Skip to content
Back to skills

Bug Forensics

ASecurity

Optional defect-repair profile for bug search, regression repair, flaky failure investigation, incidents, and security-bug fixes.

  • 18 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 4, 2026
ai-agentssecurity

Security analysis

A100/100

Scanned September 4, 2026

npx -y skills add avksp/agent-lifecycle-kit --skill bug-forensics --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Bug Forensics?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Bug Forensics
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/avksp-bug-forensics/badge)](https://www.skillsdirectory.com/skills/avksp-bug-forensics)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: bug-forensics
description: Optional defect-repair profile for bug search, regression repair, flaky failure investigation, incidents, and security-bug fixes.
---

# Bug Forensics

Use this skill only when the task explicitly asks to find or repair a bug,
regression, flaky failure, incident, or security bug. The profile is optional
and disabled by default for ordinary feature work.

Adapter task intake may recommend this profile through `bugForensicsAdvisory`.
Treat that as an advisory marker only. Do not claim the workflow gate is active
unless the reviewed task or frozen plan explicitly opts in.

## Contract

Before accepting a patch, collect schema-backed evidence for:

- symptom;
- reproduction before modification;
- stable failure fingerprint;
- hypothesis ledger with accepted and rejected hypotheses;
- root cause;
- minimal patch;
- regression proof for the same fingerprint;
- no collateral damage through the canonical fix-impact receipt.

## Inputs

Required phase-1 receipts:

- `agent-bug-reproduction-receipt.v1`;
- `agent-failure-fingerprint.v1`;
- `agent-bug-hypothesis-ledger.v1`;
- `agent-regression-proof-receipt.v1`;
- `agent-fix-impact-receipt.v1`.

Reuse `agent-fix-impact-receipt.v1` for behavior impact. Do not create a
bug-specific competing fix-impact schema.

## Gates

- Reproduction must be red before any code modification.
- Regression proof must show the same fingerprint red before and green after.
- Minimal patch gate must justify any file outside the suspect scope.
- Cross-check is advisory unless the frozen plan explicitly opts into blocking
  use. When used, reuse `agent-cross-check-receipt.v1` with token/resource
  caps.
- Suspect graph, flake detector, and bug-class classifier are phase-2 features
  and must not block the v1 profile unless a later plan enables them.

## Compact Context

For compact hosts, include only the active bug packet, reproduction command,
fingerprint fields, current hypotheses, suspect/write scope, and validation
receipts. Summarize logs by digest and short failure pattern instead of copying
large traces.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.