-31 | utils/dates.py:30-36 |
| Only Feb 29-30 and 31-day-month edge cases fail | 6 inputs tried |
Ranked hypotheses
- Day validated without considering the month (confirmed by experiment)
- Leap-year handling bug (rejected: 2024-02-29 parses correctly)
- Locale or format issue (rejected: ISO format is parsed correctly)
Minimal fix
Validate with calendar.monthrange(year, month)[1] instead of a fixed 31.
Regression test
test_parse_date_rejects_feb_30 and test_parse_date_accepts_leap_day.
Both verified to fail before the fix and pass after.
Next steps
Audit other callers of the old range check in utils/.
Connects securely to your tools. The creator never sees your data.
About this skill
Evidence-First Debugger makes your agent debug the way a careful engineer does, instead of guessing and patching.
What it does
- Reproduces the failure and captures the exact error, inputs, and environment.
- Collects evidence (stack traces, logs, failing tests, relevant code paths).
- Lists ranked hypotheses, each with the evidence for and against it.
- Runs the cheapest experiment that separates the top hypotheses.
- Applies the smallest fix that addresses the confirmed root cause.
- Adds a regression test that fails before the fix and passes after.
Why use it
- Stops the "change five things and hope" loop
- Keeps diffs small and reviewable
- Leaves behind a test so the bug stays fixed
Best for
Failing tests, runtime exceptions, flaky behavior, and unexpected output in Python, JavaScript, and other codebases an agent can run locally.
Requires execution access for full verification. When execution is unavailable, fixes are marked UNVERIFIED. Intermittent bugs may require multiple iterations to capture evidence.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 29 days ago
- Free with an Agensi account
Requires
Permissions
File Scopes