- Home
- Skills
- Agents & Orchestration
- bug-assurance
Works with the AI Agents you already use
Bug Assurance
An evidence-gated bug-fixing orchestrator that enforces requirement clarification, regression testing, and multi-agent review.
Free
bug-assurance
Example session with this skill installed
start https://github.com/org/repo/issues/402. The ticket says discount logic is applying twice on checkout for guest users.
- Read your context and instructions
- Compiled the bug-assurance
I have initialized the bug-assurance run for issue #402. The environment check passed, and I have captured the initial ticket symptoms.
Bug Assurance Status: INTAKE
| Field | Value |
|---|---|
| Run ID | 20231027-1422-discount-guest-fix |
| Status | INVESTIGATION |
| Risk Level | High (Financial/Discount Logic) |
| Ticket Reference | https://github.com/org/repo/issues/402 |
The intake process is complete. I have mapped the symptoms (double discounting) and affected components (CheckoutService, GuestCartController). I am now transitioning to the read-only investigation phase to trace the code path and identify why guest sessions trigger the discount calculation twice compared to authenticated sessions.
Next steps
- Run
BA transition INVESTIGATIONto trigger the bug investigator agent. - Observe the reproduction of the defect in a local environment.
- Review
understanding.jsonfor competing hypotheses regarding session state management.
Connects securely to your tools. The creator never sees your data.
About this skill
Shipping a bug fix that introduces a regression or violates a business rule is a costly failure. This skill replaces guesswork with an evidence-gated workflow that forces correctness through a deterministic state machine. It prevents application code changes until requirements are clarified, requires proof of a failing regression test, and mandates independent agent reviews for logic and security.
What it does
- Requirement clarification blocks code edits until every material ambiguity is resolved through stakeholder questioning and evidence.
- Root cause investigation requires competing hypotheses and direct evidence like logs or debugger traces before selecting a fix.
- Regression testing enforces the creation of a failing-before-fix test case to prove the defect exists and is later resolved.
- Multi-agent review delegates the diff to specialized agents for business logic, regression risk, and test quality.
- Evidence-based adjudication settles finding disputes using repository evidence and invariants rather than agent confidence scores.
How it works
- Intake and Investigation starts by fetching the ticket and tracing code read-only to establish a root cause with
BA start. - Clarification and Contract resolves unknowns via a Question-Challenge-Verify loop to produce a binding
contract.json. - Execution and Verification implements the minimal fix and runs a verification suite, including independent reviewer agents.
- Final Adjudication processes all findings and produces a status-ready report via
BA transition FINAL_REPORT.
Frameworks & tools
The skill uses a dedicated Python controller ba.py to manage state in .bug-assurance/. It integrates with Bash for repository checks and uses specialized agents for targeted reviews.
Why this beats prompting it yourself
Standard prompts often rush to the fix and hallucinate requirements. This skill uses a strict state machine that physically blocks the agent from editing code until gates pass, ensuring every decision is backed by a persistent audit trail.
Use cases
- Fixing high-risk defects in sensitive business logic or financial systems.
- Resuming complex debugging sessions with a full record of previous findings.
- Auditing external PRs for regressions and hidden business-logic violations.
- Standardizing bug-fix quality across a team of AI agents.
Known limitations
Requires the ba.py controller and specific project structure to be present. Does not support automated PR creation or history rewriting without manual authorization.
Requires the ba.py controller in the project root. Application code changes are physically blocked by hooks until requirements are verified. Verification is fingerprinted and goes stale on edits.
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
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 17 days ago
- Free with an Agensi account
Requires